Wesleyan University The Honors College

An Impact Study of the Village Savings and Loan Association (VSLA) Program in Zanzibar, Tanzania

by

Conner Brannen Class of 2010

A thesis submitted to the faculty of Wesleyan University

in partial fulfillment of the requirements for the Degree of Bachelor of Arts

with Departmental Honors in Economics Middletown, Connecticut April, 2010

ii

TABLE OF CONTENTS

Acknowledgements iv

Abstract v

Introduction 1

Chapter 1: Background to the Study 4 I. Context for the Study

i. Location and Physical Description ii. Historical Background iii. Economic Background a. Socioeconomic Statistics b. The Education System iv. Women’s Status in the Economy and their Access to Credit v. The Financial Sector a. Formal Sector b. Semi-Formal Sector c. Informal Sector

II. The VSLA Program i. CARE in Tanzania ii. The VSLA Methodology iii. Apex Organizations and the Sustainability of the VSLAs

Chapter 2: Literature Review 31 I. Impact of Microfinance

i. Financial Assets ii. Poverty iii. Quality of Housing iv. Education v. Nutrition and Health vi. Empowerment and Social Status of Women

II. Microsaving III. ROSCA/ASCA Participation IV. VSLA Performance

Chapter 3: Research Design, Methods, & Sample 45 I. Impact Assessment Methodologies

i. Selection Bias ii. Examples in the Literature

II. Study Design i. Sampling Strategy ii. The Individual Survey iii. Focus Group Discussions iv. Interviews with Key Informants

III. Quantitative Data Analysis i. Model Specification

iii

IV. Data Description i. Basic Characteristics of Respondents a. An Additional Test ii. Socio-Economic Status of Respondents a. Quality of Housing b. Household Assets c. Education d. Nutrition e. Health f. Sources of Income g. Social Status iii. VSLA Members Self-Reported Impacts a. Dynamics of VSLA Participation b. Impacts of VSLA Participation iv. Impacts at the Individual Level v. Impacts at the Community Level

Chapter 4: Empirical Results at the Household Level 102 I. OLS Results i. Sources of Income ii. Household Assets iii. Education iv. Nutrition and Health a. Meal Quantity b. Meal Quality c. Health Expenditure II. Probit Results i. Health

a. Use of Mosquito Nets ii. Quality of Housing a. Home Ownership

b. Housing Improvements Conclusion 131 I. Lessons Learned II. Areas for Future Research III. Implications for the Sustainability of the VSL Model i. Sustainability of JOCDO and the Apex Model in General References 140

Appendix A: Literature Review Summary 145 Appendix B: Individual Questionnaire 148 Appendix C: Focus Group Discussion Format 159 Appendix D: Statistical Tables 160

iv

ACKNOWLEDGEMENTS

First and foremost, I would like to thank my thesis advisor Damien Sheehan-Connor for his advice, support and patience throughout the entire processes. I am indebted to Elias, whose brilliance and dedication inspired me and whose awful econometrics jokes made the many hours spent in the lab more bearable. I would like to thank the beautiful ladies of 261 Pine, who listened to my concerns about my thesis for months. I especially would like to thank Emma for the camaraderie throughout the thesis process and Abby, who was always there whenever I needed a study break. I would like to thank Lev for the work parties and the constant support. Last but certainly not least; I would like to thank my family for their constant love and support. Dad – I truly could not have done this without your thoughtful guidance. Thank you for your patience and the countless hours spent proofreading. I am very grateful to the staff at CARE International and the regional apex organization who assisted with the field research in Tanzania. In particular, I would like to thank George Mkoma, the director of the VSLA program for CARE Tanzania, as well as the village trainers, who carried out the individual surveys. Finally, my sincere gratitude goes to the members of the VSLA program who participated in the survey and the focus group discussions, without whom this assessment would not have been possible.

v

ABSTRACT

In 1991, CARE International, a leading humanitarian organization, launched a unique savings-based microfinance program called a Village Savings and Loan Association (VSLA). Today, the model is being replicated across sub-Saharan Africa. Although previous studies have shown substantial benefits from participating in the VSLA program, these studies likely suffer from selection bias and other methodological weaknesses. This study attempts to improve upon the existing work by examining the impact of one of the first VSLA programs, located in Zanzibar, Tanzania, using both quantitative data from individual surveys, and qualitative data from focus group discussions and key interviews. In order to control for selection bias, this study utilizes a control group of new VSLA members who are still in the initial training phase, and also statistically controls for differences in demographic characteristics including age, gender, religion, marital status and education, which may affect program impact. The results suggest that participation in the program has an overall positive impact on various indicators of household and individual welfare, including asset expenditure levels, the development of income-generating activities (IGAs), education expenses, access to health services, nutritional levels and quality of housing. Such positive results are particularly encouraging given the long-term sustainability of the VSLA model - the program does not rely on outside donor funding and does not require continued support from the founding organization. Overall, these results suggest that the VSLA model is both successful and sustainable. Furthermore, it may offer potential teaching benefits for other microfinance programs in developing countries.

1

INTRODUCTION

Over the past twenty years, microfinance has become one of the hottest topics

in development economics. In 2007, more than 100 million of the world’s poorest

families received a microloan (Daley-Harris 2009, 1). Microfinance encompasses the

provision of financial services, including loans, savings and insurance, to low-income

clients who generally lack access to more formal banking services. The promise of

microfinance lies in its ability to empower people to work their own way out of the

poverty trap, while avoiding dependency and the ‘hand out’ shame of conditional aid.

As the number of microfinance institutions has increased across the globe, so has an

interest in understanding the nature of the clients and how they are impacted by

program participation. Although impact studies face a variety of methodological

limitations, numerous studies have found substantial positive impacts of participation

in microfinance programs, specifically in the areas of eradicating poverty, promoting

children’s education, improving health outcomes for women and children, and

empowering women.

I. Objective of the Study

Although traditionally the provision of microloans has been the dominant

feature of most microfinance programs, recently there has been an increasing

appreciation of the importance of savings mechanisms. In 2001, CARE International

2

implemented a unique savings-based microfinance program called a Village Savings

and Loan Association (VSLA) in Zanzibar, Tanzania. In 2006, Decentralized

Financial Services (DFS), a consulting group based in Kenya, carried out an impact

study of the program to examine its long-term sustainability and its impact on its

members (Anyango et al. 2006). Although their results are encouraging, the study

suffers from several methodological weaknesses. Today, as the VSLA methodology

is being replicated across not only Tanzania, but across all of Sub-Saharan Africa, it

is of utmost importance to return to one of the original projects to analyze once again

its impact and long-term sustainability, so that its operations may be better

understood, improved upon and adjusted where needed. The purpose of this study is

to expand and improve upon the study conducted in 2006 and to re-examine the

impact of CARE International’s VSLA program in Zanzibar. The results are intended

to assist CARE and other affiliated organizations to better understand the dynamics

and impact of VSLA participation so that the program might better serve its members.

II. Framework and Hypotheses

The study is comprised of an individual questionnaire administered to 170

households, including those of current members, previous members and incipient

VSLA members (who serve as a control group in order to isolate and assess the

impact of the VSLAs). The survey data is complemented by three focus group

discussions as well as several interviews with key informants within CARE and its

affiliated organizations. Finally, a thorough understanding of both the economic and

3

social setting in which the program operates, as well as of the institution itself,

facilitates interpretation of the data from the survey and focus group discussions.

The hypotheses tested are that participation in the VSLA program would

result in (1) improvements in the economic and social welfare of the household; (2)

growth and/or diversification in income-generating activities (IGAs); and (3)

increased empowerment (social, as well as economic) for members. Under each of

these broader hypotheses, a number of specific hypotheses are developed and

explored in greater detail throughout the report.

III. Organization of the Report

The next chapter provides background information for the study. After

presenting a general profile of Tanzania, it highlights the historical, economic and

social context of the study. It then describes CARE’s role in Tanzania and the VSLA

methodology. Chapter 2 investigates the findings in the literature, in order to facilitate

comparisons with the results of this study. Chapter 3 explains the survey

methodology and the sample of respondents, and presents the initial comparisons

between the statistical means of the data. Chapter 4 presents the results of the

quantitative data analysis. The final chapter reviews the findings and addresses their

significance and implications. Appendix A summarizes the impact studies referenced

in the report. Appendix B presents the format used for the individual questionnaire

while Appendix C presents that used for the focus group discussions. The data tables

referenced in the text are presented in Appendix D.

4

CHAPTER 1 BACKGROUND TO THE STUDY

I. Context of the Study

i. Location and Physical Description

The United Republic of Tanzania lies on the East African coast between

Kenya and Uganda to the north, Rwanda, Burundi and the Democratic Republic of

the Congo to the west, and Malawi and Mozambique to the south. It covers an area of

approximately 364,929 square miles (945,166 km2), which is about 1.5 times the size

of Texas. About 25 miles off the coast of Tanzania sits the semi-autonomous islands

of Zanzibar. Zanzibar is comprised of several islets and two larger islands: Unguja

(the main island, generally referred to as Zanzibar Island), and Pemba. Zanzibar

Island is about 53 miles (85km) long and between 12 and 19 miles (20-30km) wide;

Pemba is about 47 miles (75km) long and between 9 and 12 miles (15-20km) wide.

The largest settlement is Zanzibar Town, or Stone Town, on the west coast of

Zanzibar Island. Both of the larger islands are fairly flat and have a tropical climate.

Temperatures generally fall around 90˚F most days with extremely high levels of

humidity. Tanzania is too near to the Equator to experience any sort of dramatic

contrast between summer and winter. However, the months between October and

April are marginally hotter than those between May and September, with January

5

being the hottest month of the year. The rainy season is generally split into the short

rains, or mvuli, in November and December, and the long rains, or masika, from late

February to early May (Tanzania National Website).

With a population of approximately 40.4 million in 2007, Tanzania is the

second most populous country in East Africa, after Ethiopia. The total population of

Zanzibar Island is about 620,957 and Pemba is about 360,797. The majority of the

Zanzibari population (97 percent) practices Islam, owing to the centuries-long

colonization as an Omani sultanate; the remaining population is a mix of Hindus and

Christians (World Bank 2009, 1). Kiswahili and English are Tanzania’s two official

languages, but Arabic is also commonly spoken in Zanzibar (Tanzania National

Website).

ii. Historical Background

Zanzibar was formerly an Omani colony with a strict racial hierarchy in which

Arabs dominated the black majority. In the late 19th and early 20th centuries, the

power of the Omani sultans waned and they became simply puppet rulers under the

British Empire. In the early 1960s, as the nationalist tide swept across the colonies,

the British began to withdraw and on December 10th, 1963, Zanzibar became an

independent nation. A month later, the bloody Zanzibar Revolution, supported by the

black majority, overturned the largely Arab government, banished the sultan and his

family and brought the majority Afro-Shirazi Party (ASP) to power. On April 12th,

1964, the socialist-oriented president, Sheikh Abeid Amani Karume, signed a

declaration of union with Tanganyika, thus forming the United Republic of Tanzania

(Tanzania National Website).

6

In 1967, just three years after unification, with the adoption of the Arusha

Declaration, the newly-created Tanzanian government launched the Ujamaa village

development scheme across all of Tanzania, including Zanzibar. Ujamaa was

intended to rally the citizenry around the banner of socialism and to increase

productivity through the creation of communal villages. Within just seven years,

more than 9 million people (60 percent of the population) had been resettled into

6,000 villages (Ingle 1972). Rather than increasing production and generating

development as expected, these policies left the rural population worse off than

before. State marketing boards were created to act as the middleman between the

producers and consumers. However, these marketing boards simply facilitated the

overtaxation of the rural agricultural sector. Government taxed the agricultural sector

heavily through both direct taxation (usually by turning the internal terms of trade

against agriculture through such interventions as artificially low consumer prices for

food and high input prices) and indirect taxation (mainly through the impact of an

overvalued exchange rate on agricultural tradeables). The surplus generated from the

overtaxation of the rural population was not subsequently reinvested in rural

infrastructure or services, but rather in gaudy and unnecessary development projects,

primarily in urban areas, such as monuments or ill-planned industrialization projects

(Lubawa 1985).

In Zanzibar, clove production particularly suffered under the socialist policies

of Ujamaa. In the early 1970’s, Zanzibar was the world’s leading producer of cloves.

However, under Ujamaa, in a process similar to that occurring throughout Tanzania,

the large farms were split up into fewer units and it became illegal to sell cloves to

7

any buyer other than the government. As a result, farmers received a price lower than

the world market value, which caused systematic underinvestment (Lubawa 1985).

Few new trees were planted and the current trees are now coming to the end of their

productive lives. Consequently, clove production in Tanzania has never returned to its

pre-Ujamaa glory. Today, Zanzibar ranks a distant third in the world market, with

Indonesia supplying 75 percent of the world's cloves compared to Zanzibar's 7

percent (Country Report 2008).

The failure of Ujamaa goes beyond state marketing boards and diminished

incentives in agriculture. From the very beginning, the project was plagued with poor

planning and ill-suited strategies. Administrators designated a large proportion of

funding to modern technologies, which were ill-suited to the environment as well as

the subsistence-style farming. Despite the burden of such inappropriate technologies

and the artificially low prices, most farmers had few alternatives. Private

entrepreneurship was discouraged and for those who did hold strong entrepreneurial

ambitions, access to the necessary credit was severely limited, even non-existent. The

government prohibited the formation of private initiatives, such as Non-

Governmental Organizations (NGOs) and credit cooperatives, and all commercial

banks were nationalized and thus responded solely to the needs of the state rather than

the poor entrepreneur (Mutesasira 1992, 2). Therefore, though most remained

employed in agriculture, they chose to decrease production in response to the

detrimental policies of Ujamaa.

In the late 1970s, as overall agricultural output began to decline and hunger

intensified across the country, the Ujamaa program began to unravel. It subsequently

8

fell apart completely when an economic crisis struck the country in the beginning of

the 1980s. Real per capita income growth dropped from 1.9 percent between 1970-76

to negative 1.0 percent between 1980-85. Meanwhile, inflation rose unabated, spiking

to 44 percent by 1984, while internal and international deficits continued to rise. The

situation was further exacerbated by the 1978 war with Uganda’s Idi Amin and a

large drought in the 1980s (Muganda 2004, 1).

In 1986, the magnitude and intensity of the economic crisis led the Tanzanian

government to adopt the IMF-directed Economic Recovery Program (ERP), which

included economic stabilization and structural adjustment measures (Muganda 2004,

1). As government jobs and overall spending were cut, unemployment increased

significantly. More and more people were forced into self-employment and informal

business activities. However, lack of access to credit made success in the informal

sector difficult to achieve, particularly in rural areas where the majority of the

population lived. The privatization of the National Microfinance Bank (NMB) and

Cooperative Rural Development Bank (CRDB), which was part of the IMF’s

structural adjustment measures, resulted in the closure of seventy-eight branches

throughout the country, further restricting credit accessibility for the increasing

proportion of the rural population involved in the informal sector (Ssendi and

Anderson 2009, 5). Although the late 1980s also saw a shift in financial policy, with

an increasing number of private and NGO-institutions and cooperatives participating

in microcredit schemes, access to credit continues to be severely limited to this day.

As of 2007, just 10 percent of the population had access to formal financial services,

up from 6.4 percent in 2001 (World Bank 2009, x).

9

iii. Economic Background

Tanzania’s economy was slow to recover from the legacy of Ujamaa. Real

GDP growth was stagnant throughout the 1990s. However, it picked up in the second

half of the decade, averaging 4.2 percent between 1996 and 2000. Since 2000, growth

has continued to rise, despite several years of drought, reaching 7.0 percent in 2007,

making Tanzania one of the fastest growing non-oil economies in Sub-Saharan Africa

(World Bank 2009). The small economy of Zanzibar, however, has been much more

erratic, with wide swings in GDP growth rate from year to year. Although real GDP

growth averaged 7 percent between 1996 and 2000, it peaked at 16.1 percent in 1996,

but was only 1.6 percent in 1998. In 2001 and 2003, growth rebounded, with rates

around 9 percent. However, the rate slowed again in 2005 to 5.6 percent (Country

Report 2008).

Due to rapid population expansion, growth in the national GDP per capita has

not kept pace with real GDP growth. Nonetheless, there has still been substantial

improvement. In 2002, GDP per capita (measured at purchasing power parity (PPP)

in current U.S. dollars) was $594; by 2008, it had increased to $1,243 (World Bank

2009). However, this is still low compared to Tanzania’s neighbors – for example, in

2008, Uganda had a GDP per capita of $1,512, Kenya had a GDP per capita of

$1,455, and South Africa had a GDP per capita of $12,574 (Human Development

Report 2009).

The Tanzanian economy is still heavily dependent on agriculture, which in

2007 accounted for just over a quarter of GDP and employed approximately 80

percent of the labor force, mostly in subsistence farming and smallholder cash-

10

cropping. Tanzania is also endowed with substantial mineral and natural resources,

such as gold, diamonds, and several other precious and semiprecious stones,

including tanzanite, a blue-purple stone unique to the country. In 2006, Tanzania

accounted for almost 2 percent of world gold production. Tanzania, which is home to

many well-known natural wonders, including Mount Kilimanjaro, Africa’s highest

peak; Lake Victoria, Africa’s largest lake; and the plains of the Serengeti, has also

benefited from significant increases in tourism – growing form 7.5 percent of GDP in

1995 to 16 percent in 2004 (World Bank 2009).

The breakdown of the economy of Zanzibar is very similar to that of Tanzania

overall, with agriculture accounting for approximately a quarter of the economy. The

Zanzibari economy, however, is very vulnerable to fluctuations in agricultural

production, especially in clove production, which still accounts for just under 25

percent of Zanzibar’s agricultural production. Food production accounts for 60

percent of all cultivated land, with the main subsistence crops being millet, maize,

sweet potatoes, bananas, cassava, peas, rice, groundnuts (peanuts) and sorghum

(World Bank 2009). Zanzibar also has an extensive local fishing industry, and the

government is hoping to develop a modern fishing fleet. Finally, tourism is becoming

an increasingly important aspect of the economy, as Zanzibar is becoming a

progressively more popular destination in Europe and East Asia, in particular

(Tanzania National Website).

Recently, Zanzibar’s economic growth has been restricted because of frequent

power outages. At the time of this study, Zanzibar was currently in the middle of a

two-month long power outage. The islands were entirely dependent on alternative

11

methods of electricity generation, primarily diesel generators. Zanzibar suffered a

similar power outage during May and June 2008. Such power outages threaten to

shock the island's fragile economy, which is heavily dependent on foreign tourism.

However, rural areas are considerably less affected, as few households have access to

electricity anyway.

a. Socioeconomic Statistics

Zanzibar’s growth, like the growth of Tanzania overall, is also constrained by

the extremely high population growth rate. Tanzania is one of only 35 countries in

world where the total fertility rate is still higher than five children per woman.

Fertility has not declined in the past 10 years, and the UN is predicting that the

population will reach 67 million by 2050 (Ellis et al. 2007, 33). Such rapid population

growth has far-reaching implications for human capital development, employment

creation, and the environment, as well as for public services and resource

mobilization. Because of the high population growth rate, Tanzania has a larger

proportion of its population in the younger age groups than in the older age groups.

With only about half of the population in the economically productive range (15-64),

a substantial burden is placed on that age group to support older and younger

household members (NBS 2005).

The high population growth rate also puts immense pressure on the education

and healthcare sectors. As a result, the country still lags behind other developing

countries in the region in terms of demographic and socioeconomic statistics. As

measured by international poverty standards, Tanzania has the highest rate of extreme

poverty in the world, with 88.5 percent of the population subsisting on less than $1.25

12

per day and 96.6 percent on less than $2 per day (World Bank 2009, ix). Tanzania is

ranked 151st out of 177 countries assessed in the United Nations Development

Program (UNDP) 2009 Human Development Index (HDI), which is based on a

number of factors, including life expectancy and adult literacy. Its position has not

improved substantially in recent years as Tanzania ranked 151st out of 173 countries

in 2002. For the sake of comparison, war-torn Sudan is currently tied with Tanzania

in the HDI rankings (World Development Report 2009).

While the overall health status of Tanzanians remains poor, major health

indicators are generally better than the Sub-Saharan African average, although they

are worse than the low-income-country average. In 2007, life expectancy at birth was

a meager 52 years, compared with 51 years for Sub-Saharan African countries on

average, and 58 for low-income countries on average. In the U.S., life expectancy is

approximately 78 years. Tanzania has made little improvement in maternal mortality

with a significantly higher-than-average rate of maternal mortality among Sub-

Saharan African countries. Tanzanian mothers die at a rate of 950 per 100,000 live

births, compared to the Sub-Saharan average of 900 and the low-income countries

average of 780. However, substantial improvements have been made in infant and

child mortality rates. In 2007, the infant mortality rate was 74 per 1,000 live births,

compared to 94 for Sub-Saharan Africa and 85 for low-income countries. The under-

five mortality rate for Tanzania was 118 per 1,000, compared to 157 for Sub-Saharan

African countries, on average, and 135 for low-income countries (World

Development Report 2009). Nevertheless, the vast majority of child deaths are still

the result of preventable illnesses, including malaria, pneumonia, diarrhea,

13

malnutrition, HIV/AIDS, and complications from low birth weights. Malnutrition also

remains a significant problem - almost four out of every ten children under the age of

five are chronically undernourished and too short for their age (stunted) and about

one out of every five children weighs too little, given his or her height. A significant

percentage of all Tanzanians (44 percent) are energy deficient and unable to

simultaneously sustain their body and carry out even light physical activity (World

Bank 2009, 8). This has detrimental implications for the growth prospects of the

country.

b. The Education System

The Tanzanian population is also poorly educated – in 2007 only 69.4 percent

were literate (World Development Report 2009). However, Tanzania has made

remarkable progress in increasing primary school enrollment in the past several years,

from 59 percent in 2001 to more than 84 percent in 2007 (NBS 2005). The structure

of the formal education system comprises seven years of primary education, four

years of ordinary level secondary school, two years of advanced level secondary

school, and up to three or more years of tertiary education. Students must past a

national standardized exam to advance to the next stage of their education. In 2008,

49.41 percent of the 999,070 students who sat for the National Standard 7 exam, at

the end of primary school, received passing marks. Ninety percent of these students

were subsequently selected to join public secondary schools in 2009 (Tanzania

National Website).

In 2002, the federal government eliminated tuition for public primary school.

However, families still have to pay for uniforms, school supplies and testing fees.

14

Secondary schools are not tuition free, but they are subsidized by the government,

allowing tuition to remain around Tsh20,000 (US$18) per year (World Bank 2009).

Additionally, the attendant fees for secondary school are often greater than those for

primary school, which when combined with the cost of tuition, prohibits many

families from sending their children to secondary school.

Swahili is the language of instruction in public primary schools. However, by

law, all secondary and tertiary education is taught in English. This policy has caused

some controversy. While some argue that English is necessary to prepare students to

compete in the global economy, others argue that forcing students to learn in English

distracts them from concentrating on the subject matter and often causes talented

students to be left behind. Students often reach tertiary school without having attained

proficiency in English, which has a detrimental effect on their higher education.

Although substantial improvement has been made in primary school

enrollment, secondary enrollment, at only 25 percent, remains low and there is a

substantial gap across income levels. The 2005 Demographic and Health Survey

(DHS) showed only 2 percent of the poorest 40 percent of students advance to

secondary school after taking a selective exam (NBS 2005). Gross tertiary enrollment

in Tanzania is also among the lowest in Africa, at 1.5 percent in 2007, compared with

3.5 percent in Uganda, 2.8 percent in Kenya and 5.1 percent for Sub-Saharan Africa,

on average (World Bank 2009, 6).

Furthermore, there is a large gap in educational attainment between males and

females in Tanzania. The median number of years of school for Tanzanian males is

3.2, which is 33 percent more than the median number of years of schooling for

15

females, 2.4. This disparity is even greater between urban and rural residents. The

median number of years of schooling is 6.1 among both urban males and females,

compared with just 2.5 and 1.5 years of schooling for rural males and females,

respectively (NBS 2005). Obviously, these numbers will need to improve if the

Tanzanian economy is going to continue to grow.

iv. Women’s Status in the Economy and their Access to Credit

Although women are responsible for much of the country’s economic activity,

especially in agriculture and informal business, economic opportunities are often

markedly different for men and women in Tanzania. Creating opportunities for

women can help to not only empower women, but also to unlock the full economic

potential of their country.

Women constitute 50.6 percent of the employed labor force in Tanzania. Their

overall labor force participation rate (including the informal sector) is 80.7 percent,

which is slightly higher than that of men at 79.6 percent (Blackden and Rwebangira

2004, 7). Despite women’s high economic participation rate, men account for 71

percent of workers in formal sector employment and are more likely to be in paid jobs

than women (Ellis et al. 2007, 4). This may be due to traditional cultural explanations

on differing roles for men and women, or to women’s lower educational attainment.

Even for the women who do have paid jobs, in most paid labor occupations, men

have substantially higher earnings compared with women. For example, in

manufacturing, the mean monthly income paid to women is Tsh42,413 (US$38),

which is approximately 30 percent lower than the average income earned by men

(NBS 2002).

16

Because of such inequities in formal employment, women often rely on

microenterprises as a means of income generation. The International Labor

Organization (ILO) estimates that the number of women entrepreneurs in Tanzania

ranges from 730,000 to 1.2 million (ILO 2003). The majority of these

microenterprises operate in the informal sector because of the difficulties of starting a

business in Tanzania. In fact, 98 percent of all businesses in the country operate

extralegally because of the obstructive regulatory and administrative obstacles to

registering, incorporating and conducting business activities (Ellis et al. 2007, 41).

Given that women have many more competing demands on their time than men,

because of domestic responsibilities, the bureaucratic hurdles of entering into

business is likely to have a disproportionately negative impact on them. Therefore,

the vast majority of women micro-entrepreneurs operate within the informal sector.

Because women tend to be less educated and because they are also subject to

cultural and religious perspectives on the kinds of jobs “acceptable” for females, they

tend to engage in more “traditional” activities, such as street vending or charcoal

production, which have a much lower profit margin. Women are often unable to

break out of the confines of these low-margin activities because they generally lack

access to formal sources of credit.

Women’s access to formal sources of credit is restricted because they often

lack collateral, the primary source of which is land. Women are estimated to own

about 19 percent of registered land, and their plots are less than half the size of those

of their male counterparts (Ellis et al. 2007, 50). Although the law guarantees

women’s right to property ownership, customary law often overrides statutory law,

17

leaving the majority of women without the required assets to provide collateral for

loans. This is particularly the case in relation to inheritance and in circumstances of

the death of or divorce from a spouse.

In the case of death, it is not uncommon for the husband’s relatives to take the

family property, including land, homes, livestock, furniture and household items, and

leave the widow and her children without any support. The widow can choose to be

inherited as a wife by one of the relatives of her deceased husband, to go back to ‘her

people’ or to live with her children. Although efforts to reform the customary law of

inheritance have been underway since 1983, the government is reluctant to force

through reforms on laws and practices that have their roots in such strongly held

traditional, cultural and religious values (Ellis et al. 2007, 52)

Divorce is similarly devastating for most women’s economic circumstances.

The Marriage Act of 1971, which in theory supersedes customary and Islamic laws,

gives women the right to retain and control their own property whether they acquired

it before or during their marriage. If property is acquired during the marriage in the

name of either the husband or the wife, that property belongs to that person to the

exclusion of the other spouse. However, given the strong cultural inhibition against

women holding property in their own name or even jointly with the husband,

properties are customarily registered in the name of the husband. When granting a

separation or a divorce, the court is required to take into account the extent of the

contributions made by each party towards the acquisition of the major assets,

including property, but this approach tends to undervalue domestic services

performed by a wife. It can be very difficult to prove her contribution to the

18

household. Moreover, the court is required to take into account the customs of the

community to which the parties belong (Ellis et al. 2007, 54). Again, like in the case

of death of a spouse, customary law often leaves women without access to land and

thus without collateral for formal credit.

Because of these additional constraints facing women and because of the

overall importance of women in the Tanzanian economy, many microfinance

institutions have made a concerted effort to include women. Furthermore, in studies

from a variety of developing countries, loans have been shown to have a greater

effect on the household and the community, as a whole, when the borrower is a

woman (Pitt and Khandker 1998, 2003; Khandker 2005; Strauss and Beegle 1996;

Hoddinott and Haddad 1994).

v. The Financial Sector

a. Formal Sector

It is helpful to divide Tanzania’s financial sector into three general categories:

formal, semi-formal and informal. The formal financial sector is comprised of

licensed commercial, regional and rural banks, which fall under the supervisory and

regulatory jurisdiction of the Bank of Tanzania. These institutions tend to be

concentrated in urban areas, especially after the financial sector restructuring in the

late 1980s. They are also primarily designed for use by the wealthier segment of

society. The poor and women, in particular, rarely rely on the formal financial sector

for a myriad of reasons, including high account opening balances, high minimum

balances, unrealistic limits on withdrawals, complicated procedures that are

19

incomprehensible for the illiterate population, inaccessibility and high transaction

costs (Mutesasira 1999, 15).

b. Semi-Formal Sector

The semi-formal financial sector includes both Savings and Credit

Cooperative Organizations (SACCOs) and NGO-based microfinance institutions

(MFIs). A SACCO is a semi-formal savings device, in which members contribute a

weekly savings share to a central fund. Eventually, the fund may be used to grant

short-term loans to members, at a chosen interest rate. Although there is a similar

mechanism within the informal sector, SACCOs fall into the semi-formal sector as

they must be legally registered with the government. While this allows for greater

scale of operations, it also involves greater transaction costs. Formal registration

requires a higher level of bookkeeping skills, which makes SACCOs less user-

friendly for the poor, who are often illiterate. Furthermore, unlike their informal

counterpart, SACCOs are not self-sustaining as they generally rely on external capital

injections from donors, rather than the savings deposits of other members (Johnson et

al. 2005).

There are over 20 NGO-MFIs in Tanzania (Mutesasira 1999, 13). These

organizations are operated by a paid professional staff and can provide more

sophisticated financial services compared to SACCOs. They are also considered to be

a lower risk alternative for the borrower, especially if the MFI is large and well-

established. NGO-MFIs operate on a much larger scale, typically serving thousands

of clients. However, because of certain government regulations, including a

requirement of owning a business to access loan services, many poor people are

20

unable to access the services of NGO-MFIs (Mutesasira 1999). These organizations

are also limited in the breadth of services they provide. They are not allowed, by law,

to accept savings deposits except those used as collateral, which denies the poor the

opportunity to save (Gallardo et al. 2005). Because they are more heavily regulated

and employ a professional staff, NGO-MFIs also involve high transaction costs. As a

result, in order to function on a sustainable basis, they tend to concentrate in more

urban or suburban areas, thereby limiting their rural outreach.

c. Informal Sector

The informal sector is by far the largest and most important in Tanzania. Out

of Tanzania’s approximately 1.8 million enterprises, only 0.4 percent obtains their

credit from formal sources and less than 0.3 percent from semi-formal sources

(Mutesasira 1999, 5). The informal sector has emerged to satisfy the financial needs

of the majority of the population, who are left behind by the formal and semi-formal

sectors.

There are a variety of mechanisms for accumulating capital available in the

informal sector. Saving at home is arguably the most prevalent savings mechanism in

Tanzania but is rarely successful as a long-term strategy because savings are

susceptible to outside demand from one’s family and neighbors. This mechanism

involves no entry barriers but high risk. Reciprocal lending among friends and

relatives is another prevalent mechanism but is on the decline because of the

increasing dishonesty and lack of trust among many people (Kashuliza et al. 1998).

This method also presents barriers because one needs to know and trust the

counterparty in order to have access to financial exchange. However, risks involved

21

are relatively low since people tend to lend only to those they know. Moneylenders

provide another alternative. They generally demand relatively high interest rates but

transaction costs are low and disbursement is normally quick. Moneylenders tend to

operate in highly localized markets and have close relationships with their debtors,

which allows for flexible lending arrangements. However, there is a general feeling

among potential borrowers that moneylenders’ behavior is exploitative and that they

should be avoided. Furthermore, there is a general understanding that moneylenders

will not lend to the poorer members of the community (Buckley 1997).

Within the informal sector there are two more formal alternatives: Rotating

Savings and Credit Associations (ROSCAs) and Accumulating Savings and Credit

Associations (ASCAs). ROSCAs, which are called upatu in Tanzania, are the

simplest form of financial intermediation. 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. The system involves a high

degree of flexibility, with the participants determining the size of the group, the

amount to be saved, the frequency of contributions, and how the funds can be used

(Johnson et al. 2005). Although ROSCAs may provide a variety of social benefits and

impose savings discipline, they do not accrue interest and therefore may be relatively

ineffective for productive investment.

An ASCA is essentially an unregistered and informal version of a SACCO.

They are very similar to ROSCAs but, like a SACCO, involve a central fund into

which the weekly contributions are deposited. Instead of the fund being automatically

22

distributed to each member in turn, members can take out loans at an agreed interest

rate (Mutesasira 1999). Members can theoretically take out a loan at any time and in

amounts aligned to their actual needs and opportunities. Furthermore, through the

interest paid on loans, members can earn a substantial return on their savings

contributions. However, ASCAs require slightly more complex record keeping than

ROSCAs.

The structure of ROSCAs and ASCAs provide a variety of benefits. Because

they are largely self-operated, transaction costs are relatively low and, therefore, they

are able to reach poorer individuals living in less densely populated areas (Johnson et

al. 2005). Both ROSCAs and ASCAs are also self-sufficient. They do not rely on

external infusions of capital, which may lead to a dependent relationship, decrease

members’ incentives to save and to monitor operations, or lead to investments in

projects that are too big and will not survive based on local demand and resources.

Risks are also relatively low because the process of self-selection allows for a high

level of mutual understanding and trust. However, the method of self-selection also

creates a risk of excluding the poorest members of society.

II. The VSLA Program

i. CARE in Tanzania

CARE is a non-political and non-sectarian, leading humanitarian organization

dedicated to the fight against global poverty. CARE was originally founded in 1945

to bring emergency relief to the survivors of WWII in Europe and East Asia, but over

the years the organization has expanded its work and now operates in more than 65

23

developing countries across the globe. CARE’s mission is to help tackle the

underlying causes of poverty so that people can become self-sufficient and live in

dignity and security. In service of that mission, CARE launched its first Village

Savings and Loan Association (VSLA) in Niger in 1991. Since then, CARE has

established more than 54,000 microfinance groups in twenty-one African countries,

serving over 1 million members (Allen and Staehle 2007).

CARE arrived in Tanzania in 1995 in order to assist with the influx of

refugees from neighboring Rwanda and Burundi. After its arrival, CARE recognized

a need for a greater variety of services, including establishing institutions to promote

sustainable development in the region. In April of 1995, in a partnership initiative

with the Department of Commercial Fruits and Forestry (DCFF), CARE established

the Jozani-Chwaka Bay Conservation Project (JCBCP) in Zanzibar, with the goal of

improving the livelihoods of communities adjacent to the area. The conservation area,

which sits about 22 miles (35km) south of Stone Town and covers approximately

6,200 acres, is an extremely rich mosaic of Zanzibar's diverse natural habitat,

including groundwater forest, mangroves, coral rag forest and salt marshes. The coral

rag zone serves as a haven for a variety of wildlife, including rare, endemic and

endangered species, such as the Zanzibar Red Colobus Monkey, Ader’s duiker and

the Zanzibar leopard (CARE Tanzania 2003).

JCBCP initiated a savings and credit scheme in August 1999 to assist

community members in financing conservation-friendly enterprise activities. The

program followed a Grameen Bank-type model in which loans were dispersed to

individuals organized in groups of five. Although the program was extended to

24

almost 720 clients by April 2000, loan repayment quickly fell below 50 percent

(CARE Tanzania 2003). Under Ujamaa, the people of Tanzania had become

accustomed to government handouts, and they subsequently misinterpreted the loans

from CARE as simply a continuation of the government program. In response, CARE

restructured the program, eventually adopting the VSLA methodology, which was

piloted in Niger.

ii. The VSLA Methodology

The costs of bringing microfinance services to Africa is often considered

prohibitive, because of the abundance of sparsely populated areas, the higher rates of

illiteracy and HIV/AIDS, and a widespread lack of identity papers, all of which serve

to increase credit risk and transaction costs. The VSLA model overcomes many of

these obstacles and promises to reach the very poor and rural population better than

formal, centralized microfinance institutions. It essentially enables the poor to

become their own bankers.

A VSLA is an Accumulating Savings and Credit Association (ASCA), which

requires no external borrowing by, or donations to, the loan portfolio – it is entirely

self-sufficient. Its work, therefore, falls within the informal sector. It differs from a

Savings and Credit Cooperative Organization (SACCO) in that it is does not receive

external funding, only training, and is not formally registered with the government,

which allows it to operate with less formal bookkeeping and thus be more user-

friendly for illiterate members. A VSLA allows for variable savings, unlimited

savings withdrawal, and loans with variable terms and flexible repayment conditions.

A single association consists of 15 to 30 people who save a small amount every week.

25

A share is usually Tsh1,000 (US$0.90) with members contributing up to three shares

every week.1 However, each group is able to determine their own share value and the

maximum number members can contribute each week. The value of each share

remains low so as 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 and Staehle 2007).

In addition to the savings fund, the cash box holds the social fund and the

education fund. The social fund is a self-insurance mechanism, which can provide

members with a small amount in the case of emergencies. Each member contributes a

set value every week, usually between Tsh200 (US$0.18) and Tsh400 depending on

the group. In the event of an emergency such as a fire or the death of a family

member, the fund dispenses a fixed amount, generally between Tsh10,000 (US$9)

and Tsh20,000 (US$18). No interest is charged for loans from the social fund and,

although members are expected to pay back the loans, repayment is not strictly

enforced. The social fund is managed separately from the savings and loan fund and

is not shared out at the end of the cycle and is thus carried over to the next cycle. Like

the social fund, weekly contributions to the education fund vary between groups, but

generally fall in the range of Tsh100 to Tsh200 per week. The education fund is used

to pay the nominal monthly training fee and to buy the necessary materials, such as

passbooks and the lockbox.

1 In a county where the average weekly income is around Tsh26,400 (US$24), this represents a saving rate of approximately 8 percent. For the sake of comparison, the U.S. savings rate is around 5 percent.

26

After several months, the savings shares 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

percent per month – this is low compared to moneylenders who often charge up to 30

percent per month, but slightly higher than NGO-MFIs, which generally charge less

than 4 percent per month (Mutesasira 1999, 10). Each group is able to set their own

repayment terms. However, a VSLA never fines borrowers for late loan repayment as

this may aggravate any underlying crisis the household may be facing. It is assumed

that the embarrassment of being late is sufficient penalty (Allen and Staehle 2007,

10).

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 so as 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.

27

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 the starting and closing balances of the social

fund as well as loan disbursement is recorded in the group ledger (Allen and 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 organization.

CARE’s role is to supply extensive training on group dynamics, governance and

money management. VSLA training is based on a four-phase curriculum. During the

first phase - an intensive, three-month period - a field officer from CARE visits the

group every week and holds training sessions on group dynamics. The field officer

also selects and trains a Community Contact Person (CCP or village trainer) who

lives in the target community. The CCP is paid by the VSLA not by CARE. In the

second phase, the field officer visits the groups once or twice a month as they begin

to rely more on the CCP. In the third phase, after approximately a year of supervision,

if the CCP passes a certification test, the field officer will move on to another area

and start the process again. In the fourth phase, in the original VSLA model, once a

group is mature, it can function with no external support (Training Guide… 2004).

However, CARE quickly realized that even mature groups could benefit from

additional monitoring and technical support. In order to respond to such needs while

maintaining a degree of self-sustainability, CARE developed a system known as an

Apex Organization, which would support and monitor existing groups while fostering

the growth of new groups, allowing CARE to move onto new areas.

28

iii. Apex Organizations and the Sustainability of the VSLAs

As CARE prepared to leave Jozani Bay in 2003 after two years of successful

VSLA implementation in the area, the organization, together with the CCPs,

developed the Jozani Credit Development Organization (JOCDO). Today, JOCDO is

often sited as the prime example of an Apex organization. The Apex model, generally

lauded as a paradigm of sustainability, has been expanded throughout Tanzania and is

beginning to move across Sub-Saharan Africa.

An Apex organization is owned by the affiliated VSLA groups. It provides a

number of services to member VSLAs, but, like CARE, it does not provide direct

loan capital. The objectives of an Apex organization are: to serve as an umbrella

organization by promoting and protecting the interests of existing VSLA groups; to

assist in formation of new groups; to provide VSLA kits (cash boxes, pass books,

etc.) and other materials; to supervise the quality and standards of performance of

affiliated VSLA groups; to monitor groups and collect monthly data; and, finally, to

assist affiliated VSLAs through capacity building.

The Apex organization is responsible for the selection and training of new

Community Contact Persons (CCPs), who, in turn, are responsible for the formation

and training of new VSLA groups. The training of new groups under the Apex

organization essentially follows CARE’s four-phase curriculum described above. The

greatest difference is that after a group is fully mature, they continue to receive

monthly visits from a CCP, either the original CCP who carried out the initial training

or the newly trained and certified CCP from the local community. At this point, the

role of the CCP becomes one of monitoring and technical assistance. He is expected

29

to visit a mature group once a month to ensure quality control and before the final

payout of each cycle.

The organizational structure of an Apex organization is made up of a General

Assembly, a Board of Trustees, an Executive Committee, and a Director. However,

the VSLAs are always the main building blocks of an Apex organization (Mkoma

2009). Currently, JOCDO maintains a board of trustees selected from a group of

important members of the community; however, the board members are not involved

in the operations of the organization. All decisions are carried out by the executive

committee, which has twenty members, including a chairperson, secretary, assistant

secretary, treasurer, assistant treasurer, and an executive director. The position of

executive director is currently empty and will not be filled again until the election in

October 2010.

Every Apex organization, including JOCDO, is sustained by several income

sources. Each VSLA group pays an initial entry fee of Tsh60,000 (US$55) to the

Apex organization. Following the first year, the annual subscription fee for member

groups is Tsh15,000 (US$14) per year. In most groups, members contribute up to

three shares of Tsh1,000 (US$0.90) every week, yielding a total possible share value

of Tsh156,000 (US$142) per member and Tsh4,680,000 (US$4,255) per group.

Therefore, the initial entry fee represents, at a minimum, 1 percent of a group’s

annual savings, while the yearly subscription fee corresponds to 0.3 percent of a

group’s savings. As there is no external funding, these nominal fees denote the only

transaction costs of the VSLA program. The subscription fee covers basic support and

monitoring services. There are currently 233 VSLA groups in Zanzibar. However,

30

only 106 of these groups are registered and paying members of JOCDO. As JOCDO

still provides a variety of support services for non-registered members, this presents a

major financial obstacle for JOCDO. However, JOCDO may also make a profit from

the sale of VSLA kits to groups at a small margin, usually around Tsh10,000 (US$9),

and also draws a cut of the training fee paid to the CCP from each VSLA – for

example, from the Tsh3,000 (US$2.70), Tsh2,000 (US$1.80) is pocketed by the CCP

and Tsh1,000 (US$0.90) is sent to the Apex organization (Mkoma 2009).

31

CHAPTER 2 Literature Review

As interest in microfinance has grown over the past three decades, so has the

compilation of related literature. The major findings of the key studies in the field are

presented below. These findings are summarized in Appendix A for convenience.

I. Impact of Microfinance

i. Financial Assets

Most studies have found that microfinance allows the poor to protect,

diversify and increase sources of income, which helps to smooth out income

fluctuations and to maintain consumption levels even during times of crisis. Zaman

(2000), who examines the Bangladesh Rural Advancement Committee (BRAC)’s

impact on the welfare of its clients, finds that participation in micro-credit programs

reduces vulnerability by smoothing consumption, building assets, providing

emergency assistance during natural disasters, and empowering females. The

methodology of each study will be discussed in greater detail in Chapter 3, but the

results are generally considered robust. In addition to using a control group, Zaman

uses a Heckman two-step procedure, an advanced econometric technique, to control

for any biases in his estimation.

32

MkNelly and Dunford (1999) also find a positive impact on income. They

control for potential biases by assigning communities to either a program or control

group following baseline data collection, thereby allowing program impact to be

measured through simple comparison between the treatment and the control group.

Their results show that the majority of participants (67 percent) of the CRECER

Credit with Education Program in Bolivia feel that their incomes have ‘increased’ or

‘increased greatly’ since they joined the program. Additionally, MkNelly and

Dunford find that clients of Lower Pra Rural Bank Credit with Education Program in

Ghana have increased their incomes by $36 compared to $18 for non-clients. Clients

have also significantly diversified their income sources – eighty percent of clients

have secondary sources of income compared to fifty percent of non-clients. Dunn and

Arbunkle (2001), who control for potential biases with the use of a control group and

a combination of advanced quantitative and qualitative methods, find that

microfinance clients in Lima, Peru have over 50 percent higher income than non-

participants.

Household income is often very difficult to measure in a survey format.

Therefore, household expenditure level is often used as a substitute for income to

determine overall program impact. Pitt and Khandker (1998) find that for participants

of the Grameen Bank, the Bangladesh Rural Advancement Committee (BRAC),

annual household consumption expenditure increases 18 taka for every 100 additional

taka borrowed by women, compared with 11 taka when the borrowers are men. They

control for participation endogeneity by using the specific design of the credit

programs to identify the effect of program credit, by gender of participant, in a

33

limited-information-maximum-likelihood framework, and by controlling for

nonrandom program placement by using village-level fixed effects. Khandker (2005),

using a household-level fixed-effects model with panel data, which resolves both

household- and village-level endogeneity, finds that Pitt and Khandker (1998)

actually underestimated program impacts. He finds each additional 100 taka of credit

to women increased total annual household expenditure by more than 20 taka, but

finds no returns to male borrowing at all.

A few studies, however, have failed to find positive impacts on income from

microfinance participation. Masanjala and Tsoka (1997) find little impact of FINCA-

Malawi on living standards and expenditure patterns. Ssendi and Anderson (2009)

also find little long-term effect, as measured by increases in household assets.

However, both studies use a much less robust methodology and make little attempt to

control for selection bias.

ii. Poverty

A number of studies have found that access to microfinance services

decreases the incidence of poverty. Dunn and Arbunkle (2001) find that only 28

percent of microfinance clients in Lima, Peru live below the poverty line compared to

41 percent of non-clients. Khandker (2005) also finds positive effects on poverty

rates. He finds that between 1991/92 and 1998/99, moderate poverty in all villages

declined by 17 percentage points: 18 points in areas where Grameen Bank or BRAC

was active, and 13 points in non-program areas. Among program participants who

had been members since 1991/92, poverty rates declined by more than 20 percent –

about 3 percentage points per year. Khandker estimates that more than half of this

34

reduction is directly attributable to microfinance, and finds the impact to be greater

for extreme poverty than moderate poverty. Khandker further calculates that

microfinance programs reduce average village poverty level by one percentage point

each year in program areas. Microfinance thus helps not only poor participants but

also the local economy. Overall, Khandker finds that microfinance accounts for 40

percent of the entire reduction of moderate poverty in rural Bangladesh.

iii. Quality of Housing

Considering the difficulty in obtaining other measures of welfare, such as

income or even expenditure, in the majority of developing countries, the quality of

housing is often used as a proxy for a household’s socio-economic status. Overall, the

literature suggests a positive impact of microfinance program participation on both

the quality of housing as well as the level of investment. Hossain (1988), who

compares Grameen Bank members to both eligible non-participants in Grameen

villages and target non-participants in comparison village, finds that members spend

six times more on housing investments than non-members.

Neponen (2003), who uses a control group of new members to avoid selection

bias while monitoring the performance of microfinance program participants in

Trihcirappalli, India, finds that members of the microfinance program live in much

higher quality housing. Sixty-four percent of members live in tile roof and concrete

houses, which is considered to be the highest quality material available, compared to

only 50 percent of new members (the balance live in mud and thatch houses).

35

iv. Education

In general, studies have found a positive impact of microfinance program

participation on education - children of microfinance clients are more likely to go to

school and stay in school longer (Neponen 2003; Littlefield et al. 2003). Barnes

(2001), who, like Dunn and Arbunkle (2001), controls for potential biases with the

use of a control group and a combination of advanced quantitative and qualitative

methods, finds that the Zambuko Trust program in Zimbabwe has a positive impact

on the education of boys aged 6 to 16. However, the program has no effect on the

education of girls within the client-household. Pitt and Khandker (1998), however,

find that microfinance program participation increases the probability of enrollment

for girls. On the other hand, Coleman (1999), who controls for participation

endogeneity through the use of a quasi-experimental design, finds little impact on

education expenditures, which may be seen as a proxy for either access to or quality

of education.

v. Nutrition and Health

Households of microfinance clients, particularly those of female clients,

appear to have better nutrition and health statuses compared to non-client households

(Pronyk et al. 2007; Littlefield et al. 2003; Hossain 1988). Pitt et al. (2003) find that

women’s credit has a large and statistically significant impact on two of three

measures of children’s health. A 10 percent increase in credit provided to females

increases the arm circumference of their daughters by 6.3 percent - twice the increase

that would be expected from a proportionately similar increase in credit provided to

men. Female credit also has a significant and positive, but somewhat smaller effect on

36

the arm circumference of sons. Female credit is estimated to have large, positive and

statistically significant effects on the height-for-age of both boys and girls. However,

no statistically significant effects are found for body mass index (BMI) of boys or

girls.

Barnes (2001) finds that participation in Zambuko Trust in Zimbabwe has a

positive impact on the frequency with which food is consumed in extremely poor

households as well as on the quality of food. Specifically, participation has led to a

positive impact on the consumption of high protein foods (meat, fish, chicken and

milk). MkNelly and Dunford (1999) also find that children of participants of the

Lower Pra Rural Bank Credit program in Ghana experience significant improvements

in feeding frequency compared to children of non-clients. However, positive impacts

on the nutritional status of clients of the CRECER Credit program in Bolivia and their

children are not evident. Deeper analysis of the client group alone, however, reveals

that children’s weight-for-age is positively related with the quality of education

services provided. This finding suggests that without important improvements in

caregiver practices, increases in income and even empowerment are unlikely to bring

about marked improvement in children’s nutritional status.

vi. Empowerment and Social Status of women

Numerous studies have found that targeting women as clients is an effective

method of ensuring that benefits of increased income accrue to the general welfare of

the family (Pitt and Khandker 1998, 2003; Khandker 2005; Strauss and Beegle 1996;

Hoddinott and Haddad 1994). Such gender-targeted microfinance has also been

37

shown to have a positive effect on the empowerment and equality of women

(Mwenda and Muuka 2004).

Hashemi, Schuler and Riley (1996), in an attempt to deal with the

complexities and ambiguities of the meaning of empowerment, create a composite

empowerment indicator based on eight components: mobility, economic security,

ability to make small purchases, ability to make larger purchases, involvement in

major household decisions, relative freedom from domination within the family,

political and legal awareness, and involvement in political campaigning and protests.

A woman is considered empowered if she scores positively on 5 out of the 8

components. Using a combination of sample survey and case study data and

controlling for selection bias by statistically controlling for differences in

demographic characteristics such as age, education and wealth, Hashemi et al. find

that membership in either the Grameen Bank or the BRAC has significant effects on

all eight dimensions. They find that each year of membership in either program

increases the likelihood of a female client being empowered by 16 percent. Even

women who do not participate in the program are more than twice as likely to be

empowered simply by living in Grameen villages. The authors argue that credit

programs empower women by strengthening their economic roles, increasing their

ability to contribute to their families’ income, enabling them to establish an identity

outside of the family, and giving them experience and self-confidence in the public

sphere.

Terry (2006) finds that loans from FINCA-Tanzania create major positive

changes in the lives of female borrowers, including an improvement in social status

38

and self-esteem, and an increase in confidence. Women also feel empowered through

an increase in income and the ability to accumulate savings, purchase household

assets and contribute towards children’s education. The findings also suggest that

members of the household and the community, at large, view female participants in a

more positive way. However, Terry relies completely on qualitative data and does not

include a control group. Therefore, the results of the study are not necessarily reliable.

II. Microsaving

Recently, practitioners have begun to increasingly acknowledge the

importance of savings mechanisms. Research has even found that most people prefer

savings to credit (Hirschland 2005). Furthermore, small loans are not always

appropriate for poor women (Kabeer 2001). A loan becomes a debt, and the poor

often face a crisis if an expected source for repayment evaporates. Therefore,

borrowing is often much riskier than saving. Because starting a new business is risky

and sustainable providers of credit cannot afford to lose money, credit is generally not

used to start a new business but rather to expand an existing one. Therefore, most

people must rely on savings to start up new business ventures. Savings enables future

investment, by giving access to lump sums of money. These large sums of money can

be used for investment opportunities, for life cycle events, such as marriages,

funerals, etc., or for emergencies. Savings can also be used to smooth consumption.

Furthermore, while borrowers pay interest, savers can earn interest. Finally, although

not everyone is creditworthy or is willing to take such risk, all people are deposit-

worthy and want to develop assets.

39

Savings clearly offers substantial benefits and correspondingly, in general,

savings programs have been shown to have a positive impact on participants. Dupas

and Robinson (2009), who used a unique study design that controlled for potential

biases while allowing for the use of simple regression analysis, find that access to a

formal savings account has substantial positive impacts on women’s productive

investment levels and expenditures, and also makes women less vulnerable to shocks

from illness.

Chen and Snodgrass (2001), who rely on a similar method to that of Barnes

(2001) and Dunn and Arbunkle (2001) to control for potential biases, also find a

positive impact of savings. Although the impact of savings is smaller than that of

borrowing, income of savers is more stable than that of borrowers. Chen and

Snodgrass compare the impact of SEWA Bank on clients who borrow to those who

save without borrowing, and compare both groups to non-clients (who are drawn

randomly from women engaged in the informal sector in the same neighborhood as

clients in Ahmedabad, Gujarat, where SEWA is based).

In round 1, the borrowers were shown to be considerably better off than

savers, who were in turn better off than non-participants. Some of these differences

may be attributable to participation in SEWA prior to the round 1 survey. However,

between the two rounds, the savers showed the fastest rate of income growth. Still,

borrowers income was over 20 percent greater than that of savers, and 40 percent

higher than that of non-participants’. Savers, however, enjoy an income, which is 20

percent greater than that of non-participants. For borrowers, the findings show a

mixed report of the impact on poverty - the numbers of households with incomes

40

below $1/day and those above $2/day (the World Bank’s official cut-off line for

‘absolute poverty’ and ‘moderate poverty’ respectively) both increased between

rounds. Borrowers experienced the largest increase in the number of non-poor

households between rounds, but they also had the most households that slipped to a

lower poverty category. Saver householders were less volatile with the numbers in

both the $1-2 range and above $2 rising, demonstrating again the lower risk involved

in saving. Overall, the results suggest that the use of either credit or savings services

raises household income, both total and per capita. The use of financial services, in

general, is also associated with increased spending on housing improvements,

consumer durables and school enrollment, especially for boys.

III. ROSCA/ASCA Participation

Recently, an increasing amount of literature has attempted to specifically

explain ROSCA participation. As there is no interest to be gained by saving in a

ROSCA, the question is, why do individuals choose to save through a ROSCA

instead of individually accumulating savings? Besley, Coate and Loury (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. Anderson and Baland (2002) argue that

ROSCA participation is a strategy used by women to protect their savings against

claims from their husbands. Dupas and Robinson (2009) expand upon this theory by

suggesting that women also face constant demands from other relatives and neighbors

and may find it difficult to refuse requests if the money is available in the house.

41

Bauer and Morduch (2008), Gugerty (2007), and Dagnelie and 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. Although ASCAs,

unlike ROSCAs, do generally provide interest on savings that comes from interest

payment on loans, it is often a small amount. As such, much of the literature on

ROSCA participation probably can be applied to participation in ASCAs as well.

IV. VSLA Performance

Despite the apparent success of the VSLA model, few detailed studies of the

model’s performance have been undertaken. Allen and Hobane (2004) conclude that,

in Zimbabwe, membership in a VSLA contributes to an increase in household

productive and non-productive asset levels among the majority of participants, as well

as to some improvement in quality of housing. The findings also suggest that program

participation has led to an increase in the number of income-generating activities

(IGAs) and to an increase in stability of such activities. Households also allocate

more labor to IGAs. Furthermore, 81 percent of respondents feel that their status in

community has improved. However, it is difficult to attribute these results to the

interventions of the VSLAs alone. The study has no control or comparison group and

relies on recall data for a period of four years, which may not yield accurate

information, as people tend to forget what their status was four years ago.

However, Anyango (2005) reaches similar conclusions as Allen and Hobane

(2004) when studying the VSLA program in Malawi. He finds that program

participation has helped to improve the livelihoods of its members and to alleviate

42

poverty, particularly for women who constitute the majority of the groups. Number

and magnitude of economic activities has increased as a result of participation in the

program. However, members have divested away from certain economic activities

that require greater capital. The study also does not have a control group, although it

does have a baseline. The baseline, however, is taken at the community level.

Therefore, there may be a selection bias if the members of the VSLA systematically

differ from the members of the community.

There have been two major studies done on the VSLA program in Tanzania.

The most extensive study is the Women’s Empowerment Strategic Impact Inquiry

(SII), which was completed in 2006. The study incorporates a quantitative

questionnaire, which was given to 181 women, including 134 VSLA members and 47

non-members. In addition to several case studies, a series of focus group discussions

were carried out to explore more deeply the issues raised in the questionnaire.

Baseline data was not available, but the authors attempt to address the problem by

asking questions linking participation to changes in the impact variables, and

requesting respondents to compare their current situation to their situation prior to

joining VSLA group. The greatest weakness of the study is that the quantitative data

are not tested for statistical significance. The assessment of significance is therefore

based on the judgment of the research team and is thus largely subjective.

The study finds, in terms of short-term economic empowerment, VSLA

women benefit more than non-VSLA members from increased savings, more IGAs,

greater food security and health, and increased education expenditures. Most VSLA

women (75 percent) have increased their savings since joining VSLA group. VSLA

43

women are more likely (68 percent) to be engaged in an IGA than non-VSLA women

(13 percent), with most VSLA women reporting the VSLA group as the source of

funds for their IGA. Furthermore, VSLA women’s households experience greater

food security compared to non-VSLA households. More VSLA households also

report a great improvement in the quantity and quality of meals over the last 3 years,

compared to non-VSLA households. Thirty-four percent of VSLA households report

improved family health over the last three years compared to 22 percent of non-

VSLA households. About twice as many VSLA households (57.3 percent) as non-

VSLA households (30.4 percent) report an improvement in the education status of

family members over the past three years. A higher proportion of VSLA households

(79.9 percent) have made expenditures on education over the last 12 months, than

have non-VSLA households (65.2 percent).

In terms of long-term economic empowerment, the SII finds more VSLA

women own household and productive assets than non-VSLA women. For each asset

category, between 40-77 percent of the women have acquired assets with funds (loans

or payouts) directly from the VSLA group. The other women have acquired assets

with revenue from IGA or some other source. Moreover, compared to non-VSLA

women, almost twice as many VSLA women have made investments in housing

during the last three years. Although VSLA membership is not a necessary condition

for women’s investment in long-term assets, the study indicates that VSLA

participation increases women’s chances of making such investments.

In the case of women’s social empowerment, the SII study finds that VSLA

women demonstrate more confidence than their non-VSLA counterparts and appear

44

more motivated to take action to improve their lives. VSLA women also have more

freedom to participate in community social activities than non-VSLA women.

Finally, VSLA women have more control over decisions to engage in income-

generating activities and to spend time income accruing than do non-VSLA women.

The other major study in Tanzania, and the precursor to this study, was

Anyango et al.’s (2006) examination of the performance of VSLA groups in

Zanzibar. The study, however, does not have a baseline and does not use a control

group. No tests of statistical significance were performed. The study finds that

VSLAs in Zanzibar have performed well in terms of growth and sustainability. Total

membership rose 258 percent from 2002 to 2006. They have also performed well in

terms of profitability – during the last payout for all 25 groups, members received up

to a 53 percent rate of return on savings. Respondents also name improved standard

of living (22 percent), improved housing (21 percent) and increased income (20

percent) as three major changes as a result of VSLA program participation.

Although the results from Anyango et al. (2006) are promising, the study

suffers from several methodological weaknesses and therefore its results are not

entirely reliable. This study therefore aims to expand and improve upon this work,

and the entire set of literature concerning VSLA performance and microfinance

impact.

45

CHAPTER 3 Research Design, Methods, & Sample

I. Impact Assessment Methodologies

From the wealth of literature referenced above it is clear that in the last fifteen

years impact assessment has become an increasingly important aspect of development

activity as international agencies and aid donors, in particular, have sought assurance

that aid funds are well spent. There are several methodological options for conducting

impact assessments, which can be roughly grouped into two different paradigms: the

scientific method and the humanities tradition (Hulme 2000).

The scientific method seeks, through experimentation, to ensure that

outcomes can be directly attributed to inputs. In the social sciences, however,

controlled experiments are difficult and often impossible to arrange. Therefore, most

social scientists have come to rely on the control group method, which involves

comparisons between a ‘treatment’ group and an identical group (or as nearly

identical as possible) that did not receive the treatment. This method allows for

stronger estimations of program impacts and more robust conclusions of causality.

By contrast, the humanities tradition eschews statistical tools and “proof” of

impact. Rather, the humanities tradition seeks to interpret the processes involved and

to explore the range of plausible impacts, using mainly qualitative data. Although it is

considerably more difficult to quantify impacts of program participation and to

46

attribute cause and effect, the power of the humanities approach lies in its ability to

present a thorough report, describing and examining the process itself, rather than just

the outcomes. The method grants the reader the opportunity to explore different and

often conflicting accounts of the processes involved and what the program has

accomplished. Overall, the humanities approach, though it may be criticized to lack

of rigor and excessive subjectivity, may facilitate a deeper understanding of the

dynamics of program participation (Hulme 2000).

The scientific method, with its power of quantification, is often considered to

be more robust than the humanities approach. However, there is a potential problem

inherent in the scientific method - selection bias, from both self-selecting into the

program as well as non-random program placement, must be addressed before the

analysis and results can rightfully claim the scientific rigor necessary to be useful.

i. Selection Bias

The primary source of selection bias in impact assessments is self-selection

into the credit or savings program. Self-selection bias may occur if members of the

treatment group systematically possess unobserved attributes, which those in the

general population lack – such as entrepreneurial drive or ability, superior health, or

specific preferences – which make the results of the program difficult or impossible

to generalize to a broader population of potential participants. For example, if

program participants are naturally more entrepreneurial or more dedicated than non-

participants, program impacts may be vastly overestimated. However, self-selection

bias may go in the opposite direction as well. For example, Pitt and Khandker (1998)

find that poorer households are more likely to be Grameen borrowers than their

47

neighbors. This may lead to a downward bias on the estimated effect of the program,

giving the impression that participation in the program makes clients poorer relative

to the population as a whole.

The second source of selection bias is non-random program placement. Many

programs are set up in convenient locations where there is complementary

infrastructure or previous program activity. If the pre-existing factors tend to make

area residents better off, with or without the implementation of the program in

question, this may lead to an upward bias in the estimates of program impact.

Alternatively, if programs are designed specifically to assist the underserved, the poor

and the disenfranchised – for example, in predominantly rural areas – the relative

disadvantages of program participants, compared with the larger population, may lead

to apparent negative impacts relative to the control group (Morduch 1999).

Selection bias may be addressed in the design of the study. One option is the

use of a quasi-experimental technique, which attempts to simulate the situation that

would have prevailed if there had been no credit program. Baseline data are collected

and then participants are randomly assigned to a treatment or to a control group,

thereby guaranteeing random selection into the program and more robust conclusions

of program impact. However, these studies are generally very time consuming and

expensive, since the study has to be conceived, designed and implemented over the

life of the program being study. Alternatively, many quasi-experimental studies rely

on an exogenous, one time, expansion of a credit or savings program and are thus, not

easily replicable.

48

The method most commonly employed to address selection bias is by use of a

carefully selected control group. The control group is often randomly drawn from

other members in the community who are eligible to participate in the credit or

savings program. To control more stringently for potential systematic differences in

unobserved attributes, many studies employ a control group of “clients-to-be,” who

have been accepted into the credit program but who have not yet received

microfinance services. This method assumes that all participants – those who have

not received the treatment but are about to, as well as those in the treatment group –

share the same, or at least similar, unobserved characteristics, since they have all

joined the program at one point or another. However, it is still subject to a myriad of

potential biases, including the impact of time-varying unobservables. Complex

econometric techniques are thus often used, in combination with the use of a control

group, to further control for selection biases.

Basic regression analysis allows for direct measurement of program impact on

a specific set of outcome variables, while controlling for (observable) individual or

household characteristics that might also impact the outcome variables. However, this

approach cannot control for the unobserved characteristics, which may cause

selection bias. In fact, the model relies on an assumption that these unobserved

attributes are uncorrelated with membership in the credit or savings program, an

assumption which may be (and often is) violated in reality.

However, there are three more advanced econometric techniques, which do

not rely on this assumption, while addressing the issue of selection bias (Hulme

2000). These methods are briefly summarized here but are discussed in greater detail

49

in the next section through examples in the literature. The first method assumes an

error distribution, typically assumed to be normal, of the outcome variable without

treatment. The effect of treatment is then determined by measuring the deviations

from normality of the outcome within the treatment group. However, there are several

problems inherent in this method. First, there is usually no good basis on which to

make the initial assumption about the error distribution, and the results are highly

sensitive to such assumptions. Moreover, in the case of censored dependent variables,

identification of the treatment effect is sometimes still impossible. The second

standard econometric method to control for selection bias is by use of instrumental

variables. The identifying instrument would have to be a determinant of joining the

credit program, but not a determinant of the outcome variable, such as household

income or expenditure levels. Such instruments are very difficult to find. Given the

limitations of the first two econometric methods, the third option, the use of panel

data, is often considered the best means to control for selection bias. Panel data allow

for the use of the household-level fixed-effect method, which resolves both

household- and village-level endogeneity. However, panel data are difficult and

expensive to collect, and as such only a few studies have been able to use this

method.

ii. Examples in the Literature

Previous studies have employed a variety of methods to deal with the inherent

problems involved in impact assessments, particularly selection bias. These studies

and the employed methods are summarized in Appendix A. The most

methodologically complex and time-intensive studies have chosen to rely solely on

50

the scientific method, and subsequently attempted to control for selection bias by

using large data sets with carefully constructed control groups, and advanced

econometric techniques.

Zaman (2000), when examining the impact of the Bangladesh Rural

Advancement Committee (BRAC), uses the Heckman two-step procedure to control

for selection biases. Zaman relies on a large cross-sectional data set, consisting of

1,072 individuals, including 547 members of BRAC and a control group of 525

eligible non-members in ten villages where BRAC operates. The first stage of the

Heckman two-step procedure models a participation equation, which attempts to

capture the individual, household and village characteristics that affect the probability

of participation in the program. From the coefficients of the participation equation, he

then derives maximum livelihood estimates. These estimates are then used to

construct a selectivity term known as a Mills ratio. The second stage involves adding

the Mills ratio to the consumption equation and estimating the equation using

Ordinary Least Squares (OLS). If the coefficient of the selectivity term is significant,

then the hypothesis that an invisible selection process biases the initial participation

equation is confirmed. If the coefficient of the selectivity term is insignificant, OLS

estimates can safely be used for the model.

A major problem with the Heckman procedure, however, is identification. An

appropriate identification variable needs to influence participation but not poverty,

and this variable is not easy to find. Zaman attempts to use the number of eligible

households in each village in 1992. His reasoning is that a larger number of potential

51

members in a village reduces the chance of any one eligible household participating

in BRAC, but should not affect individual household’s poverty status.

Zaman then utilizes the identification on functional form procedure to test for

the sensitivity of the estimates from the Heckman procedure to the particular

identification variable used. The procedure exploits the fact that the Mills ratio term is

a non-linear function of the exogenous variables used in the first stage equation.

Therefore, all variables in the first stage equation can enter the second stage, along

with the Mills ratio term, in order to identify the selectivity effect. Identifying on

functional form also allows Zaman to incorporate village-level effects by including a

dummy variable for each village. The results of the functional form procedure raises

doubts as to whether eligible number of households is an appropriate identification

variable. Zaman then asks, in view of the lack of an ideal way of correcting for

selectivity, whether it is better simply to use OLS. The two different methods produce

slightly different results, and therefore, Zaman leaves the final decision to the reader

as to which econometric specification is more valid.

Pitt and Khandker (1998) rely on an even larger cross-sectional data set,

composed of 1,789 households, including 1,538 “target” households, meaning they

live in villages where one of three credit programs (Grameen Bank, Bangladesh Rural

Advancement Committee (BRAC), or Bangladesh Rural Development Board’s

(BRDB) Rural Development RD-12 program) is available, and 260 “non-target”

households. Among the target households, 905 were credit program participants.

Analyzing program impacts by comparing households in villages with credit

programs and households in villages without programs suffers from the possibility

52

that program placement is endogenous. Pitt and Khandker therefore use a village-

level fixed-effects method to avoid the problem of village unobservables biasing

estimates of the impact. However, even with village-fixed effects, the endogeneity

problem still remains if there are common household-specific unobservables affecting

demand for credit and household outcomes. In order to address this issue, Pitt and

Khandker constructed the sample survey to allow creation of an identifying variable.

The effect of participation on the specified outcome can be identified if the

sample also includes households in program villages that are excluded from making a

treatment choice by either random assignment or some exogenous rule. Pitt and

Khandker use the restriction that any household owning more than half an acre of

land is excluded from joining any of the three credit programs as their exogenous

rule. The programs’ effect on the specified outcome is then estimated using a limited

information maximum likelihood framework by comparing the outcome between

households with a program choice and households without a program choice,

conditioning on village fixed effects and observed household and individual

attributes. Pitt and Khandker then estimate a reduced-form credit equation

disaggregated by gender in order to identify the impact of gender-specific credit.

Because men can join only men-only groups and women can join only women-only

groups, the gender-based restriction is easily enforceable and thus observable.

In order to demonstrate the importance of controlling for endogeneity, Pitt and

Khandker also present alternative estimates of the programs’ impact on a variety of

household and individual behaviors using simpler approaches that do not control for

varying levels of endogeneity. A comparison of their more advanced econometric

53

method with the simpler alternative approach clearly indicates the importance of

attentiveness to endogeneity.

However, Morduch (1998) has criticized Pitt and Khandker’s (1998)

methodology. When the data are examined more closely, he finds that over 30 percent

of Grameen borrowers and 28 percent of BRAC borrowers are above the half-acre

cut-off. Some borrowers from Grameen hold over five acres. Nonparametric

regression yields no obvious discontinuity in the probability of borrowing for

households across the relevant range of landholdings. When Morduch looks at the

data again, focusing on simple comparisons across treatment and control villages,

while controlling for household-level and village-level characteristics, he finds little

program impact, considerably different results from those of Pitt and Khandker.

In order to address these concerns, Khandker (2005) re-examines the issue

using panel data drawn from the same households surveyed in Bangladesh in

1991/92, which were then surveyed again in 1998/99. He examines whether the

estimated impacts of microfinance found in the earlier study using cross-sectional

data analysis can be corroborated using an alternative method. Khandker begins by

classifying each village by women-only and men-only groups, which helps to identify

program impacts by gender. Again, the village-level fixed-effect method can resolve

the problem of endogeneity of program placement, but another exogenous eligibility

requirement is still needed at the household level to determine why certain

households participate while others do not. Given the sensitivity of the results to the

instrument used, Khandker uses an alternative method – household-level fixed

54

effects, using panel data. This method appears to resolve both household- and village-

level endogeneity.

The household-level fixed-effect method still may not yield consistent

estimates of credit impacts for two reasons: the unobserved factors at the household

and village level may vary over time; and if credit is measured with errors, the error

gets amplified when differencing over time, especially with only two time periods.

This measurement error will bias the impact estimates toward zero. A standard

correction for both types of bias is the reintroduction of instrumental variable

estimation. Khandker develops the instrument by interacting the choice variable for

both 1991/92 and 1998/99, to deal with the differential impacts of the two periods,

with household-level exogenous variables and village fixed effects. A specification

test (Wu-Hausman test) is then performed to determine whether the household-level

fixed-effect or the household-level fixed effect instrumental variable method is more

appropriate to estimate household consumption behavior. The test results suggest that

the credit volume, as used in the fixed-effect method, is not endogenously determined

by factors such as the time-varying heterogeneity or the measurement errors

associated with credit variables, and, therefore, the household-level fixed-effect

method is more appropriate. The results from Khandker (2005) are widely accepted

as valid, as it is generally considered to be the most econometrically advanced study

to date.

Several studies have been able to produce similarly robust results while using

less advanced econometric techniques by exploiting exogenous program expansion.

Coleman (1999) circumvents issues of self-selection and endogenous program

55

placement by using data from a quasi-experiment conducted in northeast Thailand in

1995-96. Member and nonmember households in 14 villages were surveyed four

times over the course of a year. At the time of the first survey, seven of the villages

had had a village bank for 2-4 years, and one village began its village bank

immediately after the first survey. Six “control” villages were identified to receive

NGO support for a village bank to begin one year after they were identified. In these

control villages, villagers were allowed to self-select to be village bank members or

nonmembers. The “old” village bank members in the eight “treatment” villages can

then be compared with the “new” village bank members in the six control villages.

This survey design allows for the use of relatively straightforward estimation

techniques.

Coleman estimates a single impact equation with a vector Xij of household

characteristics, a vector Vij of village characteristics, a membership dummy variable

Mij equal to 1, if a household self-selects into credit program, and 0 otherwise, and

another dummy variable Tij, which is equal to 1, if a self-selected member already

had access to program loans, and 0 otherwise. The membership dummy Mij can be

thought of as a proxy for the unobservable characteristics that lead households to self-

select into the village bank. Its coefficient can thus be interpreted as the impact on

outcomes due to these unobservables. The dummy variable Tij measures availability

of the program to members who have self-selected, which is exogenous to the

household, but which may not be exogenous with respect to the village. The possible

correlation between Tij and the unmeasured household and village characteristics that

is due to self-selection at the household level is eliminated because unobservable

56

household characteristics are captured by Mij. Furthermore, because the control

villages are also program villages (which have not yet started making loans), there is

no longer any concern with nonrandom program placement. Yet, the order in which

villages receive village banks may not be random. The inclusion of nonmembers in

all villages, however, allows for the use of village-level fixed effects estimation to

control for the possibility that the order in which the 14 villages receive program

support is endogenous. Coleman compares the results from this “correct” empirical

specification to three other specifications that fail to correct for the biases resulting

from self-selection and subsequently demonstrates that the more “naïve”

specifications significantly overestimate program impact.

Similar to Coleman (1999), Aportela (1999) relies on a natural experiment

based upon the planned expansion of a Mexican savings institute targeted to low-

income people to examine the effects of increasing financial access on the saving rate

of households. The natural experimental nature of the study helps to solve the

problem of savings heterogeneity. Households’ preferences for savings would be

expected to be correlated with the use of formal financial instruments and, in general,

with access to these kinds of instruments. However, the natural expansion of the

savings institute allows Aportela to control for this possibility. The expansion, which

was occurred in 1993, was carried out only in select cities in Mexico. Therefore, it is

possible to use the savings rates of households in non-expansion towns as a control.

Aportela uses the 1989 and 1992 Mexican National Household Income and

Expenditure surveys to perform a reverse experiment in order to test the validity of

the expansion as a natural experiment. He finds that the expansion is not related to

57

households’ savings and preferences. He also finds no correlation between the

opening of an office and the income level of the state. Moreover, none of the

documents describing the bank’s expansion (official or unofficial) reference any

specific selection method for the location of the new branches. Location was not

chosen based on the bank’s previous presence or on convenience, as the bank chose

to expand in several states in which its presence was previously very limited.

Consequently, it seems reasonable to consider the 1993 expansion as exogenous and a

valid natural experiment.

Therefore, Aportela is able to use difference-in-difference estimation to

compare household data before the expansion with data after the expansion for both

the treatment and control group, and is able to perform simple OLS regression

analysis to determine the characteristics that affect savings rates. Aportela also

compares the OLS estimates using two other estimation techniques. The first, a robust

regression method, deals with the presence of gross outliers in the data. The method

estimates an OLS regression and then performs Cook’s outliers test. After eliminating

gross outliers, the regression is preformed again and weights are calculated based on

the absolute residuals. The regression is then performed again using those weights.

The process continues until the change in weights drops below a specific tolerance

level. The second method, the median regression, attempts to deal with the non-

normality of the savings distribution. This method describes the behavior at the center

of the population distribution, thus avoiding the sensitivity to extreme values.

However, in general, the OLS results were considered stronger than the results from

either the robust regression or the median regression.

58

Similar to Aportela (1999), MkNelly and Dunford (1999), in their study of the

Credit with Education Program in Ghana and Bolivia, apply a quasi-experimental

design at the community level in order to minimize possible bias. Following baseline

data collection, study communities were assigned to either a program or control

group. The control communities did not receive Credit with Education services until

the evaluation research was completed. Baseline respondents in the treatment group

were later classified as “future participants” or “nonparticipants” depending on

whether or not they joined the program (when and if it was offered in their

community). Baseline and follow-up information was collected over a three- to four-

year period with annual visits to assess the quality of the implementation and to

conduct qualitative research. Like Aportela (1999), the design of the survey allows

MkNelly and Dunford to evaluate program impact by simply comparing the

magnitude and direction of change in the responses across the two rounds between

program participants and nonparticipants, and residents in control communities.

Dupas and Robinson (2009) performed a quasi-experiment in Kenya to test

whether savings constraints prevented the self-employed from increasing the size of

their business. Trained enumerators identified a group of market vendors, bicycle taxi

drivers, hawkers, barbers, and other artisans. Those who already had a savings

account or who were uninterested in opening a savings account were excluded from

the sample. From the remaining group, participants were randomly selected to receive

an interest-free savings account. This survey design again allows for the use of simple

OLS analysis. To further control for any pre-treatment differences between the

treatment and control groups, however, Dupas and Robinson control for gender, years

59

of education, marital status, occupation, age, literacy and ROSCA contributions in the

last year.

All of the previously mentioned quasi-experimental studies, including

Coleman (1999), Aportela (1999), MkNelly and Dunford (1999), and Dupas and

Robinson (2009), were able to use relatively simple estimation techniques and are

still considered highly robust and reliable studies. However, these studies are not

easily replicable, as they rely on an exogenous expansion of a credit or savings

program. Furthermore, they still require large amounts of data collection, which

demands immense resources and time.

In order to deal with budget and time constraints, while still producing a

reliable and robust product, impact assessment studies are increasingly moving away

from relying on any single method, instead choosing a mix of survey and qualitative

techniques. In 1995, the United States Agency for International Development

(USAID) developed a method titled “Assessing the Impacts of Microenterprise

Services” (AIMS). It is comprised of five tools (2 quantitative and 3 qualitative) and

is designed to provide practitioners a low-cost way to measure impact and improve

institutional performance. AIMS core impact assessments were carried out in India,

Zimbabwe and Peru. Although these studies are often considered less econometrically

robust than some of their counterparts, such as Zaman (2000) or Khandker (2005),

they still use impressively large data sets and advanced quantitative data analysis and

then crosscheck these results with qualitative sources.

Chen and Snodgrass (2001), in their study of SEWA Bank in India, use a

sample of 900 households, including 600 clients (300 borrowers and 300 savers) and

60

300 non-clients. Two years later, they were able to re-interview 798 women,

including 276 borrowers, 260 savers, and 262 controls. Chen and Snodgrass follow

the core AIMS data analysis plan, which, in addition to simple descriptive statistics,

calls for two types of statistical analysis: gain score analysis and ANCOVA. Gain

score analysis compares amounts of change over time in an impact variable between

treatment and control groups. Analysis of covariance (ANCOVA) controls for the

possible influence of various personal characteristics on the impact variables. It

statistically “matches” observations in the treatment and control groups that have the

same baseline measures on the impact variables and on several moderating variables.

It then looks for systematic differences in second-round outcomes. It allows the

researcher to see whether borrower, saver, or client status in the bank is a statistically

significant determinant of round 2 values for the impact variables. To minimize the

risk of selection bias, Chen and Snodgrass include a broad list of control variables,

including age, marital status, educational attainment, religion/caste, employment

status, trade, household size and number of economically active household members.

In addition to using these control variables, the authors are careful in reporting the

analytical results so as not to overstate their findings. Furthermore, to strengthen their

conclusions, they supplement their quantitative findings with case studies.

In her study of the impact of Zambuko Trust in Zimbabwe, Barnes (2001)

uses statistical methods similar to those of Chen and Snodgrass (2001). However, she

uses a slightly different methodology to select the control group. To account for

potential biases, a control group of non-clients was selected according to the basic

criteria used by Zambuko loan officers: have an enterprise that is at least six months

61

old, be the sole or joint owner of that enterprise, and not be employed fulltime

elsewhere. Loan officers from Zambuko Trust also assess the viability of the

enterprise, but since that requires more time, the criterion was not used. The non-

clients were then matched with clients on the bases of gender and enterprise sector.

Like the study done in India, the survey was conducted in 1997 and then repeated in

1999 with the same respondents.

In addition to gain score analysis and ANCOVA, Barnes employs chi-square

tests and t-tests. Simple chi-square tests are used for two types of analysis: to

determine whether the comparison groups differ significantly on the direction of

change between 1997 and 1999, and to analyze the distribution of each sample group

on particular variables. Independent sample t-tests are used to determine whether

there is significant difference between the treatment and the control group from 1997

to 1999. The quantitative analysis is further supplemented by information from case

studies of nine clients.

Dunn and Arbunkle (2001), in their study of the impact of Acción

Communitaria del Perú (ACP), employ a methodology very similar to Barnes (2001).

The data set is compiled from two rounds of interviews of 701 entrepreneurial

households in Lima, of which 400 were ACP clients and 301 were non-clients. In

order to select the sample of households, a two-stage sampling approach was used. In

the first stage, two regions within Lima were selected as most representative of

ACP’s operations and the overall ACP client base. The second stage consisted of

random selection of the client and non-client households. Like the other AIMS

studies, Dunn and Arbunkle use gain score analysis and ANCOVA, and attempt to

62

control for selection bias by including an extensive list of control variables. The

quantitative analysis findings are further supported using case study analysis.

Like the AIMS studies, Hashemi, Schuler and Riley (1996) rely on a

combination of quantitative and qualitative measurements. In an effort to measure the

impact of the Grameen Bank and BRAC on women’s empowerment, Hashemi et al.

attempt to control for selection bias by statistically controlling for differences in

demographic characteristics such as age, education and wealth. They also attempt to

control for selection bias by including nonparticipants as well as participants in

Grameen bank villages, and by comparing them with women residing in villages

where the program is not operating. The study begins using logistic regression models

to explore whether Grameen Bank and BRAC affect different dimensions of

empowerment. Economic case studies and qualitative analysis are then used to further

explore the question of how credit empowers women, starting with the impact on

women’s economic role and proceeding to discuss other aspects of women’s lives,

such as physical mobility, interaction in the public sphere, and domination and

violence within the household. Although Hashemi et al. do not employ the most

scientifically and statistically robust methodology, by employing a greater variety of

methods, there are able to crosscheck their results.

Unfortunately, many impact assessments have made little effort to control for

any of the inherent biases. For example, in their analysis of the Small Entrepreneur

Loan Facility (SELF) in Tanzania, Ssendi and Anderson (2009) compare income

levels of SELF participants to a control group randomly drawn from the population.

Randomly drawing a control group from the larger population does little to control

63

for selection biases. The authors also employ fairly basic statistical tests, simply

comparing statistics, including incomes, asset indexes and enterprise gross margins,

between the two groups using a t-test and analysis of variance (ANOVA). From these

tests they find no significant difference between those who received a loan and those

who did not. Although the authors make no attempt to control for selection bias, they

are also very cautious when referring to the robustness of the conclusions that they

draw. The study could have gained much from the addition of qualitative analysis to

support their weak quantitative findings.

Allen and Hobane (2004), Anyango (2005), and Anyango et al. (2006) do not

use any form of control group. They simply explore the characteristics of a group of

program participants. However, the studies do use a variety of methodological

techniques, including focus group discussions, semi-structured interviews and case

studies, to strengthen their conclusions.

II. Study Design

This study attempts to find a middle ground between the subjective and

completely qualitative studies, and the econometrically rigorous and

methodologically sound, but expensive and time-consuming quantitative studies. As

discussed above, each approach offers certain advantages and disadvantages. A

sample survey and the attendant statistical approaches offer representativeness,

quantification, and attribution, while the humanities approach grants the reader the

ability to uncover processes and to capture the diversity of perceptions, views of

minorities, unexpected impacts, etc.. This study seeks to combine these unique

64

advantages by using a composite of the two methods. The quantitative aspect of the

study focuses on a small-scale sample survey of current VSLA members and a

control group of incoming members. The results from the survey are then

crosschecked using information gathered from focus group discussions and

interviews with key informants.

Impacts are assessed at both the individual and the household level, primarily

using the data gathered from the questionnaire. As it is very difficult to measure

community-level impacts in a quantitative survey without the use of a control group

of eligible non-members, impacts at the community level are addressed briefly, based

primarily on information gathered through focus group discussions. The community-

level assessment is predominantly focused on capturing any major externalities of the

program intervention.

This study focuses on economic and social impacts and, in conclusion, briefly

examines the sustainability of the VSLA model. Economic impact is measured

principally through expenditure levels, the accumulation of household assets and the

development of income-generating activities (IGAs). To estimate social impact, the

study relies on a variety of indicators, including educational expenses, access to

health services, nutritional levels, and quality of housing. The use of anthropometric

measures would be ideal, but these are more difficult and time-consuming to collect.

The study also briefly addresses empowerment through questions relating to

involvement in household decision-making, levels of participation in community

activities, and electoral participation. The responses to these questions are further

supplemented through the focus group discussions.

65

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 marginalized or vulnerable may be

overlooked. The very poor also may be excluded due to their inability to finance the

purchase of shares. In order to address this problem, this study follows the procedure

implemented under the AIMS methodology and utilizes a control group of new

VSLA members who are still in the initial training phase and have, therefore, not yet

felt any impact from participation in the program. To further address self-selection

bias, the study also statistically controls for differences in demographic characteristics

including age, gender, religion, marital status and education, which may affect

program impact.

Comparing program veterans to new participants theoretically eliminates the

bias that occurs when the treatment group systematically possesses unobserved

attributes, which the control group lacks. As all study participants have joined the

program at one point or another, it is assumed that they share the same, or at least

similar, entrepreneurial drive or preferences. 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 also makes two major assumptions,

which may not be true: first, the approach assumes that either no one drops out or that

66

dropping out of the program occurs randomly; second, the approach assumes that

there is no change in how selection of VSLA members occurs over time. These

assumptions, if false, could cause two major problems: incomplete sample bias and

attrition bias (Karlan 2001). Incomplete sample bias refers to the fact that those who

drop out were presumably impacted differently and potentially made worse off than

those who remained. By ignoring dropouts in the sample there is a possibility of over-

or underestimation, depending on whether the cause of dropping out was success or

failure. Attrition bias suggests that those who drop out are different from those who

remain, irrespective of the program impact. For example, if poorer individuals are

more likely to dropout – for example, because their lack of access to the minimal

capital required to participate meaningfully in the VSLA creates a barrier to enjoying

the benefits of participation - this will cause an upward bias on the estimates of

program impact. However, if richer individuals are more likely to drop out – perhaps

because they do not perceive the benefits of participation will equal or exceed other

opportunities available to them - there will be a downward bias on the estimates of

program impact.

Bias may also be introduced by changing selection effects over time. If the

first to join the program are wealthier, or more entrepreneurial, and are perhaps

considered by their peers to be more reliable and trustworthy, program 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, program impact may be underestimated if the poor are the

67

first to join, if, for example, they are willing to take greater risks than their wealthier,

more conservative neighbors.

In addition to the two major assumptions discussed above, 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 vs. veteran participant

pool. The credit or savings program may change its strategy and/or client

identification process. Program placement also may change – for example, the

programs 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 neighborhoods. Program 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.

This study attempts to tackle each of these potential issues. The dropout biases

are problematic but solvable. In order to address the issue, 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 program. The dropout selection problem is addressed

by controlling for client characteristics, such as age, educational attainment and

number of children, at the time of joining the VSLA group. Comparisons are also

made between time invariant characteristics of the treatment and control groups to

look for evidence of changing selection.

The problems of institutional dynamics, however, are slightly more difficult to

address. 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

68

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 ten years. In the past, CARE, and now

JOCDO, approaches the leadership of every village in the area to explain the

program. The village leader is then responsible for informing his community of the

opportunity. If there is a group of 15-30 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, as all villages in

the area are informed of the program, there is little reason to believe that the nature of

the communities involved in the program has changed over time. Though this

evidence is unavoidably anecdotal, it suggests that changes in the selection process or

institutional dynamics will not bias the results of this study.

i. Sampling Strategy

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 organization 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

program participation. The control group is made up of five new VSLA groups that

began training in early January. The survey took place late in the same month;

69

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.

ii. The Individual Survey

From the sample of 73 groups, 25 groups spread across 13 different villages

were randomly chosen. Four members (with two alternates) were then randomly

selected 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 randomly chosen and therefore, the average length of membership is

only five years. In addition to the four current members from each group, twenty

dropouts were randomly selected from the overall group, based on an approximate

attrition rate of 20 percent, to be interviewed in order to control for potential dropout

biases.

The questionnaire tool in Appendix B, which was translated into Kiswahili,

covered the basic socioeconomic characteristics of the respondents and their

households: participation in the VSLA program, 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). In order to further ensure compatibility with both the literature and the local

environment, the format of the questionnaire was also largely based on that used in

CARE’s Strategic Impact Inquiry (SII), which took place in Tanzania in 2006.

The questionnaires were administered by twenty carefully selected

Community Contact Persons (CCPs), who, because of their prior work training and

supporting the VSLA members, knew and were known to the survey participants.

70

Prior to the data collection, the author conducted a one-day training session at

JOCDO’s headquarters in Stonetown, Zanzibar. The training familiarized the CCPs

with the questionnaire and provided an opportunity to refine the survey. The

questionnaire was tested to ensure that the questions and sentence structure were not

too complex or technical for respondents and interviewers alike.

iii. Focus Group Discussions

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 randomly selected from the original group 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 survey. The format

of the focus group discussions, which may be found in Appendix C, included twelve

open-ended questions that were intended to generate an open discussion. The

questions covered issues such as group formation and membership, general group

dynamics, challenges and limitations, behavioral changes, social and economic

impact, benefits and/or negative consequences of participation, impact on the

community, and the sustainability and effectiveness of the apex organization. In

addition, the author visited each of the three groups included in the focus group

discussions during its weekly VSLA meeting, in order to observe the methodology

and activities of each group as well as general group dynamics. Though the author is

conversational in Kiswahili and ran the focus group discussions, understanding of the

information obtained during the discussions was enhanced by the assistance of a well-

qualified interpreter, who was present for all sessions.

71

iv. Interviews with Key Informants

Key informant interviews were also arranged with several CCPs and several

members of JOCDO’s executive committee. The interviews were largely

unstructured. However, the general format covered, among other things, the nature of

services supplied to current groups, the formation of new groups and institutional

dynamics. Again, although the author personally conducted each interview, an

interpreter was present to facilitate more open conversation and a more complete

understanding.

III. Quantitative Data Analysis

i. Model Specification

Assuming that the treatment and the control group are identical in terms of all

observed and unobserved characteristics, simple comparisons of the means across the

treatment and the control group, which are presented in the next section, allow for

initial estimations of program impact. Regression analysis is then used to further

explore certain hypothesized outcome variables that “pass” this test. Regression

analysis permits the reviewer to draw more robust conclusions by measuring the

impact of program membership on a specific set of outcome variables, 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,

yi = ! + "1membership + "2gender + "3age + "4religion + "5maritalstat +

"6educatt + "7children + "8priorsav + "9prioraccess + ui

72

where "1 is the parameter of interest as it measures the impact of the VSLA on the

outcome variable, and yi is an outcome variable. Membership is a binary variable,

representing whether or not the respondent is in the treatment or control group.

Gender is also a binary variable, equal to 1 for female. Age is a continuous variable,

which corresponds to the current age of respondents. Religion is a non-continuous

variable, and thus is broken down into two variables: Christian and Other – the

Muslim population is represented by a zero in both categories. Maritalstat is also a

non-continuous variable and is subsequently broken down into four binary variables:

Married, Widowed, Divorced, and Separated, while a zero in all categories

corresponds to single. Educatt is broken down into three binary variables: Primary,

Ordinary level, and Secondary level, while those with no education are represented by

a zero in each category. Children is a continuous variable representing a respondent’s

current number of children. Priorsav is a binary variable equal to 1 if the respondent

saved prior to joining the VSLA. Prioraccess is also a binary variable, which is equal

to 1 if the respondent had access to loan services prior to joining the program.

It is assumed that membership in the VSLA is uncorrelated with the omitted

unobservable variables that are contained in the error term, u. There is a possibility

that this assumption may be violated in reality, introducing heteroskedasticity into the

model. Ideally, one of the three more advanced econometric techniques discussed

previously in the chapter would be used to address the selection bias that might arise

as a result of such unobservable variables. However, these methods require a much

larger data set than is available for this study. Therefore, this study must rely on

robust standard errors in order to address this possibility of heteroskedasticity. Note

73

that the standard errors are further adjusted through the use of clusters at either the

VSLA group and village level.

The outcome variables to be analyzed include number of income-generating

activities (IGAs), asset expenditures, health expenditures, education expenditures,

number of meals per day, number of times the household had fish or meat in the last

seven days, and the likelihood of using mosquito nets, of owning one’s home and of

making improvements in the quality of one’s housing. These variables were chosen

on the basis of initial mean comparisons and previous findings in the literature,

described in Section IV.

It is reasonable to hypothesize that the effect of VSLA program participation

on the set of outcome variables will be different for male and female members.

Several studies have found a differential impact of microfinance programs across

gender, with services targeted to women as an overwhelmingly more effective

method of ensuring that the program benefits reach the entire family (Pitt and

Khandker 1998, 2003; Khandker 2005; Strauss and Beegle 1996; Hoddinott and

Haddad 1994). The literature suggests that women are more likely than men to invest

in the welfare of the household, however, only if they have access to the necessary

resources. In theory, membership in the VSLA program improves women’s access to

such required resources. Therefore, we may find a greater positive impact of program

participation on the welfare of households of female members than on those of male

members. In order to account for this possibility, an interaction term between

program participation and gender is included in the basic model.

74

In order to establish whether program participation has a discrete, all-or-

nothing sort of effect or whether benefits are linked to the amount of time exposed to

the program, an additional specification is included in each table, which incorporates

dosage - a continuous variable representing the number of years in the groups.

Membership continues to capture the discrete effect of program participation, while

dosage is used to capture the repeated effect.

Several of the expected outcome variables are recorded in a binary format.

Although these variables are analyzed using two different methods: the Linear

Probability Model (LPM) and the discrete probit model, the basic model remains the

same. Under the LPM, "1, the coefficient on membership, can be interpreted as the

change in the probability of achieving success of the dependent variable (i.e. a value

of 1) if the respondent is a member of a VSLA. Although the interpretation of the

coefficients under the LPM is very convenient, there are several potential problems,

including heteroskedasticity and non-normality of the distribution of errors. A

violation of the linearity assumption may produce estimates that indicate a negative

probability or a probability greater than 100 percent – neither of which is a statistical

possibility. Even if the assumption of linearity is violated, the estimates produced

under the LPM will tend to have the correct sign for the effect of the independent

variables on the binary outcome variable. However, the estimates may grossly

understate (or overstate) the magnitude of the true effects and may also be highly

sensitive to the range of data observed in the sample. Furthermore, if the

distributional properties do not hold, the standard errors of the estimates may be

vastly inaccurate.

75

In order to address these potential issues, the probit model is presented, in

addition to the LPM. Probit limits the probability of achieving success of the

dependent variable to follow the standard normal distribution and, therefore, the

predicted probabilities never go lower that 0 or above 1. Unfortunately, however, the

magnitude of the coefficients from the probit model is much more difficult to

interpret. To facilitate in-depth analysis, marginal effects at the means of the

parameters of interest are included in each of the regression tables. Similar to the

results from the LPM, the marginal effects are evaluated as the change in the percent

likelihood of realizing success of the dependent variable, i.e. a value of one. The

results from the regression analyses are presented in Section VII below.

III. Data Description

i. Basic Characteristics of Respondents

Table 1 presents the means of the key characteristics of both the treatment

group and the control group. Many of these characteristics are considered to be time-

invariant and should not be affected by participation in the VSLA program. If the

control group is to be considered valid, the basic characteristics of its members should

not vary significantly from those of the treatment group. For variables that are

continuous, t-tests are used to assess whether the means of the two groups are

statistically different from each other. For the variables that are categorical, the values

are broken down and percentages then compared across the treatment and control

group using a proportion test. The results of the t-tests, as well as the values from the

proportion test, are presented in the final column. The two values may be interpreted

in the same manner – if the value is greater than 1.64, then the difference between the

76

treatment and control group may be considered statistically significant at the 10

percent level.

From Table 1, we can see that the control group has slightly more female

respondents than the treatment group. However, the difference is not statistically

significant. The treatment group is significantly older than the control group.

However, there does not appear to be a significant difference between the age at

which members of each group joined the VSLA program.

The respondents in both the treatment and control group are comparable

across marital status, religion and relation to the household head (HHH). There is a

marginally significant difference between the proportions of the two groups that are

the offspring of the head of the household, with the control group having almost twice

the proportion of respondents still living with their parents. Although this may be an

indication of the treatment group’s financial ability to move out of their parent’s

home, more likely it is simply a reflection of the younger age of the control group.

There appears to be a significant difference between the educational

attainment levels of the two groups, as the data suggest that the control group is better

educated than the treatment group. While a significantly greater proportion of the

treatment group has completed primary school, a significantly greater proportion of

the control group has completed the ordinary level of secondary school. Given the

national improvement in school enrollment levels in the past ten years (Tanzania

National Website), it is not surprising that the younger population in the control group

has achieved higher educational attainment. If higher education leads to better

outcomes among the control group, this would bias the estimates of program impact

77

downwards. Therefore, we need not be concerned with this disparity. Nevertheless,

educational attainment will be included as a control variable in the regressions.

The number of children of the respondent is significantly different between

the treatment and the control group. However, this is likely a reflection of the

difference in age between the two groups. Members of the treatment group are

appreciably older than the control group and therefore, are likely to have more

children currently. Furthermore, the difference between the number of children at

time of joining is not significant.

Given the difference in the number of children, it is interesting to note the

results for household size, which indicate that there is no statistically significant

difference between the two groups. There are two possible explanations for this

apparent discrepancy. The children of the treatment group are likely older and many

may have already moved out of the home. On the other hand, a greater proportion of

the control group is still living with their parents. Therefore, although they may not

have any children themselves, they are more likely to be living with siblings, as well

as their parents.

A slightly greater proportion of the treatment group saved in some form prior

to joining the VSLA program. The literature suggests that those who save prior to

joining any kind of development initiative, such as a VSLA, may be innately more

entrepreneurial or ambitious and thus, more likely to join the program and succeed in

the program. The difference between the treatment and control group is not

statistically significant. Nevertheless, it may be considered economically important.

We will have to bear this in mind when interpreting the results of the study.

78

Similarly, respondents who had access to loans prior to joining the program

may be assumed to be slightly better off, as access to loans, particularly those from

formal institutions, is generally severely limited by poverty. Alternatively, the loans

may have been made available by an organization which targets the poor, in which

case, those with prior access may have been worse off. A marginally greater

percentage of the treatment group had access to loans prior to joining the program.

However, again the difference does not appear to be statistically significant and thus

we need not be concerned with potential biases in either direction.

Overall, the treatment and control groups appear to be similar along most

dimensions. The only statistically significant differences are in educational

attainment, and these favor the control group. Therefore, there seems to be little risk

of a positive, or upward, bias on the estimates on VSLA program impact.

a. An Additional Test

In order to confirm this preliminary assessment, the treatment group is sub-

divided by the median number of years of program participation. This additional

specification allows us to further control for selection bias over time - specifically, the

possibility that the first people to join the VSLA program were better off or more

entrepreneurial than those who joined later, including the more recent members in the

treatment group. Table 2 presents the means and percentages for both the older and

more recent participants of the treatment group, as well as for the new members, i.e.

the control group. Again, both t-tests and proportion tests are used to assess whether

the three groups are statistically different from each other.

79

Similar to the results from Table 1, it appears that the three groups are

generally similar across basic characteristics. However, there is a difference in

educational attainment across the three groups. Although the difference is not quite

statistically significant at the 10 percent level, it appears that a greater proportion of

the new participants in the control group have no education, compared to the more

recent additions to the treatment group. However, the proportions with no education

are similar between the new members and the older members in the treatment group.

A smaller proportion of new (control group) members have a primary education,

compared to both more recent and older members of the treatment group. However, a

greater percentage of control group members have some secondary education

(ordinary level) compared to recent and older members of the treatment group. There

is no statistical difference across the groups for completion of the advanced level of

secondary school. Overall, this seems to corroborate the previous finding, which

suggests that the new members in the control group are somewhat better educated

than either the recent or older members of the treatment group, probably owing to the

nationwide improvement in school enrollment levels. A greater percentage of both the

more recent and older members of the treatment group have completed only primary

school, while a greater percentage of the new members have some secondary

education.

The results imply that the more recent members of the treatment group are

slightly more educated than the older members. A greater proportion of the older

members have no education, while a smaller proportion have a primary education.

Although, a slightly greater percentage of the older members have completed either

80

the ordinary level or advanced level of secondary school, the difference is not

statistically significant. Again, these findings run parallel to our previous theory about

the correlation between educational attainment and age. The newer members (both

the more recent members of the treatment group as well as the new members in the

control group) are, in general, younger than the older members and, as such, more

likely to have benefited from the overall national increase in enrollment.

Number of children of the respondent is significantly different between the

older members and the newer and more recent members. However, again the

significance largely disappears when using number of children at time of joining. The

difference between older members and the new members, however, remains

significant at the 10 percent level.

There are some apparent differences in prior savings and access to loans

between the three groups. A greater proportion of the more recent members of the

treatment group had savings prior to joining a VSLA. Although the difference is not

statistically significant between either of the other two groups, there is a possibility

that the more recent members are naturally more driven to save and therefore may

fare better in the VSLA program. Additionally, there may be reason to believe the

more recent members of the treatment group are better off than either the older

members or the new (control group) members, as a greater percentage of recent

members had access to loans prior to joining the program. The difference is not

statistically significant between recent and new members, but the difference between

the recent and older members of the treatment group is significant at almost the 5

percent level.

81

On the whole, the results in Table 2 confirm those found in Table 1,

reinforcing the validity of the control group. 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 confirming that the characteristics of VSLA program

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 toward finding the program to be less effective

than it truly is.

ii. Socio-Economic Status of Respondents

Given the difficulties of directly measuring poverty or income-levels, the

individual questionnaire included a variety of alternative indicators of welfare,

including physical housing characteristics, the availability of safe drinking water,

electricity, sanitation, asset ownership, access to education, household food security,

and access to health services. Now that we have evaluated the validity of the control

group, in order to begin to analyze VSLA program impact, we now look for

differences between these welfare indicators for the two groups.

a. Quality of Housing

Physical housing characteristics are a useful indicator of the socio-economic

status of the household. Table 3 presents a comparison of the basic housing

characteristics between the treatment and control group. Note that if we truly believe

82

the treatment and the control group are identical in terms of all observed and

unobserved characteristics, as suggested in the previous section, then a statistically

significant difference between the two groups suggests causation. However, a more

detailed and rigorous analysis of the impact of VSLA participation will be presented

in the next section.

In Table 3, there appears to be no significant difference between the two

groups for several categories. Access to safe drinking water appears to be similar

between the treatment and control group, with 74 and 76 percent of each,

respectively, having piped water supply. The primary source of cooking fuel for 98

percent of both the treatment group and the control group is fuel wood, with the

remaining 2 percent using charcoal. The percentages of each group that use soil,

cement or tiles as their primary flooring material are also statistically similar. Finally,

the average number of rooms for sleeping is almost identical between the two groups

– the mean for each group is approximately 2.56.

However, it appears there are substantial differences between the two groups

along several dimensions of housing characteristics. To begin with, the results

indicate a statistically significant differential in home ownership. A much greater

percentage of the treatment group (85.8 percent) own their home, compared to the

control group (only 60 percent), while 34 percent of the control group share their

home with other families, compared to only 8.3 percent of the treatment group. There

does not appear to be a significant difference between the percentages of either group

that rent. Also, a significantly greater proportion of the households in the treatment

83

group also have electricity (28.3 percent), compared to the control group (only 18

percent).

The treatment group also appears to be slightly better off in terms of sanitation

facilities. The World Bank defines sanitation in sub-Saharan Africa as a ladder, in

which each successive rung represents a higher cost but a correspondingly lower level

of health risk. Traditional pit latrines, which refer to various kinds of pits for the

disposal of excreta, represent the first rung. Improved latrines, which represent the

second rung of the ladder, include SanPlat, VIP latrines and basic pits with slabs, all

of which ensure more hygienic separation of excreta form the immediate

environment. The final rung of the sanitation ladder is the flush toilet, which may be

connected to either a septic tank or to a water-borne sewer network (Morella et al.

2008). A marginally greater percentage of households in the treatment group have no

sanitation facilities (13 percent, versus 4 percent for the control group), however, 78

percent of treatment group households have an improved latrine and 5 percent have a

flush toilet, compared to only 22 percent and 2 percent, respectively, of households in

the control group. The majority of households in the control group (54 percent) have

only a traditional pit latrine. Therefore, overall, it appears that the treatment group has

significantly better access to improved sanitation facilities.

Members of the treatment group also seem to, overall, use higher quality wall

material than those of the control group. A similar percentage of households in the

two groups use grass. A greater percentage of households in the control group use

either mud and pole, sun-dried bricks, baked bricks, or cement bricks, while the

majority of households in the treatment group use stones. While stone is considered

84

higher quality than many of the other options, cement bricks are generally considered

to be the optimum building material. The results indicate that while half of the control

group uses lesser quality building materials, the other half uses the highest quality.

Alternatively, few members of the treatment group use the lowest quality building

materials, but a smaller proportion uses the very best.

In order to better distinguish which group is better off in terms of wall quality,

the various materials are condensed into two categories: high and low quality. Eighty-

two percent of the treatment group uses high quality walling material (cement bricks

or stones), compared to only 54 percent of the control group – a difference which is

statistically significant at the 1 percent level. Therefore, overall, the treatment group

appears to be better off in terms of quality of wall construction material used.

The data for roofing material is much more straightforward. Forty-eight

percent of households in the control group use thatch, compared to only 22 percent of

households in the treatment group. The substantial majority of households in the

treatment group use corrugated iron (76 percent), compared with only 52 percent of

households in the control group. As corrugated iron is a much higher quality and

more durable material than thatch, the treatment group may be regarded as

significantly better-off than the control group.

While the findings in several categories discussed above are not entirely clear-

cut, the final row of Table 3 suggests that households in the treatment group are much

more likely to have made improvements in the quality of their housing in the past 12

months. Participation in the VSLA program gives households access to lump sums of

money that may facilitate investments in housing quality. Therefore, it is not

85

surprising that 67 percent of households in the treatment group have made housing

improvements in the past year, compared to only 16 percent of households in the

control group – a differential which is highly statistically significant at the 1 percent

level. This investment differential may help to explain some of the differences in

housing quality discussed previously.

Overall, the data suggest that members of the treatment group have higher

quality housing than those in the control group. This implies that participation in the

VSLA program facilitates investments in the quality of housing. This result confirms

the findings of Anyango et al. (2006), who find that VSLA members are more likely

to own and live in better constructed homes than the general population.

b. Household Assets

In developing countries like Tanzania, in particular, it is very time-consuming

and expensive to obtain reliable measures of household income. The majority of

households rely on small-scale agriculture for everyday consumption, and therefore,

have little to no measurable income. Measuring household consumption also requires

expensive and time-consuming monitoring techniques. Therefore, given the time and

financial limitations of this study, a household’s yearly asset expenditure is used as an

alternative to either income or consumption – an option frequently adopted in the

literature (Ssendi and Anderson 2009; Pitt and Khandker 1998). In the individual

questionnaire, respondents were asked to recall the amount they spent on household

assets, including durable household items, equipment, and means of transport, in

2009. Note that this list does not include investments in housing as such expenditures

are attended to later in the survey. Although the use of the recall technique is not

86

ideal, given the small quantity of disposable income and the correspondingly stringent

budget of the majority of respondents, recollections of expenditures in the previous

year are generally considered to be reliable (Ssendi and Anderson 2009). This also

holds true for the estimates of a household’s expenditures on education and health,

which are discussed in later sections.

From Table 4, although there does not appear to be any substantial difference

in asset ownership between the treatment and the control group, with a few notable

exceptions, there is a statistically significant difference in asset expenditures. In terms

of asset ownership, members of the treatment group own, on average, a greater

number of cows than members of the control group. There is a sub-set of the savings

literature, which discusses the role of livestock as an important form of savings.

Livestock provide a steady source of nutrition and draught power and may be easily

sold when there is an urgent need for money within the household (Deshingkar et al.

2008; Udry 1995). Therefore, the difference in livestock ownership between the

treatment and the control group may be deemed quite important.

Members of the control group appear to own, on average, more hoes than

members of the treatment group. However, given the relatively minor expense of a

hoe compared to the other items in the table, this differential may not be very

important.

The final row of the table indicates a statistically significant difference in asset

expenditures. In 2009, households in the treatment group spent, on average,

approximately Tsh138,000 (US$125), while households in the control group spent

around Tsh31,000 (US$28). This substantial differential does not seem to be

87

consistently reflected in the data, given the similarities in asset levels between the two

groups. Therefore, the greater part of the asset expenditures of the households in the

treatment group may have been used to purchase items not specified in the list. The

list is based off the study done in 2006 by Anyango et al., and is in no way

exhaustive.

c. Education

Considering the importance of education for a child’s social and economic

prospects (as well as the prospects of his parents), access to educational services is

one of the most important indicators of a household’s well-being. Education in

Tanzania is compulsory and tuition-free for the first seven years. Government

secondary schools, however, cost approximately Tsh20,000 per year (US$15), in

addition to various other expenses, including uniforms, other materials, and testing

fees. Therefore, a household’s education expenditures are used as a proxy for

educational attainment. It may also serve as a proxy for the quality of education

received. The elimination of tuition in 2002 has led to a massive increase in the

number of children enrolled in primary school and correspondingly, a substantial

shortage of teachers and materials. This has created a large market for private

schools, which can cost any where from Tsh200,000 (US$150) to almost

Tsh22,000,000 (US$20,000), for those who wish to receive a higher quality

education.

From the last row in Table 4 it appears that households in the treatment group

spend significantly more on education than those in the control group. Last year,

households in the treatment group spent approximately Tsh105,500 (US$95), while

88

those in the control group only spent Tsh34,000 (US$31), a difference that is

statistically significant at the 10 percent level. This suggests that participation in the

VSLA program increases educational attainment and/or improves the quality of

education received.

d. Nutrition

Members of the treatment group seem to fare no better than those in the

control group in terms of meal quantity, but they are significantly better off in terms

of quality, as measured by the frequency with which they consume fish and meat. As

illustrated by Table 5, there is no significant difference in average number of meals

per day between the treatment and control group. However, there is a substantial and

statistically significant differential in fish and meat consumption. Households in the

treatment group consume meat, on average, once every two weeks, while households

in the control group consume meat only once every six weeks. Households in the

treatment group also consumed fish, on average, on 4.61 days of the past week, while

households in the control group only consumed fish on 1.20 days.

Households in the treatment group also appear to have had fewer problems

satisfying food needs in the past year. The difference between the two groups is

highly statistically significant at the 1 percent level. Thirty-three percent of the

treatment group ‘never’ had problems satisfying food needs, compared to only 6

percent of the control group. Correspondingly, only 66 percent of the treatment group

‘sometimes’ had problems satisfying food needs, compared to 88 percent of the

control group. There is not a significant different between the proportion of each

group that ‘often’ experienced problems satisfying their household food needs.

89

e. Health

From Table 6, it is evident that the results for access to health services are

mixed. In the past year, a greater percentage of the treatment group ‘never’

experienced problems accessing medical services (23 percent, versus 4 percent for the

control group), while a greater percentage of the control group ‘sometimes’

experienced problems (96 percent, versus 69 percent for the treatment group).

However, a greater proportion of the treatment group reported ‘often’ experiencing

problems in accessing medical services (8 percent, versus zero percent for the control

group).

There appears to be no significant difference in the immunization of children

between the treatment and the control group. However, a significantly greater

proportion of children in the treatment group sleep under a mosquito net compared to

children in the control group (97.4 percent, versus 90.9 percent). Given the high

prevalence of malaria in Tanzania, the use of a mosquito net is an important indicator

of the general health of a household.

Households in the treatment group spent, on average, significantly more on

healthcare in 2009, than households in the control group – almost twice as much. This

may be a reflection of the greater capacity of households in the treatment group to

finance better healthcare. Alternatively, this may suggest that households in the

treatment group have poorer health and, therefore, must spend more on healthcare

costs. However, given that access to healthcare in Tanzania is generally financially

constrained, the former explanation seems more plausible. On the whole,

90

participation in the VSLA program appears to facilitate increased access to health

care services.

f. Sources of Income

The four-phase training program in the VSLA methodology involves

substantial training in Income Generating Activities (IGAs). In Table 7, we can see a

measurable impact of this training, when combined with the other benefits of

participating in the VSLA, on the number of IGAs pursued by households in the

treatment group. Households in the treatment group are, on average, involved in 1.91

IGAs, compared to the 1.39 IGAs operated by households in the control group – a

difference that is statistically significant at the 1 percent level.

The types of IGAs, however, are similar across the two groups. The main

economic activities of both groups are agriculture and business. Those involved in

business are mainly engaged in the sale of food, khangas,2 fuel wood or charcoal. A

small percentage of both groups are also involved in tailoring, which is considered to

be a highly-skilled trade and is often incorporated into VSL training. A substantial

proportion of the households in the treatment group are also involved in seaweed

farming. The seaweed is primarily exported to East Asia and Europe as an additive

for many products, including processed meat, toothpaste and mascara (Anyango et al.

2006, 27). Seaweed farming is strongly encourage in the VSLA training, as it is

generally considered to be a very sustainable source of income because it requires

little initial investment, and no land, fertilizer or irrigation. However, this activity is

largely limited to the specific context of the study in Zanzibar and therefore, the

results may not be relevant to the VSLA program, in general. 2 A piece of cloth traditionally tied around the body or head.

91

g. Social Status

In general, the literature suggests that microfinance programs have a positive

social impact on participants, particularly on the empowerment and equality of women

(Pitt and Khandker 1998, 2003; Khandker 2005; Strauss and Beegle 1996; Hoddinott

and Haddad 1994). Social impact is very difficult to quantify in a survey format.

Nevertheless, this study attempts to quantitatively address the subject with a few select

questions taken from CARE’s 2006 Strategic Impact Inquiry (SII). These questions

and the results are listed in Table 8.

From the table, there appears to be little statistically significant difference

between the two groups for the included parameters with the exception of the last

question. In the past year, 30 percent of respondents in the treatment group expressed

an opinion in a public meeting, other than the weekly VSLA meeting, compared to

only 8 percent of the control group.

Although these findings, or lack thereof, are somewhat discouraging, it may

be a reflection of the difficulty of measuring social impact in a survey format. The

information gathered through the focus group discussions, as well as the questions that

were addressed only to members of the treatment group, presented in Table 11 and

discussed below, suggest that the majority of VSLA members believe that participation

in the program has led to an improvement in their status within their family and the

community.

iii. VSLA Members Self-Reported Impact

As discussed above, in a study of this magnitude, with the attendant time and

financial constraints, it is very difficult to collect reliable data on certain variables,

92

such as income, consumption, or anthropometric measurements. In order to fill, at least

partially, the gaps left by the omission of these variables, program participants are

asked for their perceptions of program impact. Evaluating any microfinance program

on the basis of self-reported impact is clearly not ideal. Clients may not be able to

differentiate whether an outcome was the result of participation in the program or some

other force. Alternatively, they may not respond honestly for fear of affecting their

status in the program. However, allowing VSLA members to offer their own

perceptions of the benefits or disadvantages of program participation may augment the

more rigorously collected data and may provide additional insights into the complex

network of program impact. Furthermore, it may have the added benefit of improving a

member’s sense of ownership over the program. VSLA members’ own interpretations

of program impact are presented in the following section. In order to better understand

current members’ responses, it is important to first gain a better understanding of the

dynamics of VSLA participation.

a. Dynamics of VSLA Participation

As discussed earlier, in order to control for potential biases, 20 respondents

who had dropped out of the VSLA program are included in the treatment group.

Those who drop out of the program may be inherently different or may be impacted

differently by program participation than those who remain. In order to test for this

possibility, when analyzing the dynamics of VSLA participation, the treatment group

is broken down into current members and dropouts.

Table 9 breaks down the details of VSLA membership so that the reader

might better understand the dynamics of program participation. The current members

93

have been in the VSLA program for an average of approximately five years – a

reasonably sufficient period to perceive some level of program impact. However,

there is a significant difference between current members and dropouts in terms of

length of membership in the VSLA. Dropouts were only members in the VSLA

program for around three years. This divergence may simply be a result of the

dropouts exiting from the program. Alternatively, given that most of the dropouts left

the program after a relatively shorter period of time, it may be an indication that the

program was negatively impacting them.

VSLA Members appear to use the payout of each savings cycle to fill a

variety of household needs, which may suggest a positive program impact on

consumption smoothing within the household. Food appears to be the most common

use of the payout, with a substantial portion also spending on family celebrations,

housing improvements, and productive investments. The payout gives the program

participant access to a lump sum of money that would otherwise be unavailable, thus

facilitating large investments in housing and business ventures. The timing of the

payout is also generally set to coincide with occasions that require such large sums of

money, such as at the start of the school year or before a major holiday. Therefore, it

is logical that school fees and family celebrations are also two of the major uses of the

payout. The amount of the last payout and the primary uses of the payout do not vary

significantly between the current members and dropouts, with one notable exception -

a substantially greater proportion of current members use the payout to pay for their

children’s education.

94

The average amount of the last payout was approximately Tsh270,000

(US$245). In a country where the average annual income is approximately

Tsh1,367,300 (US$1,243) this represents a very substantial amount – specifically, 20

percent of average annual income (Human Development Report 2009).3 Furthermore,

in most groups, members contribute up to three shares of Tsh1,000 (US$0.90) every

week, which yields a total share value of Tsh156,000 (US$142). Therefore, the

average payout represents up to a 58 percent return on members’ savings. This is

even greater than the finding in Anyango et al. (2006), who reported an impressive

rate of return on savings of 53 percent.

Note that these numbers represent the maximum possible rate of return on

savings, which can only be experienced by net savers. The return on savings is

financed primarily through interest paid on loans, in addition to the various fines

levied during each savings cycle. Therefore, it is possible that a net borrower may

actually pay more into the system than she takes out in the final savings payout,

which essentially represents a negative return to savings. However, it seems

reasonable to suppose that the borrower receives some additional benefit from access

to the loan. Overall, as there is no external source of funds entering the system, the

average return is actually 0 percent. In fact, to the extent that there are also various

expenses for each group, such as those paid to JOCDO, returns may actually be

negative on average. On the whole, it is important to understand that, while for net

savers the VSLA program may represent a promising investment opportunity, in

3 The average income of VSLA participants may be a more relative number, as they may be wealthier, on average, than the general population. However, as discussed above, this number is very difficult to collect considering the time and financial constraints of this study.

95

general, the potential benefits of VSLA participation are derived from access to

savings and loan services.

Current members have taken an average of 6.5 loans from the VSLA. In a

country where the vast majority of the population lacks access to any loan services,

this represents a possible substantial benefit of program participation. However,

again, there is a sizeable difference between current members and dropouts in the

number of loans taken from the VSLA program. While current members have taken

an average of 6.5 loans, dropouts have, on average, only taken 3.4 loans from the

VSLA program. This differential, however, is likely only a reflection of the

difference in the length of membership in the program.

The average loan given in the VSLA program is approximately Tsh118,000

(US$107), and does not significantly differ across the two groups. Again, given the

average national income, this represents a very substantial amount. Members appear

to use the loans for a wide variety of purposes. Again, purchasing food appears to be

the most common use of the loan for both groups, with a substantial percentage of

each group also spending on school fees, family celebrations, housing improvements

and medical expenses. The primary uses of the loans also do not vary significantly

between the current members and the dropouts, with two notable exceptions. While

none of the dropouts listed payment of debts as a primary use of the loan, eighteen

percent of current members identified debt repayment as a primary use. A greater

proportion of current members (54 percent) also invested in productive activities,

compared to only 30 percent of dropouts.

96

While it is recognized that it will sometimes be necessary to use a loan for the

purposes of consumption smoothing, life-cycle events, emergencies, housing, or

education expenses, VSLA training strongly encourages the use of loans for

investment in productive pursuits, as it facilitates repayment. While both current

members and dropouts use the VSLA loans principally to support personal

consumption (food/household expenses), a greater proportion of current members use

the loans to invest in productive activities. This large proportion that invests in

productive, income-generating activities may help to explain the desire or even ability

of current members to remain in the program.

In summary, the VSLA program offers access to relatively large sums of

money, through both loans and the final savings payout. These funds are used for a

variety of purposes, primary among them being food, school fees, family

celebrations, housing improvements and productive investment. Furthermore, there

does not appear to be a large difference between current members and dropouts in

terms of the nature and extent of their VSLA participation.

b. Impacts of VSLA Participation

Overall, VSLA members report an overwhelmingly positive impact from

program participation. However, from Tables 10 and 11, it appears as if the current

members have benefited more from program participation than the dropouts.

Nonetheless, in general, dropouts do not appear to have been negatively impacted, but

rather simply experienced little effect from their participation in the VSLA program.

Without in-depth interviews it is impossible to determine whether the dropouts left

the program because they saw little change in their social status, or if they saw little

97

change because they left the program too quickly, without giving it a chance to yield

positive changes.

The majority of VSLA members report a positive impact on their household’s

diet from participating in the program. As shown in Table 10, 75 percent of current

members believe their household’s diet has improved since joining the program.

However, only 47 percent of dropouts report a positive impact. Meanwhile, only 23

percent of current members believe their household’s diet has stayed the same,

compared to 47 percent of dropouts. A similar proportion of each group responded to

questions about diet status with ‘worsened’ or ‘I don’t know.’

The health of the household also appears to be positively impacted by VSLA

program participation. The vast majority (80 percent) of current members believe that

the health of members of their households has improved since joining the VSLA

program. However, only 58 percent of dropouts mention a positive impact of program

participation on the health of their household. Just over 18 percent of current

members judge that the health status of the household has not changed since joining,

but almost 37 percent of dropouts make that statement. Furthermore, 5 percent of

dropouts believe that the health of the members of their household declined after

joining the VSLA (zero percent of current members reported worsening health status

for members of their households). Nonetheless, overall, participation in a VSLA has a

strong self-reported positive impact on the health status of the household.

In addition to improvements in member households’ diet and health, the

VSLA program has a substantial self-reported positive effect on the self-esteem and

social statuses of its participants. As illustrated in Table 11, a significant 84 percent of

98

current members believe their status in the community has improved since joining the

VSLA program, 85 percent noted an improvement in their status in their families, and

89 percent deem that their self-confidence has improved since joining the program.

However, again, current members appear to have seen greater improvements in their

social status than dropouts. Only 55 percent of dropouts report an improvement in

their status in the community, 50 percent reports an improvement in status in the

family, and 55 percent report an improvement in self-confidence. It is possible that

the dropouts did not remain in the program long enough to see improvements in their

social status or self-esteem. Alternatively, perhaps many left the program because

they did not see the expected social benefits. Either way, the overwhelming majority

of VSLA members report a positive benefit of program participation on their self-

confidence and social status within the community and the family unit.

Although, there does not appear to be a substantial difference between current

members and dropouts in terms of the specifics of VSLA participation, members of

the two groups seem to have been impacted very differently by the program.

Therefore, in order to prevent an upward bias in the estimates of program impact, it

was prudent to include the dropouts in the treatment group.

Overall, the VSLA program appears to have a positive impact on the nutrition,

health, and social status of the majority of its members. Given the self-reported nature

of the responses, too much weight should not be placed on these results. However,

they do offer an additional perspective on program impact and serve to supplement

the more robust findings in the report. These apparent positive impacts on the

nutrition and health of member households will be analyzed more rigorously in the

99

next chapter through the use of regression analysis. The social impact of program

participation will be further analyzed in the following section through the use of the

qualitative data gained during the focus group discussions.

iv. Impacts at the Individual Level

As discussed previously, the self-reported estimates of the effect of program

participation on members’ social statuses and self-esteem may not be entirely reliable.

As these impacts are insufficiently addressed by the survey data, in order to further

explore the findings in the previous section, we must rely more heavily on the

information from the focus group discussions.

Similar to the self-reported estimates in the previous section, the majority of

participants in the focus group discussions noted significant positive effects of

program participation on their social status and self-confidence. Within the family,

the majority of participants noted an improvement in communication and power-

sharing in the decision-making process. However, several participants, including a

few women, still attributed all decision making power to men. This is a reflection of

the strongly ingrained male dominant culture in Tanzania. One of the focus group

discussions was composed of only female members, while the remaining two were

split about 50/50 between men and women. In these meetings, the women generally

sat quietly, while the men controlled the conversation. This suggests that there has

been little social change or, more specifically, female empowerment in these groups.

However, the majority of focus group participants reported an improvement in

self-esteem as a result of membership in the VSLA program. Women, in particular,

noted their increased ability to address their own problems and provide for

100

themselves and their families. While many women used to sit idle, waiting for their

husband to provide for them, they have now learned how to save and how to use this

savings, in combination with the loans, to generate their own income. Many cited

their increased contribution to household income as a source of increased power in

the decision-making process.

Another commonly cited benefit of the VSLA program was an improvement

in social capital. The VSLA program allows the member to meet together every week

in an environment where they can exchange news and ideas. It creates a feeling of

cooperation and community. Several members also cited an improvement in their

status in the community. These impacts will be discussed at greater detail in the

following section.

v. Impacts at the Community Level

Many microfinance practitioners have asserted that even those members of the

community who do not participate in the program benefit simply by living in program

villages. It is again very difficult to quantitatively measure such impacts at the

community level. This type of analysis generally requires an immensely large data

set, which is expensive and time-consuming to collect, that, in addition to the

program participants, includes eligible non-participants in program villages.

However, the studies that have undertaken this method have generally found a

positive program impact on a variety of indicators, including poverty, household

income, and women’s empowerment, for other, non-participating, members of the

community (Khandker 2005; Hashemi, Schuler and Riley 1996). Although, the time

and financial constraints of this study prohibited the individual questionnaire from

101

addressing program impacts at the community level, the focus group discussions offer

some insight into such impacts.

Almost every focus group participant noted the respect given to VSLA

members by the community. The participants in the VSLA program serve as self-

proclaimed role models for the community. Several have even helped a neighbor start

his own small business in order to generate income. While this suggests that there

may be some degree of program impact on the household income of non-VSLA-

members in the community, without more detailed information, this conclusion

cannot be considered robust. Nevertheless, as they witness the success of the current

members, other members of the community have begun to express interest in joining

the VSLA program. However, in most cases, the group is at full capacity with 30

members. This may eventually cause tension within the community. However, as

non-participating members of the community were not included in the discussions, it

is impossible to assess the validity of such a possibility. In general, these self-reported

impacts at the community level may be slightly biased and should not be taken at face

value. Nonetheless, the VSLA program does appear to have a moderately positive

impact on the community as a whole.

102

CHAPTER 4 Empirical Results at the Household Level

The outcome variables of interest whose means differ significantly between

the treatment and the control group are now further examined using regression

analysis. Although mean comparisons present an initial observation of program

impact, regression analysis permits us to take these estimations a step further.

Preliminary comparisons of the basic characteristics of respondents, presented in

Table 1 and 2 in the previous chapter, suggest that the two groups are relatively

comparable, thereby validating the conclusions drawn from the mean comparisons.

However, regression analysis allows us to actually statistically control for these

characteristics, producing more robust estimations of program impact. It is also

possible to explore differential program impact by gender through regression

analysis. Given our hypothesis that program participation may have a greater impact

on the welfare of the household for female clients, this is an important benefit. By

incorporating both a discrete participation variable as well as a continuous variable

for the number of years of participation, regression analysis also allows us to examine

the timing of program impact.

103

I. Ordinary Least Square (OLS) Results

i. Sources of Income

A large portion of the literature on microfinance focuses on the impact on

household income. Due to the difficulty and expense involved in collecting such

parameters, this study did not collect data on the level of income in a household, but

rather focuses on the diversification of income sources. The literature suggests that

microfinance participation allows clients to significantly diversify their sources of

income (Zaman 2000; MkNelly and Dunford 1999). The initial training in the VSLA

program, which focuses on the promotion of Income Generating Activities (IGAs), in

combination with access to the loans and the payout from the VSLA, is expected to

facilitate diversification of member household’s income.

In Table 12, regression analysis is used to measure the impact of program

participation on the number of IGAs a household is currently involved in.4 Overall,

the results seem to confirm the basic hypothesis – in column (1) and (2), which are

clustered by VSLA group and village, respectively, program participation appears to

have a positive and significant impact on the outcome variable. Due to the inclusion

of the interaction term, the coefficient on membership, which is significant at the 5

percent level, may be interpreted as the effect of participation for male members only.

Given that the average number of IGAs per household is between 1 and 2, the size of

the coefficient may be considered practically significant as it suggests that

membership in the VSLA increases the number of IGAs of households of male

members by 0.368. The positive coefficient on gender, although neither statistically

4 Eight respondents did not answer this question and are thus omitted from the analysis presented in Table 12.

104

nor practically significant, suggests that households of female respondents, without

any program impact, operate a greater number IGAs than households of male

respondents.

The positive coefficient on the interaction term, although it is not statistically

significant, implies that program participation also has a positive impact on the

outcome variable for women. From the linear combination of membership and the

interaction variable, we see that households of female VSLA members are, on

average, involved in 0.524 more IGAs than non-member households. The coefficient

of the linear combination is highly significant at the 1 percent level. Overall, these

results suggest that program participation is significant for both men and women, and

there is no significant difference in the impact by gender.

The final two specifications, which incorporate the continuous dosage

variable, representing the number of years respondents have been members of a

VSLA group, generally support the basic hypothesis. In column (3), the coefficient on

membership, which is significant at the 5 percent level, suggests that households of

VSLA members operate on average 0.285 more IGAs than non-members. Note that

column (3) does not include an interaction variable and therefore, the coefficient on

membership must be interpreted as the program impact for all VSLA members, male

and female. The coefficient on dosage is statistically insignificant, which suggests

that VSLA program participation has an all-or-nothing effect on a household’s

number of IGAs. This may indicate that the gains for program participants are

primarily derived from the initial training received upon entry and therefore, do not

increase with time.

105

In column (4), which includes two interaction terms, the coefficient on

membership is significant at the 10 percent level. The magnitude of the coefficient

indicates that the households of male VSLA members operate, on average, 0.439

more IGAs than those of non-members. The positive coefficient on the linear

combination term, although it is statistically insignificant, suggests that the VSLA

program also has a positive impact on female members. The coefficient on the

interaction term between membership and gender is statistically insignificant meaning

there is no significant difference in impact by gender.

The coefficient on dosage is again statistically insignificant. However, both

the interaction term of dosage and gender and the associated linear combination term

are significant at the 5 percent level. This suggests that program participation has a

significant positive effect over time for female VSLA members. However, the impact

may not be considered economically significant as the magnitude of the linear

combination term implies that for each year of membership in the VSLA, the

households of female members operate 0.081 more IGAs than non-members.

The regression results, on the whole, confirm popular theory – program

participation has a practically and statistically significant effect on the number of

IGAs a household is involved in for all members, male and female. For male VSLA

members, program participation appears to have an all or nothing effect, rather than

an increasing effect with time in the program. However, for female members, there

may be both a discrete impact of the program and a marginal increase in number of

IGAs for each additional year of program participation. Given the importance of

IGAs in the VSLA methodology, as well as the apparent discrete effect of

106

participation, these results may also suggest a sizeable benefit of the basic training

module on program participants.

These quantitative results are further corroborated by the focus group

discussions. Participants of the discussions listed a variety of small businesses, each

of which were funded using a loan or savings payout, including the sale of khangas5;

selling bread, oranges, oil, etc.; transporting oranges to the market; raising ducks and

chicken to sell; and selling charcoal and firewood. One participant used a Tsh100,000

(US$90) loan to purchase a used sewing machine. Now she is one of the most

successful tailors in the region. Another member used a loan to purchase a dhow6 and

fishing nets, and now runs a small but profitable fishing operation. One VSLA group

planted a tree farm, the product of which can be sold at Tsh30,000 (US$27) for

twenty pieces for construction purposes. Overall, VSLA participation seems to have a

positive impact on the growth and diversification of income sources.

ii. Household Assets

Given the difficulties and large expense involved in measuring income, a

household’s asset expenditure is used as a proxy. Microfinance programs are

generally reported to have a positive impact on the level of household expenditures

(Pitt and Khandker 1998; Khandker 2005). The results displayed in Table 13

generally confirm this theory, suggesting that membership in the VSLA facilitates a

higher level of spending on household assets.7 In columns (1) and (2), the coefficient

on membership, which denotes program impact for male members, is statistically

5 A traditional piece of fabric worn by many East African women. 6 A traditional Swahili fishing boat. 7 Thirty-seven respondents did not report a value for their asset expenditure in 2009 and were thus dropped from Table 13.

107

significant at the 1 percent level. Furthermore, it is practically significant, as the size

of the coefficient suggests that male VSLA members spend, on average, Tsh116,227

(US$106) more on household assets than non-members. This represents a sizeable

benefit, considering the average annual income in Tanzania is approximately

Tsh1,367,300 (US$1,243) (Human Development Report 2009).

From the linear combination term, which is also highly statistically significant

at the 1 percent level, we see a positive impact of VSLA program participation on the

expenditure levels of households of female members as well. The magnitude of the

coefficient indicates that female members spend Tsh87,162 (US$79) more on

household assets than non-members.

The coefficient on gender, which is significant at the 10 percent level,

suggests that initially, without program intervention, women spend Tsh21,154

(US$19) less on household assets than men. Furthermore, although the coefficient on

the interaction term is statistically insignificant, its negative sign implies that program

participation has a smaller impact on female members. The literature suggests that

women are more likely to spend on education and health than their male counterparts

(Strauss and Beegle 1996; Hoddinott and Haddad 1994). Therefore, it is possible that

we see a lower initial expenditure level and a smaller program impact for women,

because they are spending more of their available resources on their children’s

education, nutrition and healthcare, rather than on household assets.

The overall hypothesis of a positive impact on asset expenditure is further

supported by the alternative specifications, which include dosage. In column (3), the

coefficient on membership is significant at the 1 percent level. The magnitude of the

108

coefficient indicates that VSLA members spend approximately Tsh91,120 (US$83)

more on household assets than non-members. The coefficient on dosage is

statistically insignificant.

In column (4), when the interaction terms are included, the coefficient on

membership is significant only at the 10 percent level, but this is due to a larger

standard error since the point estimate is actually higher than in column (3). The

magnitude of the coefficient indicates that male VSLA members spend approximately

Tsh129,000 (US$117) more on household assets per year than non-members. The

linear combination term is also statistically significant at the 10 percent level, which

confirms the findings in columns (1) and (2) that VSLA program participation has a

significant impact on both male and female members. The magnitude of the

coefficient on the linear combination term indicates that female VSLA members

spend Tsh71,906 (US$65) more on household assets than non-members. The

coefficients on dosage and on the associated linear combination term are both

statistically insignificant. This indicates that the VSLA program has a discrete, rather

than a continuous, effect on asset expenditures.

The information gathered from the focus group discussions confirms this

overall finding of a positive impact of program participation on household asset

expenditure. Focus group participants listed a plethora of items financed by either the

payout or a loan from a VSLA. This list includes such items as gold earrings, a

cupboard, ceramic plates, khangas, and even a goat, which was pregnant at the time

of the study. In summation, VSLA program participation may be said to have a

109

significant and positive impact on the level of asset expenditure of member

households.

iii. Education

Several studies in the literature have found a positive impact of microfinance

on education (Littlefield et al. 2003, Neponen 2003). Others, however, have found a

positive impact on the education of boys, but no effect on that for girls (Barnes 2001;

Dunn and Arbunkle 2001; Todd 2000). Participation in the VSLA program is

expected to increase the level of education attainment and/or the quality of education

received, by facilitating a higher level of education expenditure through consumption

smoothing. However, the quantitative results to support this theory are weak, though,

they tend in the expected direction. Overall, we can roughly speculate that the VSLA

program has a marginal positive impact on a household’s level of education

expenditure.

Table 14 presents the four basic specifications in columns (1) – (4) using a

household’s education expenditures in 2009 as the outcome variable.8 Neither the

coefficient on membership nor on the linear combination term is statistically

significant in any of the specifications. Although it is statistically insignificant, it is

interesting to note that the coefficient on membership in column (1) and (2) is

negative, which implies a negative program impact on the education expenditures for

households of male VSLA members. However, the coefficient on the linear

combination term is positive, indicating a positive effect for households of female

members. In column (3), when dosage is included, the coefficient on membership

8 Six respondents did not report a value for education expenditures in 2009 and are subsequently dropped from the analysis presented in Table 14.

110

comes in as positive, suggesting a positive impact on education expenditures for

households of all members. In column (4), when the interaction terms are

reintroduced, the coefficient on membership remains positive while that on the

associated linear combination term becomes negative. This is the opposite of the

direction of impact suggested in the first two columns. Although none of the

coefficients are statistically significant, this inconsistency may reflect a potential

weakness in the data. In both columns (3) and (4), the coefficients on dosage and the

linear combination term for dosage are also statistically insignificant. Overall, this

suggests that VSLA participation has no impact on a household’s education expenses,

contrary to our hypothesis.

Given the wealth of literature on the impact of microfinance on education, this

result is surprising. It is possible that educational expenses are not an appropriate

proxy for the program’s impact on education, especially considering that primary

education is provided tuition-free by the national government. However, time and

financial constraints limited the choice of education-related variables that could

feasibly be collected. Therefore, in order to explore the direction of program impact,

an alternative specification is presented in the final two columns.

With only 170 observations, the sample size is relatively small. Therefore, to

increase the power of the test, the interaction term between membership and gender is

dropped. The coefficient on membership must now be interpreted as the impact of

program participation on education expenditures for all members, both male and

female. Furthermore, the three binary variables on educational attainment are

replaced by a single continuous variable, educ attain. Although the coding of educ

111

attain is not continuous, it does increase systematically from least to most education –

the direction in which we would expect impact to run. Although this adjustment is not

very robust, it is simply used to demonstrate that the results move in the anticipated

direction.

There are only four non-Muslim households in the entire sample. The three

Christian households may be considered outliers as they spend disproportionately

more on education than non-Christian households, evident in the absurdly large and

statistically significant coefficient on Christian in columns (1) – (4). In order to avoid

any possible biases that might result from such outliers, columns (5) and (6) are run

only on the Muslim population. Therefore, these results may not generalize to all

religions.

In column (5), which is clustered by VSLA group, the coefficient on

membership is statistically significant at the 5 percent level. The magnitude of the

coefficient indicates that households of VSLA members, both male and female, spend

approximately Tsh29,869 (US$27) more per year on education than those of non-

members. Tuition for public secondary school is approximately Tsh20,000 (US$18)

per year (World Bank 2009). Therefore, considering that most families have more

than one child of school age, the size of the coefficient on membership may not be

considered practically significant. Although the coefficient on membership remains

positive in column (6) when clustered by village, it becomes statistically insignificant.

Though the results for education are not highly robust, a slight modification to

account for the small sample size causes the results to re-enter in the expected

direction and participation in the VSLA program is found to have a significant impact

112

on a household’s yearly education expenditures. Therefore, we can tentatively

conclude that the VSLA program has a marginal positive impact on either a

household’s level of educational attainment and/or quality of education. This

conclusion is further supported by the findings in the focus group discussions. The

increased ability to finance the education of their children, including tuition fees,

materials, testing fees, etc., was the most commonly cited benefit of program

participation by focus group participants. One member took of a loan of Tsh100,000

(US$90) to send her two daughters to a secondary boarding school – an opportunity

which would most likely have been closed to them under other circumstances.

iv. Nutrition and Health The majority of the related literature has found that the households of

microfinance clients have, on average, better nutrition and health statuses compared

to non-client households, especially when the client is female (Pitt et al. 2003; Pronyk

et al. 2007; Barnes 2001; Littlefield et al. 2003). This study strongly supports this

finding. The quantitative data suggests that VSLA participation has little effect on

meal quantity, but has a substantial positive impact on meal quality, evident through

an increase in consumption of both fish and meat. The VSLA program also appears to

improve access to health services for member households, by facilitating a higher

level of spending on healthcare.

a. Meal Quantity

The findings in Table 15 corroborate the general finding in the literature that

women, both VSLA members and non-members alike, are more likely to spend on

the diet of their household. However, VSLA participation appears to only impact the

113

number of meals per day for households of male members, and has no effect on the

households of female members.

In the first two columns, which are clustered by VSLA group and village,

respectively, the coefficient on membership is significant at the 5 percent level. The

size and the positive sign of the coefficient suggest that households of male VSLA

members have, on average, 0.337 more meals per day than those of non-members.

The coefficient on gender, which is significant at the 1 percent level, implies that,

without program intervention, households of female respondents consume 0.348 more

meals per day than those of male respondents. The coefficient on the interaction term,

which is significant at the 5 percent level, is negative indicating that participation in

the VSLA program has a much smaller impact on female members. The linear

combination term confirms this result as it is statistically insignificant, suggesting that

program participation has no effect on average number of meals per day for

households of female VSLA members. Overall, it appears that when men participate

in the VSLA program, their households are brought in line with those of women, both

members and non-members, in terms of number of meals. This may suggest that

households of female respondents are getting the optimal number of meals per day

even without program intervention. Women likely devote a greater share of their

available resources to nutrition, regardless of their wealth. Men appear to only spend

on their household’s diet once they have greater available resources, thus confirming

the hypothesis that men place less importance on the diet of their household than

women.

114

When dosage is included in column (3) and (4), the results change slightly. In

column (3), the coefficient on membership is statistically significant at the 5 percent

level and is of a similar magnitude to the coefficients found in the first two columns.

As there is no interaction term in the third column, the size of the coefficient on

membership indicates that for both male and female members simply being a part of

the VSLA program increases the household’s number of meals per day by 0.335. The

coefficient on dosage, which is significant at the 5 percent level, comes in as

negative, suggesting that the number of meals in the households of VSLA members

decreases with each additional year of program participation.

When the interaction terms are included in column (4), a significant and

positive impact of program participation is found for both male and female members,

which is inconsistent with the results found in the first two columns. The coefficient

on membership, which is significant at the 5 percent level, suggests that households

of male VSLA members consume 0.506 more meals per day that those of non-

members. The coefficient on gender, which is significant at the 5 percent level, is

consistent with the results in the first two columns, implying that without program

impact, households of female respondents consume, on average, a greater number of

meals per day. However, unlike in column (1) and (2), the linear combination term

between membership and the associated interaction term, is positive and significant at

the 5 percent level. The magnitude of the coefficient indicates that households of

female VSLA members consume 0.298 more meals per day that those of non-

members. The coefficient on the interaction term between membership and gender is

115

statistically insignificant, which suggests that the discrete impact of VSLA

participation does not vary significantly by gender.

The coefficient on dosage becomes statistically insignificant in column (4),

but the associated linear combination term becomes significant at the 5 percent level.

For female VSLA members, each additional year of participation seems to decrease

the number of meals consumed in the household by 0.067. The interaction term

between dosage and gender, however, is statistically insignificant which implies that

the difference in program impact over time is not significant by gender.

In summation, the quantitative results suggest that without program

intervention women invest more in the diet of their household. Participation in the

VSLA program, however, brings the households of male members up to the level of

those of female respondents in terms of number of meals per day. Again, perhaps this

signifies the existence of some optimal number of meals per day, which women may

achieve without program intervention. Meanwhile, men only reach this number

through the increased availability of resources made possible through program

participation. When dosage is included in the specification, the results are slightly

surprising as they suggest that each additional year of participation for female

members decreases the number of meals consumed in the household. However, the

estimated coefficient is very small.

b. Meal Quality

Meat Consumption

In addition to meal quantity, the quantitative results indicate that VSLA

participation has a considerable impact on meal quality, evident in an increase in the

116

quantity of meat and fish consumed in the past week. While the results for the

consumption of meat are slightly less robust that those for consumption of fish,

overall, they indicate a positive and significant impact of program participation on the

quality of a household’s diet.9

In Table 16, the VSLA program appears to have a positive and significant

impact on a household’s consumption of meat, particularly for those of female

members. In columns (1) and (2), which are clustered by VSLA group and village,

respectively, the coefficient on membership is statistically insignificant, which

suggests that VSLA program participation has no significant impact on meat

consumption for the households of male members. The coefficient on the linear

combination, however, is significant at the 10 percent level when clustering by VSLA

group and at the 5 percent level when clustering by village. The magnitude of the

coefficient implies that households of female VSLA members consume meat 0.287

more days per week than non-members.

The coefficient on gender, although it is insignificant, is negative, which

implies that, without program intervention, households of female respondents

generally consume meat approximately 0.2 less days than those of male respondents.

This runs contrary to our general hypothesis that suggests that women are more likely

than men to invest in the household’s diet. Perhaps households of female members

have lower incomes, in general. Income is not controlled for in the regression.

Furthermore, the coefficient on gender is negative for asset expenditure in Table 13,

suggesting that without program impact, women also spend less on household assets.

9 Both Table 16 and 17 are missing two observations. Again, keep this in mind when interpreting the results.

117

Together, these results may suggest that households of female VSLA members, on

the whole, have a lower-income than those of male members.

Alternatively, given the relatively high expense of meat in Zanzibar, women

may spend a greater proportion of their resources on more cost-effective food items

such as grains or fish, rather than on expensive meat. They may be more concerned

with meal quantity than quality – a theory supported by the significant and positive

coefficient on gender in Table 15. Correspondingly, female VSLA members

experience a significant program 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.

When dosage is included in column (3), the results become less robust. The

coefficients on both membership and dosage become statistically insignificant.

However, when the interaction terms are included in column (4), the results again

suggest a significant program impact on meat consumption for households of female

VSLA members. The coefficient on membership remains statistically insignificant.

However, the coefficient on the linear combination term between membership and the

associated interaction term is significant at the 10 percent level. The magnitude of the

coefficient implies that households of female VSLA members consume meat 0.439

more times per weak than households of non-members. The coefficients on dosage as

well as on the related linear combination term remain statistically insignificant. This

suggests a discrete impact of participation in the program, rather than one that varies

over time.

Overall, these quantitative results suggest that program participation has a

118

significant impact on meat consumption for households of female members but not

for those of male members. Women appear to initially spend less on meat than men,

which may imply that they are more concerned with meal quantity than quality. They

only begin to invest in meat when they are wealthier, which is why we see a positive

program impact for female members.

Fish Consumption

The regression results presented in Table 17 indicate that participation in the

VSLA program has a substantial impact on fish consumption. In columns (1) and (2),

which are clustered by VSLA group and village, respectively, the coefficient on

membership is highly significant at the 1 percent level. Furthermore, the magnitude of

the coefficient may certainly be considered practically significant, as it suggests that

households of male VSLA members consume fish 3.483 more times per week than

those of non-members – a sizeable difference. The linear combination term is also

statistically significant at the 1 percent level. The size of the coefficient indicates that

households of female VSLA members consume fish 3.691 times more per week than

those of non-members – an even greater improvement than that for households of

male members.

When dosage is included in the final two specifications, the coefficient on

membership remains statistically significant at the 1 percent level. In column (3) the

magnitude of the coefficient indicates that households of VSLA members consume

fish 3.260 more times per week than those of non-members. In column (4), which

includes the two interaction terms with gender, the coefficient on membership

indicates that households of male VSLA members consume fish 2.939 more times per

119

week than those of non-members. The linear combination term between membership

and the associated interaction term, which is also significant at the 1 percent level,

indicates that VSLA program participation also has a positive effect on the fish

consumption of households of female members. More specifically, households of

female VSLA members consume fish 3.459 more times per week than those of non-

members.

In both columns (3) and (4) the coefficient on dosage as well as on the

associated interaction term are statistically insignificant. This suggests that VSLA

participation has a discrete, all-or-nothing effect on a household’s fish consumption.

This suggests that it may simply be having access to credit and savings that increases

a household’s consumption of fish. Participation in the VSLA program may facilitate

consumption smoothing within the household, thereby allowing more to be invested

in the quality of the household’s diet.

Overall, the quantitative analysis suggests that VSLA participation has a

positive impact on a household’s fish consumption for both male and female

members. The program appears to have a strong positive effect on the quality of

meals for member households, especially those of female members.

c. Health Expenditures

Theory suggests that microfinance program participation increases a

household’s ability to finance and thus access healthcare, eventually improving the

household’s health status. The results presented below for health expenditures are

generally not statistically significant, though they move in the expected direction.

120

In columns (1) and (2) in Table 18, the magnitude of the coefficient on

membership suggests that male VSLA members spend approximately Tsh28,000

(US$26) more on healthcare than non-members.10 However, the coefficient is not

statistically significant. The coefficient on gender is also insignificant in column (1),

which is clustered by VSLA group, but is significant at the 5 percent level in the

second column, which is clustered by village. The size of the coefficient suggests that

without program intervention women spend roughly Tsh23,000 (US$21) less on

healthcare. The linear combination term, which is significant at the 10 percent level

when clustering by VSLA group, shows that women experience a positive impact

from VSLA participation. Female VSLA members spend approximately Tsh21,000

(US$19) more on heath expenditures than non-members. Overall, the VSLA program

appears to have a moderately significant impact on the health expenditures for the

households of female members only. This supports our hypothesis that women are

more likely to spend on the healthcare of their family when they have access to the

necessary resources.

When dosage is included in column (3) and (4), the coefficients on all of the

variables of interest become statistically insignificant. Although they are statistically

insignificant, the coefficients on both membership and the related linear combination

term remain positive.

Although many of the key variables are not statistically significant, overall,

the results seem to move in the anticipated direction. The coefficient on the linear

combination term between membership and the associated interaction term is

10 Ten observations were dropped from Table 18 as the respondents did not report a value for their 2009 health expenditures.

121

significant at the 10 percent level when clustering by VSLA group. Therefore,

although program participation may be insignificant for male members, it may have a

positive impact on health expenditures for female members.

Overall, the qualitative results support this finding, indicating a positive effect

of VSLA program participation on the nutrition and health statuses of member

households. Many of 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 to program membership. One focus group

participant attributes her son’s life to the VSLA program. As a child, her son was

very sick. She was able to take a Tsh100,000 (US$90) loan to bring him to Dar es

Salaam where he received treatment what would otherwise have been inaccessible.

II. Probit Results

i. Health

a. Use of Mosquito Nets

The quantitative findings in the previous section suggest that the VSL

program has a moderate positive impact on the level of health spending of the

households of female members. Here this impact on health is further explored using a

dummy variable indicating whether or not the children in the household sleep under a

mosquito net as a proxy for investments in healthcare. Given the pervasiveness of

malaria in Tanzania, a mosquito net is likely one of the most important investments a

household can make in the health of its children.

122

Overall, the results in Table 19 weakly confirm our general hypothesis of a

positive impact of program participation on the frequency of use of mosquito nets.11

The coefficient on membership is significant at the 10 percent level under the probit

model when clustering by village in column (3). The magnitude of the marginal effect

indicates that the children of male VSLA members are 5.1 percentage points more

likely to sleep under a mosquito net than those of non-members. Although the

coefficient on membership is statistically insignificant in column (1), (2) and (4), it

remains positive across the board. The coefficient on the linear combination term is

statistically significant when using the probit specification - at the 5 percent level in

column (1) when clustering by VSLA group and at the 10 percent level in column (3)

when clustering by village. This suggests that program membership has a more

significant impact on the likelihood of sleeping under a mosquito net for the children

of female VSLA members than on the likelihood for those of male members. The

marginal effect of the linear combination term implies that the children of female

VSLA members are approximately 5.9 percentage points more likely than those of

non-members to sleep under a mosquito net. Remember that 97.4 percent of the

children in the treatment group sleep under a mosquito net, compared to only 90.9

percent of those in the control group.

The marginal effect on gender, although the coefficient is not statistically

significant, implies that the children of female respondents are around 2 percentage 11 Several of the control variables were dropped from the regression because they predict success perfectly. Both Christian and Other religion were dropped – all of the children of every Christian respondent, as well as those of the one ‘Other,’ sleep under a mosquito net. All of the children of the ‘widowed,’ ‘divorced’ and ‘separated’ respondents also sleep under a mosquito net. Finally, the children of those with previous access to loans sleep under a mosquito net. Therefore, all of these variables are dropped from the regression. Note that ten respondents were omitted from the analysis presented in Table 19 as they did not answer the question.

123

points more likely to sleep under a mosquito net than those of male respondents. This

confirms the general finding in the literature, which suggests that women are more

likely than men to spend on their children’s education and health. The positive

coefficient on the interaction term, which is also statistically insignificant, implies

that the children of female VSLA members are more likely to sleep under a mosquito

net than those of male members. This further confirms our hypothesis - when women

have access to the necessary resources (i.e. through participation in the VSLA

program), they are even more likely to spend on their children’s health and well-

being, in this case through the purchase of mosquito nets.

In column (5) and (6), the coefficients on both membership and dosage are

statistically insignificant. However, when the interaction terms are included in

column (7), the results indicate a significant program impact over time for female

members. The coefficient on membership and the associated linear combination term

remain statistically insignificant. The coefficient on dosage is also insignificant.

However, the coefficient on the related linear combination term in column (7) is

significant at the 5 percent level. Nonetheless, the findings may not be considered

practically significant as the marginal effect of the linear combination term indicates

that each additional year of program participation increases the likelihood that the

children of female VSLA members sleep under a mosquito net by only 0.7 percentage

points.

Although many of the key variables are statistically insignificant, the results in

Table 19 generally appear to confirm our expectations – participation in the VSLA

program increases the likelihood of children sleeping under a mosquito net. This may

124

suggest that the children of VSLA participants are more likely to be healthy than

those of non-participants, indicating that the VSLA program has an overall positive

impact on the health status of member households. This supports the quantitative

findings from the OLS regressions in the previous section as well as the qualitative

findings from the focus group discussions.

ii. Quality of Housing

The findings in the literature suggest a positive impact of microfinance

program on both the quality of housing as well as on the level of investment (Hossain

1988; Neponen 2003). The results presented here support these findings in the

literature. VSLA members seem to be more likely to both own their own home and to

make subsequent investments in the quality of this home.

a. Home Ownership

Membership in the VSLA program is expected to increase the resources

available to a household, which often enables them to purchase their own home. In

Table 20, we see that VSLA program membership does, in fact, have a highly

significant impact on home ownership.12 Furthermore, the results do not appear to

vary substantially between the probit and the LPM specifications. In column (1) and

(2), which are clustered by VSLA group, the coefficient on membership is significant

at the 5 percent level. In column (3) and (4), which are clustered by village, the

coefficient is also statistically significant - at the 1 percent level when using probit

and at the 5 percent level under the LPM. Across all four columns, the magnitude of

the coefficient (or marginal effect) on membership is approximately 0.3, indicating

12 Note that Christian and other religion were dropped from Table 20 as each predicts success perfectly under the probit model. In order to maintain consistency in number of observations between the different methods, the two religion variables were dropped in the LPM as well.

125

that if you had two otherwise identical households, those of male VSLA members

would have a 30 percentage points greater chance of owning their own home than

those of non-members. To put this number into perspective, recall from Table 3 that

85.8 percent of the treatment group own their own home, compared to only 60

percent of the control group.

The linear combination term is also highly significant at the 1 percent level

across columns (1) – (4). The predicted size of impact is similar between probit and

the LPM. Under the probit specification in columns (1) and (3), the marginal effect of

the linear combination term indicates that, like male members, female VSLA

members are approximately 30 percentage points more likely to own their own home

than non-members. However, the coefficient under the LPM suggests that female

members are only 27.6 percentage points more likely than non-members to own their

own home.

From the results in column (1) – (4), we see little differential in program

impact between male and female members. Buying a house is a major investment. As

such, it is likely that the decision is made either together as a couple, or by the

husband alone. It is unlikely that a woman, even if she were the one with access to the

necessary resources through participating in a VSLA, would be solely responsible for

making such a weighty decision. Therefore, as the decision is likely made together, it

is reasonable that program impact would not change significantly by gender of the

participants. The consistency across male and female members is confirmed by the

extremely small magnitude and statistical insignificance of the coefficient on the

interaction term across all four specifications.

126

Nonetheless, the coefficient on gender, which is statistically significant at the

10 percent level in column (3), signifies that households of female respondents are

10.7 percentage points more likely to own their own home than those of male

respondents. The coefficient on gender is also significant in column (5), when the

continuous variable dosage is included, and the marginal effect again suggests that

households of female respondents are 10.7 percentage points more likely to own their

own home than those of male respondents.

In column (7) and (8), when dosage as well as both interaction terms is

included, the coefficient on membership is statistically insignificant. However, the

coefficient on the linear combination term between membership and the associated

interaction term is significant at the 1 percent level. This implies that the VSLA

program has a significant effect on home ownership for female members, but not for

male members. Although this finding differs slightly from that in columns (1) – (4),

as the coefficient on membership remains positive, the two findings do not contradict

one another. The marginal effect of the linear combination term in column (7)

indicates that households of female members are 34.2 percentage points more likely

to own their own home than households of non-members.

The coefficients on dosage and the related linear combination term are

statistically insignificant in columns (5) – (8). This suggests that VSLA program

participation has an all-or-nothing effect on home ownership. This result may be

surprising as it is reasonable to expect that longer participation in the VSLA would

lead to higher income, which would, in turn, increase the likelihood of owning one’s

home. Alternatively, perhaps it is merely having access to credit and savings that

127

makes home ownership possible, as opposed to increasing income. Maybe the

program allows members to spend their money on different things rather than to

spend more money overall.

The above findings generally corroborate intuition, which suggests that VSLA

participation increases the resources available to a household, allowing them to

purchase their own home. The VSLA program appears to have a significant impact on

the probability of owning a home, particularly for female members.

b. Housing Improvements

In addition to owning their own home, VSLA participants are substantially

more likely to make improvements in the quality of their housing. In Table 21, in

columns (1) - (4), the coefficient on membership is significant at the 1 percent level.13

The magnitude of the coefficient/marginal effect indicates that male VSLA members

are approximately 54 percentage points more likely to make housing improvements

than non-members. The coefficient for the linear combination term is also highly

significant at the 1 percent level. The magnitude of the coefficient under the LPM is

slightly smaller than that of the marginal effect in the probit specification. However,

on average, the numbers indicate that households of female VSLA members are 57.6

percentage points more likely to making housing improvements than non-members.

When interpreting these numbers, remember that 67 percent of the treatment group

report having made housing improvements in the past year, compared to only 16

percent of the control group.

13 Note that Christian and other religion were dropped in Table 21 as well, as both predict success of housing improvements perfectly.

128

Although it is statistically insignificant, the negative coefficient and the

marginal effect of gender imply that, without program impact, women are between 13

and 17 percentage points less likely to make housing improvements than men. Given

that women are generally believed to be more likely than men to invest in the

education, nutrition and health of their families, this finding is not entirely surprising.

Although it is statistically insignificant, the coefficient on the interaction term is

positive, which suggest that female members experience an even greater increase in

probability of making housing improvements than their male counterparts. Overall,

this may be another indication of the general lower income of the households of

female members as suggested in Tables 13 and 16. Alternatively, it may be a signal of

women’s priorities – when women have few resources, they devote a greater

proportion to their children’s education, nutrition and health. Only once they have

access to more resources through participation in the VSLA program do they begin to

make the less critical investments in the quality of their housing.

When dosage is included in columns (5) and (6) the coefficient on

membership remains significant at the 1 percent level. The marginal effect of the

coefficient in column (5) implies that VSLA members, both male and female, are

49.9 percentage points more likely to make improvements in the quality of their

housing than non-members. Under the LPM in column (6), the coefficient on

membership is slightly smaller, suggesting that VSLA members are 47.2 percentage

points more likely to make housing improvements. The coefficient on gender, which

is statistically significant at the 5 percent level, is negative in both columns (5) and

(6). The magnitude of the marginal effect again is slightly larger than that of the

129

coefficient in the LPM specification. The marginal effect suggests that women are

15.5 percentage points less likely to invest in housing improvements, while the

coefficient in column (6) suggests that women are only 12 percentage points less

likely to make housing improvements – both of which corroborate our findings from

the first four columns.

In columns (7) and (8), the coefficient on membership is significant at the 5

percent level. The marginal effect is slightly larger than the magnitude of the

coefficient, but together they suggest that households of male VSLA members are

approximately 41 percentage points more likely to invest in housing quality than non-

members. The associated linear combination term, which is significant at the 1

percent level in both of the final two columns, suggests that households of female

members also experience a sizeable positive impact from VSLA participation. The

marginal effect of the linear combination term suggests that households of female

VSLA members are 54.1 percentage points more likely to make housing

improvements than those of non-members, while the coefficient in column (8)

suggests a slightly smaller impact of 51.9 percentage points.

In columns (5) – (8), the coefficients on both dosage and the related linear

combination term come in as statistically insignificant. Again this implies that the

VSLA program has a discrete, rather than an increasing, impact on housing quality.

This is likely a reflection of the positive impact of simply having access to savings

and loan services – housing quality increases immediately, rather than increasing over

time, which would presumably be the direction of impact of income growth.

130

Overall, the results in Table 21 suggest that the VSLA program has a positive

and significant impact on the likelihood of making improvements in the quality of

housing of its members. VSLA participation allows for increased consumption

smoothing though savings and borrowing, which, in turn, increases the likelihood of

making housing improvements.

The conclusion of a positive impact of the VSLA program on home ownership

and housing quality is further supported by the information gained in the focus group

discussions. A large proportion of focus group participants reported using either the

final payout or a loan, or both, to invest in household improvements. One member

used a Tsh100,000 (US$90) loan to install electricity in his home. Another member

uses a small proportion of her payout from each savings cycle to slowly accumulate

cement blocks in order to eventually to build a new, higher-quality home. These are

just a few of the many accounts of VSLA participants’ investments in the quality of

their housing.

131

CONCLUSION

I. Lessons Learned

Microfinance makes capital available to low-income people who would not

otherwise have access to financial services and is generally believed to be a cost-

effective humanitarian intervention. However, the empirical evidence to confirm this

hypothesis of an overall positive impact is limited. This study hopes to add to and

improve upon the available evidence. The results of the study corroborate many of the

findings in the existing literature, offer some potentially new insights and suggest

several lessons for the study of microfinance in general.

While microfinance was originally focused on providing credit services to

needy recipients, in the past ten years or so, microfinance practitioners have

increasingly argued for the importance of offering and promoting savings for program

participants. This study supports the conjecture by finding an overall positive impact

of participating in the savings-based VSLA program. VSLA participants assert that

saving in the home is almost impossible given the myriad of competing demands they

face, and they consequently pronounce themselves very grateful for the opportunity to

save. It appears from the data gathered by the survey instrument, as well as from the

focus group discussions, that saving has given most members the capacity to improve

their livelihood and that of their families, independent of the benefits of borrowing

132

from the VSLA. In the VSLA program, savings also facilitates the loan function,

which presents further opportunities to improve the overall well-being of the

household. Furthermore, establishing the program around members’ savings rather

than injections of donor capital also helps to create a sense of program ownership for

the members. This, in turn, helps to build self-confidence and a sense of community

among members. It may also facilitate loan repayment by increasing members’ sense

of liability and responsibility within the group.

The findings from this study and the general experiences of the VSLA

program in Zanzibar should also encourage a broader conception of the purposes and

potential benefits of microfinance. There is a widespread assumption that the sole role

of microfinance is to promote development of microenterprises by providing essential

capital. Clients theoretically use their loans only to invest in productive enterprises

and use the ensuing cash flow to repay the loan. However, there are many other

sources of potential benefits of program participation other than investments in

productive capital. One such benefit is that of consumption smoothing. The majority

of microfinance clients, including those from the VSLA program, are relatively poor

and face a variety of competing demands on their limited financial resources.

Sometimes, additional funds are necessary to cope with major life-cycle events or

emergencies, or to fund necessary housing improvements or education expenses.

The funds from VSLA program participation are used for a wide variety of

consumption purposes, including purchasing food, paying for school fees, family

celebrations, housing improvements and medical expenses. By supplying these funds

when needed, the VSLA program enables members to maintain a steady level of

133

consumption and prevents them from slipping into a more desperate level of debt and

poverty, thereby improving their chances of eventually moving up the income ladder.

This study demonstrates the numerous benefits that may arise through the fulfillment

of such basic needs, even with no change in household income. The lack of a dosage

effect, as presented in Chapter 4, suggests that the positive effects of the VSLA

program are mediated by access to credit and savings, rather than income growth. In

other words, the observed program benefits are a result of the increased ability to

spend on different items rather than the capacity to spend more overall. Furthermore,

contrary to previous assertions, investing in consumption has not prevented the VSLA

program from achieving financial sustainability. Even though a sizeable proportion of

members do not report productive investment as one of their top three uses of either

source of funds, the program has had relatively few problems with loan repayment

and the exit rate remains very low. In conclusion, it appears that microfinance

programs should consider allowing, perhaps even encouraging, their clients to use the

money to satisfy any of their household’s basic needs, rather than limiting the use of

funds to investment in income-generating activities.

II. Areas for Future Research

A concerted effort was made during this study to measure impact on a wide

variety of measures, while controlling for selection bias, and the results are generally

encouraging. However, time and financial constraints severally limited both the scope

and the methodological strength of the study. For future research projects, there are

134

several areas in which a few adjustments or additions could improve the strength of

the results.

First and foremost, simply increasing the size of the sample would increase

the precision of the results. The study may also benefit from the inclusion of other

populations in the control group, specifically, non-members in program villages

and/or eligible nonmembers in villages where the program is not active. Including

nonmembers in program villages might provide quantitative measurements of

program externalities within the community. If there are such externalities, the

inclusion of eligible nonmembers in villages where the program is not active might

provide a more realistic baseline for the general population, thus more robustly

controlling for selection bias.

In addition to an expansion of the sample size, the inclusion of several

additional parameters may improve upon the present study. Many studies focus on the

impact of microfinance on income, consumption and/or poverty levels. These studies,

however, are limited because, as previously mentioned, these parameters are difficult

and time-consuming to collect and there exists reasonable proxies from which

impacts may be estimated. However, including these measures in the study would

facilitate a greater level of comparison with the larger scale studies in the literature.

Similarly, a deeper analysis of female empowerment and social change would greatly

improve the depth and breadth of the study. Such an analysis, however, requires more

in-depth interviews with female participants as well as other members of the family

and the community and, as such, also requires a much greater time and financial

commitment.

135

This study’s largely insignificant findings for the impact of VSLA

participation on education expenditures are puzzling, particularly given the other

significant results in the literature . Considering the importance of education for both

the current and future welfare of the household, it may be worthwhile to investigate

this issue further using alternative parameters to measure a household’s access to or

quality of education – such as children’s educational attainment. Many of the studies

that have explored the impact of microfinance on education have found a differential

impact by gender of the child. In order to further explore the possibility of such a

differential, any education related parameter should be broken down by gender.

Considering the importance of income-generating activities (IGAs) in the field

of microfinance, a similar study may benefit from a deeper analysis of enterprise

dynamics. USAID’s AIMS methodology lays out a possible approach, which

examines enterprise growth through a variety of parameters, including enterprise

revenue, the value of enterprise assets and the number of employees in the enterprise.

III. Implications of the Results for the Sustainability of the VSLA model

The VSLA program in Zanzibar has performed well in terms of outreach.

When CARE left the area in 2004, there were 61 VSL groups. Today, there are 233

groups in Zanzibar, which represents an annual growth rate of approximately 34

percent. This compares favorably with an attrition rate of around 3 percent per year or

roughly 20 percent over the six-year period. The high growth rate in the number of

groups in addition to the low attrition rate suggests that there is a high demand for the

services offered by the VSLA program.

136

The existing groups have performed well, in general, in terms of financial

sustainability. The data from the individual questionnaire suggests that net savers may

experience a rate of return on their savings of up to 58 percent. The average of the

previous payout was approximately Tsh270,000 (US$245), while the average per

member share value was approximately Tsh156,000 (US$142). Therefore, the VSLA

program appears to be offering useful and beneficial financial services in an

environment where there are few alternatives.

Despite the apparent overall success of the VSLA program, late loan

repayment was a common concern raised during the focus group discussions and key

informant interviews. In one case, the group reported one of its members to the police

in hopes of recovering the missing funds. Occasionally, groups fail to recover the

loan completely, which could have a negative impact on the long-term sustainability

of the group and the model. However, although there is no available data on the

default rate, loan repayment does not appear to be a systematic problem. The average

maximum possible return on members’ savings (for net savers) suggests that the

majority of VSLAs are very financially sustainable.

Several focus group participants also expressed concerns over weekly share

contributions. Many have had trouble finding the necessary funds to meet the

required weekly contribution. For this reason, several groups have removed the fine

assessed for failing to contribute a share every week. However, this may threaten the

financial sustainability of the VSLA model and the success of the loan services.

Another common complaint articulated during the focus group discussions

was poorly trained and irresponsible Community Contact Persons (CCPs). Several

137

participants felt as if many of the CCPs had little additional training than the average

member. This is likely a failure of JOCDO in both the selection and training process

for the CCPs, and represents one of the major potential avenues for improving the

effectiveness of the VSLAs.

The final frequently cited grievance was the average interest rates charged on

loans. Although the interest rate is determined by each individual group, most charge

around 5 percent per month. As mentioned previously, this is a much lower rate than

that of moneylenders who often charge up to 30 percent per month, but is slightly

higher than that charged by NGO-MFIs, which generally charge less than 4 percent

per month (Mutesasira 1999, 10). Several focus group participants suggested that

such an exorbitant rate of interest discourages members from taking loans and makes

them less willing to make risky, but high-yielding investments. The latter may

actually be a benefit of such an interest rate as it may make the group more

financially stable in the long-run.

i. Sustainability of JOCDO and the Apex Model in General

JOCDO clearly plays a vital part in the continued success of the VSLA model

in Zanzibar. Although JOCDO continues to receive minor support from CARE in the

form of training and business guidance, the organization allows CARE to achieve

program sustainability without a strong continued presence. In the six years since

CARE’s departure from Jozani-Chwaka Bay in Zanzibar, JOCDO has played a major

role in the formation and development of new groups and has provided these groups

with necessary training and materials, as well as some form of initial monitoring, so

that they may eventually operate more effectively on their own.

138

JOCDO, however, is currently facing numerous challenges, chief among them

being poorly educated and trained management. The majority of the leadership of

JOCDO is from rural areas and has only a primary education. They have little

knowledge of management or bookkeeping skills and resist additional training from

CARE. If the management of the organization does not have adequate training, it may

be difficult to prepare and supervise CCPs at the level necessary to initiate, train and

support the VSLA groups.

There also may be concerns over the financial sustainability of JOCDO, a

problem which may be exacerbated by the lack of formal education or training for the

leadership of the organization. JOCDO faces a classic “free rider” problem, in that

among the current 233 VSLA groups in Zanzibar, only 103 are registered and paying

members of JOCDO. This represents a substantial financial loss, as the annual

subscription fee for each group is Tsh15,000 (US$14). JOCDO still provides a variety

of support services for non-registered groups, which places a major financial burden

on the organization, though the organization has several other income sources,

including the sale of VSLA kits to groups at a small margin, usually around

Tsh10,000 (US$9), as well as a portion of the training fee paid to the CCP from each

VSLA. Nonetheless, it is unclear whether or not these sources will be sufficient to

sustain the organization in the long run.

Participation in the VSLA program appears to have an overall positive impact

at the individual, household and community level. The quantitative data from the

individual questionnaire, in addition to qualitative results from the focus group

139

discussions and key interviews, generally demonstrate a positive impact on program

participants. While, on the whole, the direction of program impact on the chosen

outcome variables is consistent with the basic findings in the literature, the VSLA

program may not have as substantial an impact on its members as many of the larger

NGO-MFI programs, such as the renowned Grameen Bank in Bangladesh or

BancoSol in Bolivia. These organizations 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. However, inasmuch as the VSLA

approach does not rely on outside donor funding and does not require continued

support of the founding organization, it may prove to be more cost-effective,

sustainable and easily replicated than many of the other larger organizations. Overall,

the VSLA model appears to be both successful and sustainable, and offers potential

teaching benefits for other microfinance programs in developing countries.

140

REFERENCES

Allen, H.and M. Staehle (2007). Village Savings and Loan Associatons (VSLAs):

Programme Guide and Field Operations Manual. CARE International. Allen, H. and P. Hobane (2004). Impact Evaluation of Kupfuma Ishungu. Arusha,

Tanzania. CARE International.

Anderson, S. and J. Baland (2002). The Economics of Roscas and Intrahousehold Resource Allocation. The Quarterly Journal of Economics 117: 963-995.

Anyango, E. (2005). CARE Malawi Central Region Livelihood Security Project

Impact Assessment Report on Village Savings and Loans Component (VS&L). Care International.

Anyango, E., E. Esipisu, L. Opoku, S. Johnson, M. Malkamaki, and C.Musoke

(2006). Village Savings and Loan Associations in Zanzibar. London: Department for International Development (DFID).

Barnes, C. (2001). Microfinance Program Clients and Impact: An Assessment of

Zambuko Trust, Zimbabwe. USAID – AIMS Paper. Washington D.C. Bauer, M., J. Chytilová, and J. Morduch (2008). Behavioral Foundations of

Microcredit: Experimental and Survey Evidence from Rural India. IES Working Paper 28. Charles University.

Besley, T., S. Coate, and G. Loury (1993). The Economics of Rotating Savings and

Credit Associations. The American Economic Review 83: 792-810. Blackden, C.M. and M. Rwebangira (2004). Tanzania Strategic Country Gender

Assessment. Poverty Reduction and Economic Management Network, Africa Region, World Bank.

Buckley, G. (1997). Microfinance in Africa: Is it Either the Problem or the Solution.

World Development 25: 1081-1093. CARE Tanzania (2003). The Jozani Chwaka Bay Conservation Project. Zanzibar,

Tanzania. Unpublished. –––––– (2006). Village Savings and Loans and Women’s Empowerment: Strategic

Impact Inquiry (SII). Dar es Salaam, Tanzania.

141

Chen, M. and D. Snodgrass (2001). Managing Resources, Activities and Risk in Urban India: The Impact of SEWA Bank. Washington, D.C.: AIMS, USAID.

Coleman, B. (1999). The Impact of Group Lending in Northeast Thailand. Journal of

Development Economics 60 (1): 105-141. Country Report: Tanzania. London: Economist Intelligence Unit. May 2008. Dagnelie, O. and P. LeMay-Boucher (2008). Rosca Participation in Benin: A

Commitment Issue. UFAE and IAE Working Paper. Daley-Harris, S. (2009). State of the Microcredit Summit Campaign Report 2009.

Washington D.C.: Microcredit Summit Campaign. Deshingkar, P., J. Farrington, L. Rao, S. Akter, P. Sharma, A. Freeman and J. Reddy

(2008). Livestock and Poverty Reduction in India: Findings from the ODI Livelihood Options Project. London, UK: Overseas Development Institute.

Dunn, E. and G. Arbunkle (2001). The Impacts of Microcredit: A Case Study from

Peru. USAID – AIMS Paper. Washington, D.C. Dupas, P. and J. Robinson (2009). Savings Constraints and Microenterprise

Development: Evidence from a Field Experiment in Kenya. Cambridge, MA: NBER.

Ellis, A., M. Blackden, J. Cutura, F. MacCulloch, and H. Seebans (2007). Gender and

Economic Growth in Tanzania: Creating Opportunities for Women. Washington, D.C.: World Bank.

Gallardo, J., K. Ouattara, B. Randhawa, and W.F. Steel (2005). Comparative Review

of Microfinance Regulatory Framework Issues in Benin, Ghana, and Tanzania. African Region Financial Sector Group. Washington, D.C.: World Bank.

Gugerty, M.K. (2007). You Cannot Save Along: Commitment in Rotating Savings

and Credit Associations in Kenya. Economic Development and Cultural Change 55: 251-282.

Hashemi, S.M., S.R. Schuler, and A.P. Riley (1996). Rural Credit Programs and

Women’s Empowerment in Bangladesh. World Development 24: 635-653. Hoddinott, J. and L. Haddad (1994). Does Female Income Share Influence Household

Expenditures? Evidence from Cote d'Ivoire. Oxford Bulletin of Economics and Statistics 57: 77-96.

142

Hossain, M. (1988). Credit for the Alleviation of Rural Poverty: The Grameen Bank in Bangladesh. Washington, D.C.: IFPRI, Research Report No. 65.

Hulme, D. (2000). Impact Assessment Methodologies for Microfinance: Theory,

Experience and Better Practice. World Development 28: 79-98. International Labor Organization (ILO) (2003). Tanzanian Women Entrepreneurs:

Going for Growth. Geneva, Switzerland. –––––– (2001). Securing Small Loans: The Transaction Costs of Taking Collateral.

Final Research Report by the Social Finance Program. Geneva, Switzerland. Ingle, C.R. (1972). From Village to State in Tanzania: The Politics of Rural

Development. Cornell University Press: London. Johnson, S., M. Malkamaki, and K. Wanjau (2005). Tackling the ‘Frontiers’ of

Microfinance in Kenya: The Role of Decentralized Services. Nairobi, Kenya: Decentralized Financial Services.

Kabeer, N. (2001). Conflicts Over Credit: Re-Evaluating the Empowerment Potential

of Loans to Women in Rural Bangladesh. World Development 29: 63-84. Karlan, D. (2001). Microfinance Impact Assessments: The Perils of using new

members as a control group. Journal of Microfinance 3(2): 75-85. Kashuliza, A.K., J.P. Hella, F.T. Magayane, and Z.S.K. Mvena (1998). The Role of

Informal and Semi-Formal Finance in Poverty Alleviation in Tanzania: Results of a Field Study in Two Regions. Dar es Salaam, Tanzania: Research on Poverty Alleviation (REPOA).

Khandker, S.R. (2005). Microfinance and Poverty: Evidence Using Panel Data from

Bangladesh. The World Bank Economic Review 19: 263-286. Littlefield, E., J. Morduch, and S. Hashemi (2003). Is Microfinance an Effective

Strategy to Reach the Millennium Development Goals? CGAP’s Focus Note Series 24.

Lubawa, C. (1985). Rural Development Strategies: The Case of Tanzania. Rep.

Michigan State University. Masanjala, W.H. and M.G. Tsoka (1997). Socio-Economic Impact Study of FINCA-

Malawi. University of Malawi, Center for Social Research. Mersland, R. and O. Eggen (2007). You Cannot Saving Alone: Financial and Social

Mobilization in Savings and Credit Groups. Norad.

143

MkNelly, B. and Christopher D. (1999). Impact of Credit with Education on Mothers and Their Young Children’s Nutrition: CRECER Credit with Education Program in Bolivia. Freedom from Hunger Research Paper No. 5. Davis, CA: Freedom from Hunger.

–––––– (1999). Impact of Credit with Education on Mothers and Their Young

Children’s Nutrition: Lower Pra Rural Bank Credit with Education Program in Ghana. Freedom from Hunger Research Paper No. 4. Davis, CA: Freedom from Hunger.

Mkoma, G. (2009). VSL Apex Organizations. Dar es Salaam, Tanzania. CARE

International. Unpublished. Morduch, J. (1999). The Microfinance Promise. Journal of Economic Literature 37:

1569-1614. Morella, E., V. Foster, and S.G. Banerjee (2008). Climbing the Ladder: The State of

Sanitation in Sub-Saharan Africa. Washington, D.C.: The World Bank, Africa Infrastructure Country Diagnostic (AICD).

Mosley, P. (1998). The Use of Control Groups in Impact Assessments for

Microfinance. Enterprise and Cooperative Development Department, International Labor Office, Geneva.

Muganda, A. (2004). Tanzania’s Economic Reforms – and Lessons Learned. The

International Bank for Reconstruction and Development/The World Bank. Washington, D.C.

Mutesasira, L. (1999). Use and Impact of Savings Services among the Poor in

Tanzania. Nairobi, Kenya: Microsave. Mwenda, K.K. and Gerry N.K. (2004). Towards Best Practices for Micro Finance

Institutional Engagement in African Rural Areas: Selected Cases and Agenda for Action. International Journal of Social Economics 31: 143-158.

NBS (National Bureau of Statistics)/United Republic of Tanzania (2005).

Demographic and Health Survey (DHS). Dar es Salaam, Tanzania. –––––– (2001). Integrated Labour Force Report. Dar es Salaam, Tanzania. Neponen, H. (2003). ASA-GV Microfinance Impact Report. Trihcirappalli, India: The

Activists for Social Alternatives (ASA). Pitt, M.M., S.R. Khandker, O.H. Chowdury, and D.L. Millimet (2003). Credit

Programs for the Poor and the Health Status of Children in Rural Bangladesh. International Economic Review 44: 87-118.

144

Pitt, M.M. and S.R. Khandker (1998). The Impact of Group-Based Credit Programs

on Poor Households in Bangladesh: Does the Gender of Participants Matter?” The Journal of Political Economy 106 (Oct.): 958-996.

Pronyk, P.M., J.R. Hargreaves, and J. Morduch (2007). Microfinance Programs and

Better Health: Prospects for Sub-Saharan Africa. JAMA 16: 1925-1927. Ssendi, L. and A.R. Anderson (2009). Tanzanian Micro Enterprises and Micro

Finance: The Role and Impact for Poor Rural Women. Journal of Entrepreneurship 18: 1-19.

Strauss, J. and K. Beegle (1996). Intrahousehold Allocations: A Review of Theories,

Empirical Evidence and Policy Issues. MSU International Development Working Paper No. 62. Michigan State University.

Terry, W. (2006). The Impact of Micro-finance on Women Micro-entrepreneurs in

Temeke District, Dar-es-Salaam, Tanzania. MA thesis, Ohio University. The United Republic of Tanzania National Website. Country Profile.

http://www.tanzania.go.tz/. Todd, H. (2000). Poverty Reduced Through Microfinance: The Impact of ASHI in the

Philippines. Washington, D.C.: AIMS. Training Guide for the Formation of Savings and Loan Associations (SLA). (2004).

CARE Uganda. Udry, C. (1995). Risk and Saving in Northern Nigeria. American Economic Review

85 (5): 1287-1300. World Bank (2009). Tanzania Country Brief. Washington, D.C. World Development Report 2009: Reshaping Economic Geography. Washington,

D.C.: World Bank. Zaman, H. (2000). Assessing the Poverty and Vulnerability Impact of Micro-Credit in

Bangladesh: A Case Study of BRAC. Washington, D.C.: World Bank. Zeller, M. and M. Sharma (1998). Rural Finance and Poverty Alleviation.

Washington, D.C.: International Food Policy Research Institute (IFPFI).

145

APPENDIX A Summary of Impact Studies Author, Study Year Program Evaluated Methods to

control for selection bias

Key Results

Anyango, “CARE Malawi Central Region Livelihood Security Project Impact Assessment Report on Village Savings and Loans Component (VSL)”

2005 VSLA Program, Malawi None Improvement in livelihood of members; decrease in poverty; increase in number and magnitude of economic activities

Anyango et al., “Village Savings and Loan Associations in Zanzibar”

2006 VSLA Program, Zanzibar, Tanzania

None VSLAs have performed well in terms of growth and sustainability; self-reported improvement in standard of living and housing, and increase in income

Aportela, “Effects of Financial Access on Savings by Low-Income People”

1999 Banco de México, Mexico City, Mexico

Quasi-experimental design

Expansion of savings program increased average savings rate, with the poorest households experiencing the greatest increase

Barnes, “Microfinance Program Clients and Impact: An Assessment of Zambuko Trust, Zimbabwe”

2001 Zambuko Trust, Zimbabwe AIMS combination of quantitative and qualitative methods; Control group

Increase in income; increase in number of years of schooling for boys aged 6-16; improvement in both quantity and quality of food consumed; increase in durable assets

CARE Tanzania, “Village Savings and Loans and Women’s Empowerment Strategic Impact Inquiry (SII)”

2006 VSLA Program, Tanzania Control group; combination of quantitative and qualitative methods

Increase in number of IGAs; greater food security and health; increased education expenditures; increase in self-confidence and role in decision-making process

146

Chen and Snodgrass, “Managing Resources, Activities and Risk in Urban India: The Impact of SEWA Bank”

2001 SEWA Bank, India AIMS combination of quantitative and qualitative methods; Control group

Greater increase in income of borrowers, but both savers and borrowers experienced increase relative to non-participants

Coleman, “The Impact of Group Lending in Northeast Thailand”

1999 The Rural Friends Association and the Foundation for Integrated Agricultural Management, Thailand

Quasi-experimental design

Little to no impact on physical assets, savings, sales, school expenditures or health expenditures; impacts are vastly over-estimated when using more naïve controls for self-selection bias

Dunn and Arbunkle, “The Impacts of Microcredit: A Case Study from Peru”

2001 Mibanco Microfinance Program, Peru

AIMS combination of quantitative and qualitative methods; Control group

Decrease in rate of poverty; increase in income; increase in employment

Dupas and Robinson, “Savings Constraints and Microenterprise Development: Evidence from a Field Experiment in Kenya”

2009 Micro-entrepreneurs in Nairobi, Kenya

Quasi-experimental design

Access to a formal savings account has substantial positive impacts on women’s productive investment levels and expenditures, and also makes women less vulnerable to shocks from illness

Hashemi, Schuler and Riley, “Rural Credit Programs and Women’s Empowerment in Bangladesh”

1996 Grameen Bank and Bangladesh Rural Advancement Committee (BRAC)

Statistically control for differences in demographic characteristics; combination of sample survey and case study data

Both programs increase likelihood of a female client being empowered by 16 percent. Even women who do not participate in the program are more than twice as likely to be empowered simply by living in program villages

Hossain, “Credit for the Alleviation of Rural Poverty: The Grameen Bank in Bangladesh”

1988 Grammen Bank, Bangladesh

Control group of both eligible non-participants in Grameen villages and target non-participants in comparison village

Increase in household income; increase in investment in housing quality; improvement in nutrition and health statuses

147

Khandker, “Microfinance and Poverty: Evidence Using Panel Data from Bangladesh”

2005 Grameen Bank and Bangladesh Rural Advancement Committee (BRAC)

Household-level fixed-effects model with panel data

Increase in household expenditures, particularly for female clients; decrease in rate of poverty among participants as well as non-participants

Masanjala and Tsoka, “Socio-Economic Impact Study of FINCA-Malawi”

1997 FINCA - Malawi None Little impact on living standards and expenditure patterns

MkNelly and Dunford, “Impact of Credit with Education on Mothers and Their Young Children’s Nutrition: CRECER Credit with Education Program in Bolivia”

1999 Lower Pra Rural Bank Credit with Education Program, Ghana

Quasi-experimental design; Control group

Increase in income and diversification of income sources; improvement in household food security; improvement in nutritional outcomes (height-for-age and weight-for-age); improvement in women’s self-confidence and self-perception

MkNelly and Dunford, “Impact of Credit with Education on Mothers and Their Young Children’s Nutrition: CRECER Credit with Education Program in Bolivia”

1999 CRECER Credit with Education Program, Bolivia

Quasi-experimental design; Control group

Increase in income; little improvement in household food security; little impact on nutritional outcomes

Neponen, “ASA-GC Microfinance Impact Report”

2003 Activists for Social Alternatives (ASA), Trihcirappalli, India

Control group of new members

Children of clients are more likely to go to school and to stay in school longer; higher quality of housing

Pitt and Khandker, “The Impact of Group-Based Credit Programs on Poor Households in Bangladesh: Does the Gender of Participants Matter?”

1998 Grameen Bank, Bangladesh Rural Advancement Committee (BRAC)

Instrumental Variable (IV) model with village-level fixed effects

Increase in household expenditures, particularly for female clients; increase in the probability of girls’ school enrollment; positive impact on children’s health

Pitt et al., “Credit Programs for the Poor and the Health Status of Children in Rural Bangladesh”

2003 Grameen Bank, Bangladesh Rural Advancement Committee (BRAC)

Instrumental Variable (IV) model with village-level fixed effects

Positive impact on children’s health (as measured by height and arm circumference) for female borrowers only

Terry, “The Impact of Micro-finance on Women Micro-entrepreneurs in Temeke District, Dar-es-Salaam, Tanzania”

2006 FINCA - Tanzania None Improvement in social status and self-esteem, and an increase in confidence

148

Todd, “Poverty Reduced Through Microfinance: The Impact of ASHI in the Philippines”

2000 ASHI, Philippines Ethnographic approach – spend two years following a total of 64 households (40 borrowers and 24 comparison households)

Decrease in rate of poverty among borrowers; improvement in educational attainment for children of borrowers; improvement in quality of housing

Zaman, “Assessing the Poverty and Vulnerability Impact of Micro-credit in Bangladesh”

2000 Bangladesh Rural Advancement Committee (BRAC)

Control group; Heckman two-step procedure

Microfinance reduces vulnerability by smoothing consumption, building assets, providing emergency assistance during natural disasters, and empowering females

149

APPENDIX B Individual Questionnaire

Statement to be read before the interview begins: The information provided during this interview will be treated as highly confidential and is collected for research purposes only. Participation in this study will not affect one’s membership or role in the VSLA program. The purpose of this study is simply to gain a better understanding of the impacts of the program, so that its efforts may be improved so as to better serve its members. Therefore, we ask you to feel at ease and to provide frank and honest answers without fearing any persecution or disclosure. Researchers are only interested in analysis of collective feed back and not individual respondent information. Section 1: Background Information

1. Date of Interview________________________________ 2. Village________________________________ 3. Name of VSL Group________________________________

Section 2: Demographic Information 4. Gender of client

1. Male 2. Female

5. Age of client_________ 6. Relation to HHH

1. Household head 4. Parent of HHH 2. Spouse 5. Other relative 3. Son/daughter 6. No relation

7. Religion 1. Muslim 2. Christian 3. Other

8. Marital status 1. Married 4. Separated 2. Widowed 5. Single 3. Divorced

9. If married, is your husband polygamous? 1. Yes 2. No

10. What is the highest level of schooling that you have reached? 1. No education 4. Completed secondary (Advanced level) 2. Primary 5. Higher 3. Some Secondary (Ordinary level)

11. How many children have you had? ________

150

12. Provide following details for each member of the household

No

Sex 1=male 2=female Age

Marital Status 1 = married 2 = widowed 3 = divorced 4 = separated 5 = single

Relation to HHH 1 = household head 2 = spouse 3 = son/daughter 4 = parent of HHH 5 = other relative 6 = no relation

Main Occupation 1 = younger than school age 2 = student 3 = self-employed farming 4 = employed (no agric.) 5 = agric. laborer 6 = own business 7 = unemployed

Able to Read and Write Swahili 1 = yes 2 = no

Highest level of education 1 = no education 2 = Primary 3 = Some Secondary 4 = Completed Secondary 5 = Higher

Is he/she attending school now? 1 = yes 2 = no

1 2 3 4 5 6 7 8 9

10

151

13. How much did your household spend on education expenses (fees, uniforms, books, or other materials) during the last 12 months?

1. Yes 2. No

14. Do you pay for these educational expenses using payout or loans from the VSLA?

1. Yes 2. No

15. Does your village have a school? 1. Yes 2. No

16. Does your village have a paved road? 1. Yes 2. No

17. How far is it to the closest market in kilometers? ________ Section 3: Client Information

18. Member of VSL group for how long 1. Less than a year 2. 1-2 years 3. 2-5 years 4. More than 5 years

19. How many cycles of the VSL have you completed? _________ 20. How many shares do you currently have in your VSL group? _______

3.1 Savings 21. Before you joined the VSLA did you have any savings?

1. Yes 2. No

22. If yes, where did you put your savings? 1. In house 4. ROSCA 2. Bank account 5. SACCO 3. Credit union 6. Other

23. Do you continue to save in any other form? 1. In house 5. SACCO 2. Bank account 6. Other 3. Credit union 7. Do not save in other form 4. ROSCA

24. Amount of last payout? __________________ 25. Please rank your three most important uses of the payout. If business or

productive investment, please specify 1. Food 7. Medical expenses/health 2. Paid off debts 8. Productive investment 3. School fees 9. Household asset 4. Family celebration/ceremony 10. Gave to spouse 5. House project/improvements 11. Lending to another 6. Savings 12. Other

152

a. Primary use of payout If #7, type of productive investment

b. Secondary use of payout If #7, type of productive investment

c. Tertiary use of payout If #7, type of productive investment

26. Who made the decision?

1. Husband 2. Wife 3. Both 4. Other

3.2 Loans 27. Did you have access to loans before joining the VSLA?

1. Yes 2. No

28. If yes, did you ever take out a loan from a different organization? 1. Yes 2. No

29. If yes, how many loans? _________ 30. Have you ever taken a loan from VSLA?

1. Yes 2. No

31. If yes, how many loans? _________ 32. Did you take out a loan in the previous savings cycle?

1. Yes 2. No

33. If yes, how many loans did you take during the previous savings cycle? _____

34. What was the value of each of the loans during the previous savings cycle?

a. Value of First Loan _______________________________ b. Value of Second Loan_______________________________ c. Value of Third Loan_______________________________

35. Please rank your three most important uses of the loan(s). If business or productive investment, please specify

1. Food/household expenses 6. Medical fees/health 2. Repaying debts/borrowing for other 7. Business/productive investment 3. School fees 8. Household assets 4. Family celebration/ceremony 9. Emergency 5. House improvements 10. Other

153

a. Primary Use of Loan If #7, type of productive investment

b. Secondary Use of Loan If #7, type of productive investment

c. Tertiary Use of Loan If #7, type of productive investment

36. Who made the decision?

1. Husband 2. Wife 3. Both 4. Other

37. Are you currently engaged in any IGA? 1. Yes 2. No

38. In how many IGA are you currently engaged in? ____________ 39. What type of IGA are you currently engaged in? (circle as many as

necessary) 1. Agriculture (including livestock-keeping, poultry-farming) 2. Business (sales and trade) 3. Fishing 4. Seaweed Farming 5. Teaching 6. Tourist Industry 7. Transport Industry 8. Carpentry, masonry 9. Tailoring 10. Other, please specify_______________________________

40. How many people in the household are engaged in work that generates income? ____________

Section 4: Impact on Welfare Household Assets

41. How many of the following does your household own? #

Type of Asset

Quantity

Were you a member of the VSL when you acquired the asset?

1= yes 2 = no

1 Livestock 1.1 Cows 1.2 Sheep 1.3 Goats 1.4 Chicken/Duck

154

42. How much did you spend on household assets, including household goods,

equipment, and means of transport, in 2009? ____________ 43. How many acres of land does your family own? ____________ 44. How would you rank your household’s wealth within the community?

1. Richest in the community 2. Among the richest in the community 3. Richer than most households in the community 4. Among the poorest households in the community 5. The poorest in the community

Housing 45. To whom does the house belong?

1. Ours 3. Rented 2. Shared 4. Other

46. Does the house have electricity? 1. Yes 2. No

2 Transportation 2.1 Car/truck 2.2 Motorcycle 2.3 Bicycle 2.6 Cart 3 Electronics

3.1 Radio 3.2 Television 3.3 Cell phone 3.4 Fan 4 Agricultural Material

4.1 Tractor 4.2 Hoe 4.3 Plough 4.4 Irrigation pump 5 Other Goods

5.1 Mosquito Net 5.2 Lantern 5.3 Sewing machine 5.4 Refrigerator 5.5 Metal cooking pots

155

47. What material are the walls in the house? 1. Grass 5. Cement bricks 2. Mud and Pole 6. Stones 3. Sun-dried (unburnt) bricks 7. Other 4. Baked (burnt) bricks

48. What material is the roof made from? 1. Thatch – grass/leaves/mud 4. Plastic Sheets 2. Corrugated iron 5. Other 3. Asbestos/tiles/concrete

49. What material is the flood made of? 1. Earth, soil 3. Tiles 2. Cement 4. Other

50. How many rooms for sleeping? ____________ 51. What is your source of water?

1. Piped supply 4. Spring, river/stream, pond/lake 2. Borehole/covered well 5. Other 3. Open well

52. What type of sanitation does the house use? 1. Bush 3. Improved pit latrine 2. Traditional pit toilet 4. Flush Toilet

53. Source of cooking fuel 1. Fuel Wood 4. Electricity 2. Charcoal 5. Bottled Gas 3. Paraffin 6. Other

54. Has your household made any improvements in the past 12 months? 1. Yes 2. No

55. Where these improvements paid for by payout or loans from the VSLA? 1. Yes 2. No

Household Diet

56. Has household diet improved since joining the VSLA? 1. Improved 2. Stayed the same 3. Worsened 4. I don’t know

57. Usual number of meals per day? ___________________ 58. Frequency of problem with satisfying food needs in past year?

1. Never 2. Sometimes 3. Often 4. Always

59. Number of days consumed meat in past week? ____________________ 60. Number of days consumed fish in past week? _____________________

156

Health Care 61. Frequency of problem with accessing medical services and medication in

past year? 1. Never 2. Sometimes 3. Often 4. Always

62. Are all of your children immunized? 1. Yes 2. No

63. Do your children sleep under mosquito nets? 1. Yes 2. No

64. Has the health of members of the household changed since joining the VSLA?

1. Improved 2. Stayed the same 3. Worsened 4. I don’t know

65. How much did your household spend on healthcare expenses in 2009? ___________________

Section 5: Social Capital

66. Has your status in the community changed since joining VSLA? 1. Improved 2. Stayed the same 3. Worsened 4. I don’t know

67. Has your status in your family changed since joining VSLA? 1. Improved 2. Stayed the same 3. Worsened 4. I don’t know

68. Has your self-confidence changed since joining VSLA? 1. Improved 2. Stayed the same 3. Worsened 4. I don’t know

69. Are you a member of any community-based organizations, associations, networks or political parties?

1. Yes 2. No

70. If yes, are you a board member or do you hold a leadership position? 1. Yes 2. No

157

71. Did you vote in the last parliamentary election? 1. Yes 2. No

72. In the last 12 months, have you expressed your opinion in a public meeting (other than a VSL regular meeting)?

1. Yes 2. No

158

APPENDIX C Focus Group Discussion Format

Verbal Consent to Participate in the Focus Group: You have been asked to participate in a focus group. The purpose of this study is to gain a better understanding of the impacts of the VSL program, so that its efforts may be improved so as to better serve its members. You can choose whether or not to participate in the focus group and may stop at any time. Although the focus group will be tape recorded, your responses will remain anonymous and no names will be mentioned in the report. There are not right or wrong answers to these questions. We want to hear many different viewpoints and would like to hear from everyone. Participation in this study will not affect one’s membership or role in the VSLA program. Therefore, we ask you to feel at ease and to provide frank and honest answers without fearing any persecution or disclosure.

1. Tell me a little about your group and how it works 2. How long has the group been in existence? 3. What are some of the challenges and limitations your group faces? 4. Tell me about your life before you joined the group and how has that changed

since you became a member of the group? 5. In what ways has your behavior changed since you joined the group? 6. What role do you play in the decision making process of your household? Has

it changed since you joined the group? 7. What do you believe the benefits are to belonging to a VSLA group? What are

your reasons for joining? 8. Have there been any negative consequences of joining the VSLA group? If so,

what are they? 9. How does the community treat VSL members? Do they treat you differently

than before you were members? 10. Have you seen an impact of the VSL on the community as a whole? 11. Do you believe that the training has been beneficial? Is the apex organization

helpful? Is there any difference between the services that CARE provided versus those that the Apex organization now provides?

12. Is there anything else you would like to say about the VSL program?

159

APPENDIX D Statistical Tables

Table 1: Basic Characteristics of Respondents

Control Variable Treatment

Group Control Group

Test Statistic

N 120 50 Gender (% Female) 67.5 72.0 0.5771 Age 37.95 33.64 2.1759*** Age at time of joining 33.19 33.64 0.2341 Relation to HHH (%) Household head (HHH) 42.5 32 1.2765 Spouse 47.5 50 0.2972 Child 9.2 18 1.6288 Other relation 0.1 0 0.6474 Religion (%) Muslim 96.67 100 0.0 Christian 2.5 0 1.1280 Other 0.83 0 0.6474 Marital status (%) Married 75.00 70.0 0.6733 Widowed 8.33 8.0 0.0720 Divorced 5.0 8.0 0.7575 Separated 1.67 0 0.9183 Single 10.0 14.0 0.7542 Educational attainment (%) No education 13.33 20.0 1.1004 Primary 47.50 20.0 3.3433*** Ordinary level 15.83 32.0 2.3753*** Advanced level 23.33 28.0 0.6428 Number of children 3.725 2.62 2.7184*** Number of children at time of joining 3.075 2.62 1.1403 Average household size 5.033 4.92 0.316 Savings prior to joining VSLA? (%) 47.9 36.0 1.4211 Access to loans prior to joining? (%) 7.62 8.0 0.0827

*** p<0.01, ** p<0.05, * p<0.1

160

Table 2: Basic Characteristics of Respondents with Treatment Group

Divided by Median Years in VSLA

Control Variable Means/Percentage Test Statistic

Older Recent New

Recent-Older

Recent- New

Older-New

n 63 57 50 Gender (% Female) 0.714 0.632 0.72 0.966 0.9727 0.067 Age 41.032 34.544 33.64 3.195*** 0.396 3.367*** Age at time of joining 34.365 31.895 33.64 1.2355 0.7732 0.332 Relation to HHH (%) Household head (HHH) 38.1 47.37 32 1.0262 1.6178 0.673 Spouse 50.79 43.86 50 0.7596 0.6352 0.0838 Child 11.11 7.02 18 0.7761 1.7349* 1.0433 Other relation 0 1.75 0 1.0557 0.941 0 Religion (%) Muslim 96.83 96.49 100 0.1018 1.3371 1.2712 Christian 3.28 1.75 0 0.4976 0.941 1.2712 Other 0 1.75 0 1.0557 0.941 0 Marital status (%) Married 74.6 75.44 70 0.1055 0.6315 0.5447 Widowed 9.52 7.02 8 0.4961 0.1928 0.2833 Divorced 6.35 3.51 8 0.7129 1.0075 0.3398 Separated 0 3.51 0 1.4993 1.3371 0 Single 9.52 10.53 14 0.1828 0.5487 0.7407 Educational attainment (%) No education 17.46 8.77 20 1.3982 1.6691 0.3447 Primary 39.68 56.14 20 1.803* 3.820*** 2.248** Ordinary level 17.46 14.04 32 0.513 2.223** 1.800* Advanced level 25.4 21.05 28 0.5619 0.836 0.3112 Number of children 4.111 3.298 2.62 1.790* 1.487 3.358*** Number of children at time of joining 3.381 2.737 2.62 1.4468 0.2638 1.7074* Savings prior to joining VSLA? (%) 46.03 50.0 36.0 0.4325 1.4518 1.0746 Access to loans prior to joining? (%) 3.17 12.73 8.0 1.9502** 0.7900 1.1362

*** p<0.01, ** p<0.05, * p<0.1

161

Table 3: Housing Characteristics of Treatment and Control Group

Treatment Group

Control Group

Test Statistic

n 120 50 Tenure (%) Owned by household 85.8 60.0 3.6745*** Shared 8.3 34.0 4.2173*** Rented 0.8 0 0.6436 Other 5.1 6.0 0.0354 Electricity (%) 28.3 18.0 1.7454* Source of Drinking Water (%) Piped supply 74.0 76.0 0.1948 Well 25.0 24.0 0.1948 Sanitation (%) Bush 13.0 4.0 1.6975* Traditional pit latrine 4.0 54.0 7.5419*** Improved pit latrine 78.0 22.0 6.8484*** Flush toilet 5.0 2.0 3.0316*** Source of Cooking Fuel (%) Fuel Wood 98.0 98.0 0.2034 Charcoal 2.0 2.0 0.2034 Flooring Material (%) Earth, soil 22.0 34.0 0.7821 Cement 76.0 66.0 0.5562 Tiles 2.0 0 0.9261 Wall Material (%) Grass 3.0 4.0 0.5083 Mud and Pole 12.0 26.0 2.2809** Sun-dried bricks 3.0 10.0 2.0753** Baked bricks 0 4.0 2.1856** Stones 59.0 2.0 6.8768*** Cement bricks 23.0 52.0 3.5874*** Other 0 2.0 1.5408 Roof Material (%) Thatch 22.0 48.0 3.3656*** Corrugated iron 76.0 52.0 3.1113*** Asbestos, tiles 2.0 0 0.9261 Avg. number of rooms for sleeping 2.566 2.56 0.1230 Improvements in last 12 months? (%) Yes 67.0 16.0 6.0844***

*** p<0.01, ** p<0.05, * p<0.1

162

Table 4: Household Assets

Treatment Group

Control Group

Test Statistic

n 120 50 Livestock

Number of cows 1.966387 0.86 1.7678* Goats 0.7142857 0.46 0.8119 Chicken/Ducks 7.798319 6.54 0.8835

Transportation Motorcycles 0.0840336 0.04 0.9336 Bicycles 0.7478992 0.8 0.4248

Electronics Radio 0.8833333 0.76 1.0153 Television 0.1092437 0.04 1.4451 Cell Phone 1.033333 0.78 1.708* Fan 0.0583333 0.04 0.3706

Other household items Hoe 0.9916667 1.64 2.6975*** Mosquito net 2.825 2.54 1.235 Lantern 1.441667 1.3 0.5374 Sewing machine 0.302521 0.28 0.2462 Refrigerator 0.0840336 0.02 1.4092 Metal cooking pots 7.525 6.14 1.6302

2009 Asset Expenditure (Tsh) 138078.2 31288.89 4.7292*** 2009 Education Expenditure (Tsh) 105,579.9 33,808.51 1.8751*

*** p<0.01, ** p<0.05, * p<0.1

Table 5: Household Food Security Treatment

Group Control Group

Test Statistic

Average number of meals per day 2.542 2.460 0.9251 Average number of days consumed meat in last week 0.5042

0.1632

2.6352***

Average number of days consumed fish in last week 4.6050 1.2041

10.381***

Frequency of problems satisfying food needs in past year (%)

Never 33.0 6.0 3.6760*** Sometimes 66.0 88.0 3.5754*** Often 2.0 6.0 1.5126 Always 0 0

*** p<0.01, ** p<0.05, * p<0.1

163

Table 6: Health Status of Household Treatment

Group Control Group

Test Statistic

Frequency of problems accessing medical services in past year (%)

Never 23.0 4.0 3.0327*** Sometimes 69.0 96.0 3.8155*** Often 8.0 0 1.9986** Always 0 0

Are all of your children immunized? (%) Yes 95.7 95.5 0.0543

Do all of your children sleep under mosquito nets? (%)

Yes 97.4 90.9 1.7825* 2009 Health Expenditures (Tsh) 69,696.4 36,947.92 2.1722***

*** p<0.01, ** p<0.05, * p<0.1

Table 7: Income Generating Activities (IGAs)

Treatment Group

Control Group

Test Statistic

Number of IGAs 1.9107 1.3877 4.5775*** Type of IGA (%)

Agriculture 75.8 68.0 1.0548 Business 46.7 58.0 1.3466 Fishing 9.2 4.0 1.1550 Seaweed farming 25.0 0.0 3.8960*** Tourism 0.83 0.0 0.6474 Carpentry 0.83 0.0 0.6474 Tailoring 5.8 2.0 1.0754

*** p<0.01, ** p<0.05, * p<0.1

Table 8: Social Status of Respondents

Treatment Group

Control Group

Test Statistic

Are you a member of any community-based organization, association, or political party? (%)

Yes 81.5 74.0 1.1000 If yes, do you hold a leadership position (%)

Yes 28.7 24.3 0.5144 Did you vote in the last parliamentary election? (%)

Yes 84.9 78.0 1.0818 In the last 12 months have you expressed your opinion in a public meeting? (%)

Yes 30.2 8.0 3.1063*** *** p<0.01, ** p<0.05, * p<0.1

164

Table 9: Specifics of VSLA Participation

Current

Members Drop-Outs

Test Statistic

n 100 20 Number of years in the VSLA program 5.06 3.25 3.1354*** Amount of last payout (Tsh) 277,125.9 234,473.3 0.9959 Primary uses of payout (%)

Food 51.0 60.0 0.7358 To pay debts 22.0 15.0 0.7037 School fees 48.0 25.0 1.8909** Family celebration/ceremony 22.0 25.0 0.2933 House improvement 29.0 30.0 0.0898 Savings 16.0 10.0 0.6860 Medical expenses 10.0 15.0 0.6568 Productive Investment 33.0 30.0 0.2615 Household Assets 6.0 5.0 0.1742 Gave to spouse 1.0 0 0.4491 Other 14.0 10.0 0.4804

Number of loans from VSLA 6.4845 3.375 3.4204*** Average value of loan (Tsh) 120,241.9 111,066.7 0.3072 Primary uses of loan (%)

Food/household expenses 47.0 45.0 0.1637 To pay debts 18.0 0.0 2.0580** School fees 35.0 25.0 0.8660 Family celebration/ceremony 18.0 20.0 0.2110 House improvement 22.0 20.0 0.1982 Medical expenses 15.0 20.0 0.5592 Productive Investment 54.0 30.0 1.9596** Household Assets 6.0 0.0 1.1239 Emergency 5.0 0.0 1.0215 Other 12.0 15.0 0.3703

*** p<0.01, ** p<0.05, * p<0.1

165

Table 10: Diet and Health Status Changes Since Joining the VSLA Program

(Current Members vs. Dropouts)

Current

Members Drop-

out Test

Statistic Has household diet improved since joining VSLA? (%)

Improved 75.0 47.4 2.4232*** Stayed the same 23.0 47.4 2.1961** Worsened 1.0 0.0 0.4377 I don’t know 1.0 5.3 1.3252

Has the health of members of the household improved since joining VSLA? (%)

Improved 80.8 57.9 2.1778** Stayed the same 18.2 36.8 1.8232* Worsened 0.0 5.3 2.2924** I don’t know 1.0 0 0.4400

*** p<0.01, ** p<0.05, * p<0.1

Table 11: Changes in Social Status Since Joining the VSLA Program (Current Members vs. Dropouts)

Current

Members Drop-outs

Test Statistic

Has your status in the community changed since joining? (%)

Improved 84.0 55.0 2.9152*** Stayed the same 15.0 35.0 2.1101** Worsened 0.0 0.0 0.0 I don’t know 1.0 5.0 1.2756

Has your status in your family changed since joining? (%)

Improved 85.0 50.0 3.5184*** Stayed the same 15.0 40.0 2.5930*** Worsened 0.0 5.0 2.2454** I don’t know 0.0 0.0 0.0

Has your self-confidence changed since joining? (%)

Improved 89.0 55.0 3.7245*** Stayed the same 11.0 40.0 3.2431*** Worsened 0.0 0.0 0.0 I don’t know 0.0 0.0 0.0

*** p<0.01, ** p<0.05, * p<0.1

166

Table 12: Number of IGAs

Column (1) clustered by VSLA group (n = 25), Column (2) – (4) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) VARIABLES Membership 0.368* 0.368* 0.285** 0.439* (0.205) (0.201) (0.124) (0.247) Gender 0.0439 0.0439 0.148 0.0356 (0.0859) (0.0768) (0.133) (0.0858) Membership*Gender 0.157 0.157 -0.296 (0.237) (0.226) (0.336) Membership + (Membership*Gender)

0.524*** (0.097)

0.524*** (0.076)

0.144 (0.147)

Dosage 0.0425 -0.0164 (0.0382) (0.0545) Dosage*Gender 0.0976** (0.0446) Dosage + (Dosage*Gender)

0.081** (0.034)

Age 0.0119 0.0119 0.0101 0.00895 (0.00754) (0.00695) (0.00652) (0.00663) Christian -0.854*** -0.854*** -0.804*** -0.687*** (0.254) (0.201) (0.184) (0.161) Other religion 1.695*** 1.695*** 1.628*** 1.366*** (0.300) (0.264) (0.231) (0.263) Married -0.371** -0.371** -0.363** -0.284** (0.157) (0.163) (0.147) (0.119) Widowed -0.450 -0.450 -0.432 -0.347 (0.284) (0.262) (0.258) (0.243) Divorced -0.332 -0.332** -0.357*** -0.256** (0.206) (0.133) (0.116) (0.104) Separated -1.379*** -1.379*** -1.263*** -1.066*** (0.207) (0.205) (0.161) (0.123) Primary 0.253 0.253 0.233 0.254 (0.179) (0.207) (0.201) (0.217) Ordinary level 0.379* 0.379 0.344 0.331 (0.197) (0.224) (0.217) (0.226) Advanced level 0.376* 0.376 0.340 0.343 (0.214) (0.248) (0.226) (0.239) Children 0.0433 0.0433 0.0380 0.0272 (0.0344) (0.0344) (0.0349) (0.0316) Prior savings -0.0210 -0.0210 -0.0265 -0.0248 (0.115) (0.0784) (0.0775) (0.0787) Prior access -0.190 -0.190** -0.135** -0.134* (0.122) (0.0806) (0.0625) (0.0642) Constant 0.913*** 0.913*** 0.929*** 1.005*** (0.304) (0.266) (0.286) (0.268) Observations 162 162 162 162 R-squared 0.211 0.211 0.222 0.239

167

Table 13: 2009 Asset Expenditures

(1) (2) (3) (4) VARIABLES Membership 116,227*** 116,227** 91,120*** 128,624* (40,516) (40,088) (29,146) (68,072) Gender -21,154* -21,154* -40,477 -21,434* (11,355) (10,083) (32,150) (10,444) Membership*Gender -29,064 -29,064 -56,717 (48,964) (45,971) (83,807) Membership + (Membership*Gender)

87,162*** (26,662)

87,162*** (29,355)

71,906* (36,643)

Dosage 874.6 -2,433 (4,598) (10,407) Dosage*Gender 5,370 (10,434) Dosage + (Dosage*Gender)

2,937 (2,766)

Age 38.36 38.36 21.59 -8.994 (1,339) (1,458) (1,629) (1,592) Christian 164,439*** 164,439** 168,946** 168,311** (56,687) (61,307) (60,456) (63,181) Other religion -4,321 -4,321 5,458 -22,753 (60,898) (45,399) (40,736) (56,243) Married -6,384 -6,384 -4,311 -2,531 (27,744) (24,505) (24,546) (24,519) Widowed -21,550 -21,550 -20,248 -17,594 (45,576) (42,200) (41,475) (43,692) Divorced -37,531 -37,531 -32,270 -34,308 (40,208) (39,158) (37,071) (39,185) Separated -43,495 -43,495 -38,017 -31,077 (50,487) (42,091) (41,301) (43,826) Primary 32,998 32,998 36,996 34,199 (41,641) (45,701) (51,050) (47,721) Ordinary level 80,871* 80,871 85,248 79,612 (46,630) (53,025) (54,531) (53,606) Advanced level 37,293 37,293 41,995 36,682 (34,769) (35,817) (38,095) (35,993) Children 5,138 5,138 5,084 4,620 (5,205) (3,976) (3,912) (4,072) Prior savings 5,623 5,623 5,832 5,839 (25,617) (21,591) (21,911) (21,876) Prior access -64,847 -64,847** -61,002** -66,440** (44,717) (22,896) (20,700) (25,084) Constant -399.5 -399.5 8,083 -99.41 (55,487) (64,711) (58,287) (68,648) Observations 133 133 133 133 R-squared 0.274 0.274 0.273 0.276

Column (1) clustered by VSLA group (n = 25), Column (2) – (4) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

168

Table 14: 2009 Education Expenditures

(1) (2) (3) (4) (5) (6) VARIABLES Membership -19,370 -19,370 4,324 29,988 29,869** 29,869 (41,026) (35,584) (47,473) (90,989) (14,311) (21,520) Gender -42,082 -42,082 -8,637 -42,720 -23,883 -23,883 (32,955) (31,322) (29,611) (32,236) (36,352) (33,678) Membership*Gender 46,294 46,294 -61,422 28,038 28,038 (54,582) (47,439) (96,361) (64,117) (60,273) Membership + (Membership*Gender)

26,924 (20,567)

26,924 (25,502)

-31,433 (44,793)

38,992** (16,831)

38,992* (20,941)

Dosage 2,143 -10,671 (10,892) (14,913) Dosage*Gender 22,338 (13,421) Dosage + (Dosage*Gender) 11,667

(10,408)

Age 2,425 2,425 2,298 2,188 1,855 1,855 (1,984) (2,022) (2,276) (2,191) (1,615) (1,595) Christian 563,220*

* 563,220

*** 562,971** 583,189**

(268,021) (78,083) (266,920) (268,520) Other religion 99,691 99,691 85,248 21,480 (59,551) (71,269) (65,555) (66,535) Married -41,579 -41,579 -40,719 -25,265 -22,225 -22,225 (28,338) (33,284) (30,386) (26,355) (21,669) (23,236) Widowed -27,356 -27,356 -25,416 -9,764 1,627 1,627 (69,021) (60,548) (73,323) (72,823) (71,690) (71,865) Divorced 4,128 4,128 -942.41 15,075 -19,775 -19,775 (46,198) (33,334) (44,338) (47,694) (49,527) (45,706) Separated -46,840 -46,840 -40,813 11.668 -63,565 -63,565 (54,058) (57,432) (72,915) (66,019) (62,980) (68,385) Primary 101,995* 101,995 96,543 104,720* (56,216) (70,020) (58,075) (55,992) Ordinary level 81,115 81,115 74,270 73,664 (48,328) (61,428) (48,868) (46,698) Advanced level 109,893* 109,893 103,258 105,591* (61,133) (70,750) (61,463) (62,112) Educ. Attainment 25,449 25,449 (18,322) (20,504) Children 24,320**

* 24,320*

** 24,082*** 22,396** 22,424*** 22,424***

(7,964) (7,746) (8,325) (8,520) (7,509) (7,268) Prior savings 11,096 11,096 11,127 10,483 -5,768 -5,768 (37,761) (23,418) (37,252) (37,067) (37,626) (28,679) Prior access -42,077 -42,077 -41,210 -42,134 -40,864 -40,864 (45,271) (52,732) (51,265) (53,098) (44,345) (50,401) Constant -129,506 -129,506 -144,809 -127,043 -117,584 -117,584 (95,836) (107,768

) (97,010) (104,053) (90,577) (108,285)

Observations 164 164 164 164 160 160 R-squared 0.294 0.294 0.293 0.303 0.128 0.128

Column (1) & (5) clustered by VSLA group (n = 25), column (2) – (4) & (6) clustered by village (n = 13) Column (5) & (6) run only on the sub-sample of Muslims

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.

169

Table 15: Number of Meals per Day

(1) (2) (3) (4) VARIABLES Membership 0.337** 0.337* 0.335** 0.506** (0.151) (0.160) (0.133) (0.195) Gender 0.348*** 0.348** 0.0972 0.355** (0.111) (0.136) (0.0885) (0.122) Membership*Gender -0.358** -0.358** -0.208 (0.149) (0.144) (0.162) Membership + (Membership*Gender)

-0.020 (0.106)

-0.020 (0.124)

0.298** (0.131)

Dosage -0.0547** -0.0361 (0.0234) (0.0326) Dosage*Gender -0.0314 (0.0266) Dosage + (Dosage*Gender)

-0.067** (0.023)

Age 0.00119 0.00119 0.00383 0.00385 (0.00413) (0.00397) (0.00412) (0.00405) Christian 0.545*** 0.545*** 0.500*** 0.444*** (0.134) (0.143) (0.135) (0.137) Other religion -0.00654 -0.00654 0.124 0.114 (0.186) (0.221) (0.207) (0.247) Married 0.151 0.151 0.152 0.113 (0.171) (0.201) (0.166) (0.186) Widowed 0.0416 0.0416 0.0293 -0.00678 (0.249) (0.254) (0.245) (0.259) Divorced -0.252 -0.252 -0.212 -0.281 (0.189) (0.187) (0.163) (0.164) Separated -0.277 -0.277 -0.430* -0.509** (0.206) (0.245) (0.211) (0.233) Primary 0.126 0.126 0.151 0.104 (0.131) (0.142) (0.136) (0.139) Ordinary level 0.0921 0.0921 0.139 0.107 (0.130) (0.167) (0.160) (0.152) Advanced level 0.244* 0.244 0.294* 0.249* (0.130) (0.142) (0.145) (0.134) Children -0.0209 -0.0209 -0.0159 -0.0131 (0.0172) (0.0181) (0.0183) (0.0191) Prior savings -0.107 -0.107 -0.108 -0.107 (0.0832) (0.0976) (0.0962) (0.0937) Prior access 0.192 0.192 0.132 0.117 (0.167) (0.224) (0.219) (0.227) Constant 2.036*** 2.036*** 2.082*** 1.958*** (0.226) (0.190) (0.217) (0.191) Observations 170 170 170 170 R-squared 0.133 0.133 0.153 0.175

Column (1) clustered by VSLA group (n = 25), Column (2) – (4) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

170

Table 16: Number of Times had Meat in Last 7 Days

(1) (2) (3) (4) VARIABLES Membership 0.337 0.337 0.311 0.205 (0.311) (0.206) (0.213) (0.286) Gender -0.209 -0.209 -0.245* -0.207 (0.272) (0.155) (0.117) (0.154) Membership*Gender -0.0494 -0.0494 0.205 (0.355) (0.229) (0.334) Membership + (Membership*Gender)

0.287* (0.144)

0.287** (0.102)

0.439* (0.244)

Dosage -0.00215 0.0292 (0.0463) (0.0510) Dosage*Gender -0.0541 (0.0521) Dosage + (Dosage*Gender)

-0.025 (0.055)

Age -0.0107 -0.0107 -0.0106 -0.0104 (0.00662) (0.00634) (0.00610) (0.00609) Christian 0.872 0.872*** 0.873*** 0.828*** (0.632) (0.189) (0.148) (0.139) Other religion -0.621* -0.621** -0.606*** -0.463 (0.349) (0.216) (0.200) (0.296) Married 0.0424 0.0424 0.0441 0.0101 (0.233) (0.229) (0.223) (0.225) Widowed 0.321 0.321 0.322 0.296 (0.343) (0.328) (0.319) (0.316) Divorced -0.169 -0.169 -0.163 -0.196 (0.226) (0.190) (0.190) (0.187) Separated -0.133 -0.133 -0.136 -0.233 (0.395) (0.342) (0.321) (0.346) Primary 0.0411 0.0411 0.0468 0.0286 (0.246) (0.241) (0.226) (0.242) Ordinary level -0.132 -0.132 -0.126 -0.124 (0.288) (0.295) (0.279) (0.292) Advanced level 0.0404 0.0404 0.0480 0.0458 (0.262) (0.267) (0.254) (0.268) Children 0.00129 0.00129 0.00150 0.00643 (0.0316) (0.0315) (0.0286) (0.0283) Prior savings 0.178 0.178 0.178 0.180 (0.141) (0.159) (0.159) (0.159) Prior access 0.0820 0.0820 0.0817 0.0866 (0.253) (0.155) (0.197) (0.194) Constant 0.580 0.580 0.593 0.578 (0.396) (0.384) (0.420) (0.366) Observations 168 168 168 168 R-squared 0.137 0.137 0.136 0.141

Column (1) clustered by VSLA group (n = 25), Column (2) – (4) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

171

Table 17: Number of Times had Fish in Last 7 Days

(1) (2) (3) (4) VARIABLES Membership 3.483*** 3.483*** 3.260*** 2.939*** (0.743) (0.771) (0.649) (0.932) Gender -0.762 -0.762 -0.622* -0.766 (0.765) (0.633) (0.293) (0.638) Membership*Gender 0.208 0.208 0.520 (0.814) (0.717) (0.811) Membership + (Membership*Gender)

3.691*** (0.341)

3.691*** (0.457)

3.459*** (0.677)

Dosage 0.0794 0.118 (0.100) (0.0928) Dosage*Gender -0.0671 (0.107) Dosage + (Dosage*Gender)

0.051 (0.127)

Age 0.00366 0.00366 2.44e-05 0.000450 (0.0146) (0.0150) (0.0149) (0.0149) Christian -0.771* -0.771** -0.666 -0.698 (0.446) (0.311) (0.392) (0.400) Other religion -1.879*** -1.879** -1.993*** -1.740** (0.576) (0.659) (0.619) (0.738) Married 0.332 0.332 0.360 0.333 (0.677) (0.795) (0.781) (0.802) Widowed -0.139 -0.139 -0.116 -0.136 (0.845) (1.084) (1.070) (1.079) Divorced -0.246 -0.246 -0.248 -0.251 (0.840) (1.059) (1.082) (1.083) Separated -0.290 -0.290 -0.0167 -0.112 (0.833) (0.946) (0.842) (0.889) Primary 0.480 0.480 0.488 0.502 (0.497) (0.522) (0.478) (0.518) Ordinary level 1.101 1.101 1.075 1.112 (0.690) (0.689) (0.650) (0.697) Advanced level 1.675** 1.675** 1.654** 1.694** (0.669) (0.735) (0.670) (0.747) Children -0.0997 -0.0997 -0.110 -0.104 (0.0806) (0.0847) (0.0852) (0.0892) Prior savings -0.208 -0.208 -0.209 -0.206 (0.250) (0.300) (0.299) (0.310) Prior access -0.584 -0.584 -0.475 -0.452 (0.373) (0.474) (0.472) (0.463) Constant 0.876 0.876 0.908 0.974 (0.867) (0.997) (1.012) (1.056) Observations 168 168 168 168 R-squared 0.491 0.491 0.495 0.496

Column (1) clustered by VSLA group (n = 25), Column (2) – (4) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

172

Table 18: 2009 Health Expenditures

(1) (2) (3) (4) VARIABLES Membership 27,988 27,988 27,250 45,139 (29,809) (29,629) (28,702) (70,215) Gender -23,474 -23,474** -28,367 -23,544** (15,530) (9,966) (21,666) (10,203) Membership*Gender -7,114 -7,114 -30,263 (33,707) (33,063) (75,434) Membership + (Membership*Gender)

20,875* (11,930)

20,875 (13,574)

14,876 (21,181)

Dosage -919.0 -3,692 (4,863) (9,712) Dosage*Gender 4,838 (9,725) Dosage + (Dosage*Gender)

1,146 (3,480)

Age 378.3 378.3 431.1 394.5 (674.4) (699.6) (696.5) (673.3) Christian 41,080 41,080 40,210 43,173 (66,426) (24,344) (26,743) (25,180) Other religion -50,816 -50,816 -47,856* -64,790 (33,131) (30,274) (23,131) (53,753) Married 23,843 23,843 23,739 26,164 (21,163) (19,985) (19,296) (22,381) Widowed 36,869 36,869 36,414 38,967 (32,061) (30,925) (30,497) (32,942) Divorced 20,423 20,423 21,029 22,429 (23,394) (19,151) (18,868) (20,390) Separated 43,040 43,040 40,354 47,820 (27,077) (25,704) (25,640) (29,004) Primary 40,977** 40,977* 41,823* 41,213* (19,336) (21,562) (21,524) (22,002) Ordinary level 19,974 19,974 21,211 18,993 (18,315) (18,716) (18,641) (18,852) Advanced level 11,067 11,067 12,403 9,882 (15,281) (16,136) (16,312) (16,539) Children -113.1 -113.1 -4.432 -427.8 (2,974) (2,883) (2,879) (3,074) Prior savings -13,640 -13,640 -13,698 -13,661 (17,093) (17,340) (17,507) (17,541) Prior access 65,777 65,777* 65,096* 62,962 (41,305) (36,710) (36,078) (39,259) Constant 1,435 1,435 1,996 679.0 (29,367) (28,693) (29,155) (29,179) Observations 160 160 160 160 R-squared 0.145 0.145 0.145 0.148

Column (1) clustered by VSLA group (n = 25), Column (2) – (4) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

173

Table 19: Do All of Your Children Sleep Under Mosquito Nets?

(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES Probit LPM Probit LPM Probit LPM Probit LPM Membership 0.778 0.118 0.778* 0.118 0.379 0.0481 0.550 0.0899 (0.551) (0.120) (0.459) (0.0894) (0.490) (0.0581) (0.614) (0.135)

Marginal effect 0.051 0.051 0.017 0.013 (0.058) (0.049) (0.032) (0.022) Gender 0.617 0.123 0.617 0.123 0.703** 0.0693 0.636 0.122 (0.460) (0.116) (0.414) (0.0948) (0.346) (0.0471) (0.439) (0.117)

Marginal effect 0.036 0.036 0.037 0.015 (0.032) (0.025) (0.025) (0.011) Membership*Gender 0.0722 -0.0709 0.0722 -0.0709 -0.941 -0.0527 (0.639) (0.125) (0.625) (0.108) (0.902) (0.140)

Marginal effect 0.003 0.003 -0.018 (0.028) (0.001) (0.023) Membership + (Membership*Gender)

0.851** (0.409)

0.043 (0.044)

0.851* (0.449)

0.043 (0.056)

-0.391 (0.675)

0.038 (0.057)

Marginal effect 0.059 0.059 -0.005 (0.048) (0.057) (0.008) Dosage 0.117 0.00364 0.0557 0.0060 (0.108) (0.0059) (0.111) (0.012)

Marginal effect 0.004 0.001 (0.004) (0.002) Dosage*Gender 0.381* -0.0039 (0.206) (0.013)

Marginal effect 0.006 (0.005) Dosage + (Dosage*Gender)

0.436** (0.188)

0.001 (0.006)

Marginal effect 0.007 (0.005) Age 0.0172 0.0013 0.0172 0.0013 0.0153 0.0012 0.0151 0.0012 (0.0141) (0.0015) (0.0130) (0.0015) (0.0126) (0.0015) (0.0123) (0.0015) Primary -0.247 -0.0020 -0.247 -0.0020 -0.296 0.0077 -0.252 -0.0022 (0.289) (0.0349) (0.370) (0.0425) (0.334) (0.0269) (0.393) (0.0347) Ordinary level -0.0891 0.0197 -0.0891 0.0197 -0.162 0.0288 -0.125 0.0186 (0.372) (0.0270) (0.402) (0.0321) (0.450) (0.0370) (0.409) (0.0269) Advanced level -0.0970 0.0030 -0.097 0.0030 -0.186 0.0122 -0.147 0.0024 (0.375) (0.0298) (0.421) (0.0353) (0.466) (0.0368) (0.458) (0.0301) Children 0.151 0.0093 0.151 0.0093 0.147 0.0092 0.137 0.0094 (0.117) (0.0083) (0.117) (0.0079) (0.125) (0.0082) (0.119) (0.0084) Prior savings -0.403 -0.0418 -0.403 -0.0418 -0.435 -0.0428 -0.485 -0.0420 (0.369) (0.0400) (0.317) (0.0384) (0.329) (0.0402) (0.324) (0.0405) Constant 0.202 0.75*** 0.202 0.75*** 0.291 0.78*** 0.358 0.75*** (0.644) (0.132) (0.482) (0.0647) (0.466) (0.101) (0.404) (0.134) Observations 160 160 160 160 160 160 160 160 Pseudo R2/R-squared 0.082 0.084 0.082 0.084 0.081 0.081 0.085 0.086 Columns (1) - (2) clustered by VSLA group (n = 25), columns (3) – (8) clustered by village (n = 13)

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

174

Table 20: Do You Own Your Home? (1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES Probit LPM Probit LPM Probit LPM Probit LPM

Membership 1.000** 0.304** 1.000*** 0.304** 0.852** 0.248** 0.594 0.206 (0.430) (0.147) (0.355) (0.106) (0.383) (0.0908) (0.593) (0.187)

Marginal effect 0.301** 0.301*** 0.252** 0.169 (.139) (0.104) (0.118) (0.181)

Gender 0.389 0.101 0.389* 0.101 0.389* 0.0791 0.396 0.101 (0.382) (0.143) (0.215) (0.0638) (0.229) (0.0652) (0.382) (0.144)

Marginal effect 0.107 0.107* 0.107 0.108 (0.112) (0.062) (0.068) (0.112)

Membership*Gender 0.00829 -0.0280 0.00829 -0.0280 0.533 0.0837 (0.495) (0.164) (0.332) (0.0993) (0.596) (0.185)

Marginal effect 0.002 0.002 0.135 (0.128) (0.086) (0.148)

Membership + (Membership*Gender)

1.008*** (0.265)

0.276*** (0.076)

1.008*** (0.272)

0.276*** (0.078)

1.217*** (0.383)

0.290*** (0.075)

Marginal effect 0.304*** 0.304*** 0.342*** (0.082) (0.085) (0.117) Dosage 0.033 0.00771 0.0945 0.0217

(0.063) (0.0142) (0.0892) (0.0215) Marginal effect 0.009 0.024

(0.017) (0.023) Dosage*Gender -0.117 -0.0240 (0.0830) (0.0193)

Marginal effect -0.030 (0.021) Dosage + (Dosage*Gender)

-0.023 (0.062)

-0.002 (0.013)

Marginal effect -0.006 (0.016) Age -0.00516 -0.00206 -0.00516 -0.00206 -0.00682 -0.00248 -0.00666 -0.0024

(0.0142) (0.00405) (0.0151) (0.0041) (0.0133) (0.0039) (0.0136) (0.004) Married -0.356 -0.0864 -0.356 -0.0864 -0.350 -0.0815 -0.396 -0.0942

(0.507) (0.140) (0.631) (0.176) (0.507) (0.137) (0.519) (0.141) Widowed -0.537 -0.111 -0.537 -0.111 -0.537 -0.104 -0.552 -0.115 (0.613) (0.160) (0.398) (0.0984) (0.619) (0.163) (0.613) (0.163) Divorced -0.883 -0.242 -0.883 -0.242 -0.900 -0.240 -0.939 -0.251 (0.674) (0.210) (0.540) (0.153) (0.669) (0.207) (0.686) (0.214) Separated -1.792 -0.509 -1.792 -0.509 -1.705 -0.479 -1.740 -0.488

(1.157) (0.436) (1.265) (0.452) (1.148) (0.432) (1.216) (0.453) Primary -0.560 -0.126 -0.560 -0.126 -0.553 -0.119 -0.583 -0.128

(0.369) (0.0834) (0.363) (0.0837) (0.340) (0.0792) (0.377) (0.086) Ordinary level -0.339 -0.0759 -0.339 -0.0759 -0.346 -0.0736 -0.344 -0.0727

(0.379) (0.0917) (0.273) (0.0607) (0.367) (0.0929) (0.379) (0.093) Advanced level -0.199 -0.0340 -0.199 -0.0340 -0.202 -0.0293 -0.189 -0.0305

(0.479) (0.106) (0.430) (0.0920) (0.444) (0.100) (0.485) (0.107) Children 0.0370 0.0102 0.0370 0.0102 0.0365 0.00976 0.0434 0.0110

(0.0508) (0.0144) (0.0573) (0.0161) (0.0506) (0.0146) (0.0505) (0.0145) Prior savings -0.0734 -0.0138 -0.0734 -0.0138 -0.0725 -0.0141 -0.0677 -0.0127

(0.200) (0.0497) (0.180) (0.0441) (0.199) (0.0490) (0.197) (0.0491) Prior access 0.865*** 0.148*** 0.865*** 0.148*** 0.896*** 0.161*** 0.916*** 0.164**

(0.333) (0.0500) (0.291) (0.0451) (0.342) (0.0571) (0.355) (0.0601) Constant 0.693 0.719*** 0.693 0.719*** 0.746 0.741*** 0.755 0.731***

(0.787) (0.237) (0.885) (0.239) (0.744) (0.215) (0.767) (0.234)

Observations 170 170 170 170 170 170 170 170 Pseudo R2/R-squared 0.130 0.130 0.130 0.130 0.131 0.131 0.136 0.135

Columns (1) - (2) clustered by VSLA group (n = 25), columns (3) – (8) clustered by village (n = 13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

175

Table 21: Have You Made any Housing Improvements in the Last 12 Months? (1) (2) (3) (4) (5) (6) (7) (8)

VARIABLES Probit LPM Probit LPM Probit LPM Probit LPM

Membership 1.582*** 0.530*** 1.582*** 0.530*** 1.390*** 0.472*** 1.152** 0.407** (0.509) (0.144) (0.428) (0.119) (0.383) (0.117) (0.580) (0.171)

Marginal effect 0.552*** 0.552*** 0.499** 0.427** (.1299) (0.111) (0.108) (0.182)

Gender -0.440 -0.130 -0.440 -0.130 -0.393** -0.118** -0.445 -0.130 (0.421) (0.111) (0.310) (0.0820) (0.189) (0.0568) (0.421) (0.111)

Marginal effect -0.173 -0.173 -0.155** -0.175 (0.161) (0.118) (0.073) (0.161)

Membership*Gender 0.0961 0.0212 0.0961 0.0212 0.388 0.112 (0.484) (0.138) (0.387) (0.118) (0.642) (0.197)

Marginal effect 0.038 0.038 0.154 (0.193) (0.154) (0.251)

Membership + (Membership*Gender)

1.678*** (0.301)

0.551*** (0.074)

1.678*** (0.237)

0.551*** (0.072)

1.540*** (0.457)

0.519*** (0.150)

Marginal effect 0.576*** 0.576*** 0.541*** (0.072) (0.067) (0.122) Dosage 0.0591 0.0155 0.103 -0.0201

(0.0504) (0.0168) (0.0851) (0.0356) Marginal effect 0.024 0.041

(0.020) (0.034) Dosage*Gender -0.0694 -0.0201 (0.113) (0.0356)

Marginal effect -0.028 (0.045) Dosage + (Dosage*Gender)

0.034 (0.068)

0.007 (0.025)

Marginal effect 0.013 (0.027) Age -0.036*** -0.016*** -0.036*** -0.011*** -0.040*** -0.011*** -0.034*** -0.011***

(0.0120) (0.0036) (0.0104) (0.0032) (0.0119) (0.0036) (0.0119) (0.0036) Married 0.317 0.0954 0.317 0.0954 0.343 0.100 0.322 0.0921

(0.447) (0.148) (0.504) (0.161) (0.438) (0.146) (0.439) (0.149) Widowed 0.773 0.239 0.773 0.239 0.828 0.248 0.810 0.240 (0.550) (0.173) (0.499) (0.158) (0.538) (0.171) (0.550) (0.177) Divorced 0.540 0.184 0.540 0.184 0.531 0.179 0.535 0.177 (0.665) (0.214) (0.647) (0.211) (0.670) (0.208) (0.670) (0.215) Separated -0.325*** -0.579*** -0.325** -0.579** -0.318** -0.536** -0.316** -0.534**

(0.118) (0.207) (0.213) (0.225) (0.211) (0.215) (0.209) (0.214) Primary -0.108 -0.0172 -0.108 -0.0172 -0.0924 -0.0154 -0.0946 -0.0170

(0.378) (0.118) (0.377) (0.122) (0.355) (0.110) (0.384) (0.119) Ordinary level -0.0789 -0.0102 -0.0789 -0.0102 -0.0961 -0.0155 -0.0793 -0.00970

(0.357) (0.0981) (0.284) (0.0811) (0.334) (0.0924) (0.354) (0.0980) Advanced level -0.0611 -0.00916 -0.0611 -0.0092 -0.0708 -0.0121 -0.0502 -0.0068

(0.407) (0.121) (0.399) (0.124) (0.380) (0.113) (0.405) (0.119) Children 0.0277 0.00858 0.0277 0.00858 0.0225 0.00735 0.0267 0.00844

(0.0572) (0.0181) (0.0607) (0.0199) (0.0587) (0.0182) (0.0568) (0.0179) Prior savings 0.204 0.0634 0.204 0.0634 0.196 0.0631 0.199 0.0645

(0.192) (0.0652) (0.169) (0.0571) (0.186) (0.0638) (0.191) (0.0653) Prior access 0.0741 0.0172 0.0741 0.0172 0.163 0.0387 0.178 0.0433

(0.419) (0.145) (0.420) (0.145) (0.404) (0.140) (0.402) (0.141) Constant 0.0564 0.473** 0.0564 0.473*** 0.121 0.490** 0.146 0.495**

(0.576) (0.189) (0.384) (0.127) (0.561) (0.182) (0.562) (0.187)

Observations 170 170 170 170 170 170 170 170 Pseudo R2/R-squared 0.224 0.287 0.224 0.287 0.229 0.290 0.230 0.291

Columns (1) - (2) clustered by VSLA group (n = 25), columns (3) – (8) clustered by village (n=13) Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1