Abstract
1 Background
1.1 DESCRIPTION OF THE CONDITION
Over the past two decades, microfinance activities have spread across the globe, reaching tens of millions of poor households with tailored financial services. Current estimates vary between 133 and 190 million microfinance borrowers worldwide. This growing importance has resulted in a vast number of research and evaluation studies, including impact studies. Microfinance can best be described as a field of intervention rather than a particular instrument. Initially, microfinance usually meant microcredit for working capital and very small investments, but increasingly it has been broadened to include savings/deposits, (a limited range of) micro-insurance and payment services as well as a somewhat broader range of credit products for more substantial investments. In this study we will only focus on microcredit activities, constituting the bulk of microfinance activities across the globe.
1.2 DESCRIPTION OF THE INTERVENTION
Microcredit is not the panacea for the poorest, but for millions of poor and very poor households it can constitute a potentially powerful tool for development and an escape from the poverty trap. In the literature, claims and supporting evidence can be found of microcredit – with or without complementary services – resulting in a range of positive effects, from income growth and consumption smoothing effects to gender and social empowerment (e.g. Armendariz de Aghion and Morduch, 2005). At the same time, concerns about the independence of some of the impact studies commissioned by microfinance organizations 1 , as well as unease with important blind spots like employment effects (Balkenhol, s.d.), limited macro-economic impact (Honohan, 2004) and an increasing body of evidence on the negligible and even negative effects (mainly on the extreme poor) and lack of rural outreach of microfinance are some of the more recent critical signals emerging in the debates (e.g. Dichter and Harper, 2007).
The amount of studies addressing the role of microcredit in development processes at micro and meso scale, as well as more specific studies addressing the impact of credit on livelihood processes and small enterprise development has increased substantially over the years. Microcredit activities have affected the lives of clients (and others) in multiple ways. The most frequently reported types of effects of credit at individual, enterprise and household level are the following: income, expenditure smoothing, and poverty alleviation effects; business growth and employment effects; schooling effects; and effects in terms of women's empowerment. One of the first comparative studies addressing effects of microfinance using quasi-experiments was Hulme and Mosley's (1996) Finance against Poverty, bringing a new critical voice to the debate by showing the limitations of microfinance in bringing about poverty alleviation. Subsequently, more and more quasi-experimental and regression-based analyses have followed (e.g. Pitt and Khandker, 1998; Khandker, 2003) 2 . More recently, in a few occasions randomized controlled designs have been used to assess particular aspects of microcredit activities (e.g. Giné and Karlan, 2008).
1.3 HOW THE INTERVENTION MIGHT WORK
Despite the diversity in microcredit schemes, many share two characteristics: they target poor women and often rely on some type of group-lending technology. Women's empowerment in relation to microcredit has been studied extensively within the context of this type of microcredit scheme. Most of these studies have been carried out in the context of microcredit group schemes in South Asia. Women's empowerment can be broadly defined as an “expansion in the range of potential choices available to women” (Kabeer, 2001: 81). It has been argued that access to microcredit can foster changes in individual attitudes of women (e.g. increased self-reliance), power relations within the household (e.g. control over resources) and social status (e.g. Malhotra et al., 2002).
Mayoux (2006) argues that the interlinkages between microcredit and women's empowerment are delineated differently by existing paradigms. The financial sustainability paradigm as well as the feminist empowerment paradigm emphasize women's income-earning activities, whereas the poverty alleviation paradigm emphasizes the effects on household expenditures and particularly the use of loans for consumption purposes. Consequently, in the literature one can identify a wide range of measures that try to capture the effect of microcredit on women's empowerment (e.g. Kabeer, 2001).
An important dimension of empowerment concerns women's control over household spending (e.g. Pitt et al., 2003). The main assumption is that by providing credit to poor women, their direct control over expenditures within the household increases, with subsequent implications for the status of women and the well-being of women and other household members. Given the fact that women's control over household spending (through different proxies) is a frequently recurring aspect analyzed within the context of microcredit interventions allows us to study the following question: Does microcredit targeted to women affect women's control over household spending decisions? In what circumstances?
1.4 WHY IT IS IMPORTANT TO DO THIS REVIEW
To our knowledge no systematic review of microcredit activities has been conducted. Instead, one can find a substantial number of synthesis studies which discuss a set of microcredit interventions which are intended to generate claims on impact with a certain degree of external validity. However, most of these studies face certain limitations in terms of depth of empirical assessment, attribution of impact to microcredit, and/or the external validity boundaries of particular contexts. Moreover, methodological principles regarding comprehensive searches and principles of selection, coding, extraction and aggregation are often lacking in these studies. Examples of studies with crossintervention perspectives on impact are for example Hulme and Mosley (1996), Snodgrass and Sebstad (2002), Goldberg (2005), Armendariz de Aghion and Morduch (2005) and EBI (2008).
Overall, these reviews suggest that effects of microcredit are mixed. In part this can be explained by the heterogeneity in microcredit interventions, contexts and target groups. With respect to women's empowerment Kabeer (2001) argues that the divergence in results between different impact studies is mainly due to differences in methodology as well as conceptualization and measurement of the concept of empowerment. Preliminary reading of the literature on microcredit and empowerment points to the prevalence of control over household expenditures as a key (partial) indicator of empowerment. Despite the central and recurrent role (across studies) of this aspect of women's empowerment in relation to microcredit activities there has been no review on this topic.
Methodological quality of microcredit impact studies has often been contested (e.g. Armendariz de Aghion and Morduch, 2005; Karlan and Goldberg, 2006). Randomized controlled trials (RCTs) in microcredit are very scarce. The few RCTs conducted in the context of microcredit interventions that cover aspects of women's empowerment (Banerjee et al., 2009; Kim et al., 2007) in fact look at different outcomes. Banerjee et al. (2009) look at the effect of microcredit on women's control over household spending, whereas Kim et al. (2007) take on board a much larger set of empowerment indicators ranging from changes in self-confidence, to changes in gender norms and partner relationships.
There is a substantial amount of quasi-experimental and regression-based studies available studying the relationships between microcredit and women's empowerment (including women's control over household spending). These studies have been criticized regarding their limitations in addressing selection bias issues. Selection bias is likely to arise in microcredit programs targeting the poor, as it may be expected that women who choose to participate in the program are on average more empowered than those who do not (Goldberg, 2005). An example of a methodological critique comes from a recent paper by Roodman and Morduch (2009). They replicate what was long considered to be the most rigorous study on microcredit impact in the field, i.e. Pitt and Khandker's (1998) 3 study on microcredit in Bangladesh. While Roodman and Morduch's paper does not arrive at substantially different findings, they do point out several methodological flaws endemic to quasi- and non-experimental impact studies on microcredit.
2 Objective of the review
To assess the effects of microcredit on women's control over household spending in developing countries. We will also look at how contextual factors and intervention mechanisms affect outcomes.
Despite the abovementioned limitations in non-randomized designs, this review will include not just randomized experiments but also quasi-experimental (e.g. propensity score matching) and regression-based studies (e.g. panel data regressions). The main reasons for doing so are the following. The first reason is the virtual absence of randomized experiments covering microcredit and empowerment. Second, despite certain methodological limitations, the substantial number of quasi-experimental and regression-based studies available offer the potential to develop (to some extent) generalizable insights on the effects of microcredit in changing women's control over household expenditures as an intermediary step towards women's empowerment. In addition to the main review, based on studies that specifically address the attribution problem in evaluation through counterfactual analysis, we will also consult qualitative studies to enrich the interpretation and analysis of the findings from the main review.
3 Methods
3.1 CRITERIA FOR CONSIDERING STUDIES FOR THIS REVIEW
3.1.1 Types of studies
Analyses based on (combinations of) the following designs are eligible:
Randomized designs Quasi-experimental designs Pipeline approach Propensity score matching Cohort design Ex ante - ex post measurements (reflexive controls), including interrupted time series Other ex post only measurements (with matching) Regression discontinuity Regression-based approaches Cross-section data Panel data (cross-section plus multiple observation points in time; including difference in difference analysis) Instrumental variable analysis
Non-experimental studies without counterfactual analysis are discarded (e.g. single ex post measurement of microcredit beneficiaries only). Qualitative studies addressing the relationships between microcredit and empowerment will be analyzed separately. All matters pertaining to the selection and analysis of qualitative studies will be discussed in the final section (on qualitative synthesis).
3.1.2 Types of participants
Microcredit organizations across the globe cover heterogeneous groups of clients. We look at studies that analyze the effects of microcredit schemes targeting poor women. We will only consider studies that concern microcredit initiatives in low and middle income (or developing) countries, as defined by the World Bank.
3.1.3 Types of interventions
Microcredit interventions targeting women differ in terms of the following characteristics. First, in terms of group-based loans versus individual loans, group-based credit schemes usually refer to a system in which credit is provided to women organized in solidarity groups. Within these groups women receive individual loans. Repayment is enforced through mechanisms of group pressure based on principles such as joint liability and contingent renewal. Individual loans are directly provided to the (female) client. Second, in terms of credit-only versus “credit plus”, in many cases credit is linked to other services such as training or financial services such as savings. While primary studies based on experimental designs can isolate the credit component, this may not be the case in all quasi- and non-experimental studies.
Comparisons in primary studies usually concern groups of women with credit versus women without credit.
3.1.4 Types of outcomes
Changes in the intervention group and changes relative to control or comparison groups will be assessed. Measures relating to women's control over household spending are the following: ability to make small purchases; ability to make large purchases; decisions on expenditures regarding clothing, education, health, food, house repairs; ability to purchase small livestock; ability to purchase large livestock; ability to purchase land. Other relevant outcome measures of interest may be identified during the course of the study and will be included as well.
The present review will not explicitly focus on proxies of women's empowerment beyond the control over household expenditures. However, we do aim to shed light on the causal relationships between control over household expenditures and deeper-lying dimensions of women's empowerment such as self-confidence, changes in intra-household power relations and changes in the social status of women from a theory of change perspective (see below).
Changes in expenditures which usually (traditionally) are under the control of women (given a specific context) are not taken into account if there is no explicit mention of changes in decision- making power or control over expenditures.
Looking at the range of microcredit experiences around the world, there is not always a gender division in the control over household expenditures. In such contexts microcredit may still affect women's empowerment in alternative ways. The present review will not cover this.
3.2 SEARCH METHODS FOR IDENTIFICATION OF STUDIES
Previous searches on microcredit impact studies by team members generated close to 300 studies (see Vaessen et al., 2009). Approximately 60 studies were identified that were related to women's empowerment. The current review will include these studies and expand the evidence base using three parallel search strategies: database searches of published literature, internet searches for published and unpublished research, and manual searches of books. In addition, in all three cases we will use back-referencing from recent studies as well as (where possible) citation-tracking to identify additional relevant studies.
3.2.1 Electronic searches
The following databases will be initially included in the search process.
University of Maastricht library network:
EBSCO Business Source Premier Econlit (EBSCO) Econpapers IBSS (EBSCO) JSTOR PsycINFO (EBSCO) SocINDEX (EBSCO) Source OECD ISI Web of Knowledge
External:
BLDS: LILACS: Googlescholar:
3.2.2 Search terms
Search strategies in databases and journals will be adapted for each source. Where possible we will use the existing keyword indices of particular databases. In general, the following keywords will be used:
a) microcredit
a) microfinance
a) microlending
a) *credit
a) *finance
a) *lending
a) *loans
b) women
b) gender
c) decision-mak*
c) empower*
c) bargain*
c) expense*
c) expenditure*
c) spend*
Keyword combinations relating individual words from a + b and a + c and a + b + c will be used.
3.2.3Searching other resources
Initial searches will include the following websites. When during the search process new relevant sites will be discovered, those will be included as well.
Portals:
CGAP: www.cgap.org
Microbanking Bulletin: www.mixmbb.org
Microfinance Gateway: www.microfinancegateway.org
Microfinance Network: www.mfnetwork.org
Multilateral and bilateral and non-governmental donor organizations: African Development Bank, Asian Development Bank, DFID, EBI, Inter-American Development Bank, USAID, World Bank. Research institutes, foundations and networks: Center for Global Development, Grameen Foundation, BRAC Research and Evaluation Division, IDS, IDEAS, IFPRI, J-PAL.
To the extent that the previously mentioned databases do not cover microfinance-related journals, the following journals will be screened manually:
African Development Review Development in Practice Journal of Development Entrepreneurship Journal of Development Effectiveness Journal of Development Studies Journal of International Development ESR Review (previously Journal of Microfinance) Microcredit review Savings and development Enterprise Development and Microfinance (previously Small Enterprise Development) World Bank Research Observer World Bank Economic Review World Development
We will search the tables of contents of books on microfinance for relevant chapters. Book collections from Maastricht University, University of Antwerp as well as other university libraries in The Netherlands and Belgium will be searched.
Finally, authors of studies which we are unable to retrieve will be contacted. In addition, we will contact experts on microcredit (and women's empowerment) for additional references, which we may have missed.
3.3 DATA COLLECTION AND ANALYSIS
3.3.1 Selection of studies
Studies must be published later than 1980 4 . No restrictions will be placed on language of the study.
Titles and abstracts identified in the search process described above which appear to meet the inclusion criteria will be entered (where possible) in ProCite and subsequently in an Excel sheet. Subsequently, all studies will be independently coded by two reviewers on the basis of the following classification:
Priority 1: article is on the impact of microcredit on women's empowerment; involves original empirical analysis.
Priority 2: article is on the impact of microcredit, covering multiple outcome measures which include aspects of women's empowerment; involves original empirical analysis.
Priority 3: article is on the impact of microcredit and women's empowerment but does not rely on original empirical analysis.
Priority 4: all other studies.
Where necessary full-text copies will be obtained to facilitate coding. Full-text copies will be obtained for all priority 1 and 2 articles. Priority 1 and 2 articles will be screened for methodological design. All purely qualitative studies will be transferred to a separate database. All other studies will be screened for the nature of outcome measures. Articles that do not cover the dimension of women's control over household expenditures will be transferred to a separate database. All other studies will be retained for further analysis.
3.3.2 Data extraction and management
Recorded data will include:
Study characteristics:
authors publication type date of publication country(ies), region(s) evaluation conducted internally (by MFI) or externally (e.g. by independent researchers) identification of purely qualitative studies
5
Intervention characteristics: name(s) of microfinance organization(s) type of microfinance organization(s) format independent variable: access (yes/no), loan amount, loan history independent variable modality (e.g. solidarity group versus individual contract) microcredit exclusively targeting women (yes/no) associated interventions Other intervention(s) received, including in the control or comparison group Client characteristics: gender, age, ethnic group, religion, marital status, income level, main income activities, other outstanding loans, credit history, use of loan, level of indebtedness, etc.
Other moderator variables:
microfinance program integrity
6
, nature of gender relations, regional characteristics
Outcome variables:
indicators of women's control over household expenditures other indicators of women's empowerment
Methodology:
design (estimation method, data structure, method for addressing selection bias) sample size(s) combination of methods used (quantitative and qualitative)
All selected studies will be independently coded by two reviewers and disagreements will be discussed. For guidelines on coding (and extraction) the reviewers will consult Lipsey and Wilson (2001). The microfinance and gender specialist will read a selection of the retrieved articles to provide feedback on updating the information to be retrieved.
3.3.3 Assessment of risk of bias in included studies
Methodological quality will be assessed independently by two reviewers, including a specialist in quantitative methods. Disagreements will be discussed. The following dimensions will be assessed 7 :
clarity of methodological approach reliability of measurement of outcome indicators clarity regarding the effect of microcredit separate from or linked to associated interventions (e.g. education, training) quality of and technique used in addressing selection bias problem (i.e. through randomization, matching, instrumental variables technique) (where appropriate) quality of matching (e.g. in propensity score matching: checking for the assumptions of conditional independence and assumption of common support) (where appropriate) quality of addressing identification problem (which is related to the quality of the theory underlying the causal model)
Studies of low methodological quality will be discarded. A list of discarded studies will be provided along with an explanation.
Qualitative studies will not be used in the first part of the synthesis, but only to complement the findings. The assessment of quality of qualitative studies is discussed in the final paragraph.
3.3.4 Measures of treatment effect
The prospects for statistical aggregation through meta-analysis roughly depend on the following aspects:
the number of impact studies based on randomized or quasi-experimental design that report one or more effect sizes, or from which effect sizes may be calculated; homogeneity in ‘treatment’; or sufficient studies on impact of microcredit with similar ‘treatment’ characteristics; homogeneity in outcome measure; or sufficient studies reporting on the same type of outcome measures; outcome measures are likely to diverge between studies on empowerment; homogeneity in type of target group and context; the characteristics of clients, their livelihood conditions and existing gender relations may differ substantially across studies.
We expect to find sufficient quasi-experimental studies on microcredit. However, heterogeneity in methodological design, outcome measures or other variables may reduce the feasibility of performing a reliable and meaningful meta-analysis. Where possible, we will calculate risk or odds ratios for dichotomous outcomes and standardised mean differences for continuous outcomes, with corresponding 95% confidence intervals for comparison across studies.
3.3.5 Assessment of heterogeneity
We will assess heterogeneity using I-squared and Tau-squared statistics.
3.3.6 Assessment of reporting biases
Provided sufficient studies are available, we will examine presence of file-drawer effects using funnel plots.
3.4 DATA SYNTHESIS
3.4.1 Subgroup analysis, moderator analysis and investigation of heterogeneity
Where possible, effect estimates will be synthesised using inverse-variance weighted random effects meta-analysis. We will use STATA software for all quantitative analysis (STATA Corp, TX). We will conduct separate meta-analyses for appropriate groups of outcomes variables, corresponding to different dimensions of empowerment such as control over household spending, decision-making power, and other dimensions. Where sufficient studies exist, we will also test for effect moderators according to study characteristics, intervention characteristics and client characteristics, as defined above.
3.4.2 Sensitivity analysis
We will examine sensitivity of results to study quality, based on the risk of bias assessment.
3.4.3 Synthesis of qualitative data
The context-specific nature of existing gender relations and the implications for how microcredit might affect these justify a theory-based approach to synthesis. We will use the realist notion of context, mechanisms and outcomes, or the CMO principle (Pawson and Tilley, 1997) as a format for synthesis. Van der Knaap et al. (2008) discuss a systematic review process which combines Campbell standards of systematic review with the CMO principle as a basis for aggregation. In other words, first, studies based on designs with a high degree of internal validity and of high relevance to the topic are identified. Subsequently, CMO principles are applied to the selected studies. In our case, information on context, mechanism and outcomes is (partly) covered by the variables described above: intervention characteristics, client characteristics, (other) moderator variables and outcome variables. During the course of the study we will update the information to be retrieved in order to strengthen this synthesis. This is especially important regarding moderator variables. When building our CMO theory, we intend to use our coded information on independent and dependent variables (type plus effect) in a grounded theory type of approach where we iteratively gather more information from our sample of studies on important moderator variables, gradually trying to uncover “demi-regularities” (Pawson, 2009) on what works in what circumstances.
Subsequently, having used all eligible quantitative studies to develop a CMO theory, we will consult additional qualitative studies identified in our search to enrich the theory. As discussed in the Cochrane Handbook (Cochrane, 2009), there are several options for extending the review using purely qualitative studies. There are at least two important reasons why it may be important to include qualitative studies. First of all, quantitative studies with a high degree of internal validity (e.g. randomized designs) are often not particularly informative regarding the nature of the context of interventions and target groups. By contrast, qualitative studies often present a richer picture of the processes of change that are induced by women's access to and use of microcredit. A synthesis of qualitative studies will complement our analysis by providing additional information on the nature of the relationships between microcredit, women's control over household spending and other dimensions of women's empowerment. In our case, we will conduct a small complementary review using purely qualitative studies.
We will only consider studies that include original empirical analysis on the relationship between microcredit and women's control over household spending. Moreover, only qualitative studies that cover the same interventions as in the studies selected for the main review will be considered. Thirdly, inclusion also depends on a study's potential to further our understanding of particular aspects of the CMO constructions developed in the main review (as described in the previous paragraph 8 ). Finally, we will perform a succinct assessment of the study's quality. There are no universally accepted standards for assessing the quality of qualitative research. We assess the quality of studies by looking at two aspects: quality of the theoretical model underlying the analysis (are the researchers aware of and build on existing work in the field of microcredit and women's empowerment), transparency in the description of the study process 9 .
In addition, we will apply the principle of triangulation in order to decide whether or not to include particular information from qualitative empirical studies in the CMO theory. This works in two ways. First, the validity of a particular finding from a study increases if more than one study report similar findings on a change process. Second, if our CMO theory (see above) is reflected in a particular qualitative study, then additional information from that study is more likely to be valid and applicable to enrich the interpretation of our overall theory. Both principles will be applied with caution (see also Pawson, 2003).
Footnotes
5 Sources of support
1
The microfinance sector counts an increasing number of mid-size and big microfinance organizations many of which continue to depend (in part) on development assistance, private capital and other sources of finance to fund their growing portfolios.
2
USAID founded a special group called ‘Assessing the Impact of Microenterprise Services’ (AIMS) which among others focused on quasi-experimental impact analysis of microfinance.
4
Most microfinance institutions were founded in the 1980s or later. A few were founded in the 1970s. The roots of the Grameen Bank, a pioneer institution in the field of microfinance, can be traced back to 1976, but it became an independent microfinance institution in 1983.
5
These will be stored in a separate database for qualitative synthesis (see below).
6
Performance of the microfinance organization (e.g. quality of service delivery) can also affect outcomes.
7
We will consult the Cochrane Collaboration's tool for assessing risk of bias for providing guidance on quality assessment of randomized designs and where appropriate other designs (see Cochrane, 2009).