Value chain interventions for improving women's economic empowerment: A mixed-methods systematic review and meta-analysis

Value chains

Traditionally, agricultural interventions have tried to improve farmers' incomes by increasing farm productivity. This has been pursued, for example, through the introduction of new farming technologies and agricultural extension, and by providing access to inputs such as fertiliser, seeds, and loans. However, developments in global market integration over the last 30–40 years suggest that farmers will have a chance to access better income opportunities by strengthening their positions within agricultural product markets. For example, farmers may gain considerably by supplying their goods to a large retailer located in a high-income country, or they may obtain better prices after storing and processing their produce.

The ‘value chain’ is a conceptual framework for understanding agricultural product markets and identifying unexploited non-farm opportunities. Although the concept of a value chain is multivariate in how it is used, for example, by different UN agencies (Stamm & Von Drachenfels, 2011); it is primarily used to represent a set of primary and support activities undertaken by an organisation to create value for its customers and, thus, to enhance the organisation's competitiveness (Porter, 1985).

To this end, Kaplinsky and Morris (2000) explained a value chain as ‘the full range of activities which are required to bring a product or service from conception, through the different phases of production (involving a combination of physical transformation and the input of various producer services), delivery to final consumers, and final disposal after use’. The simplest characterisation of a value chain, therefore, includes four links: design, production, marketing and consumption.

However, agricultural value chains are more complex than this simple characterisation. For example, the FAO guidelines on ‘developing sustainable food value chains’ include two additional stages between production and marketing: ‘aggregation’ and ‘processing’ (Neven, 2014). Aggregation refers to the process of aggregating and storing the produce of a multiplicity of small farmers by intermediaries, while processing refers to the physical processing of agricultural goods (e.g., pulping and drying coffee beans).

Figure 1 illustrates the typical nodes of an agricultural value chain, as well as a tentative classification of ‘value chain interventions’ indicating the node of the chain in which they primarily operate. The tangle of arrows leading from interventions to the nodes in the chain shows how interventions can address constraints at different points in the chain simultaneously. Value chains are embedded within institutional, economic, social and natural environments that condition their operation. For example, the movement of products along the chain will be conditioned by legal frameworks on quality standards, availability of transport infrastructure, or vulnerability to droughts.

OPEN IN VIEWER

Figure 1

Agricultural value chain and associated interventions. FBOs, farmer-based organisations; SME, small and medium-sized enterprise.

Value chains may develop different levels of complexity. A useful distinction is made between traditional, transitional and complex value chains (De Brauw & Bulte, 2021; Barrett et al., 2022). Traditional value chains, often related to subsistence crops, are localised and include few nodes, sometimes directly connecting producers to consumers. Transitional value chains, often involving ‘cash crops’, include more nodes in the chain, such as storing and processing, and a complicated distribution system.

Complex value chains, normally based on internationally traded crops, are articulated in several nodes, and include the participation of several actors, some of which are large firms. In complex value chains, goods travel long distances, and relationships between actors are regulated by contracts and other institutional arrangements that certify the quality of the products or insure their value.

In practice, value chains are much more complex. For example, Figure 2 provides a representation of a value chain of green beans in Kenya. It displays a higher degree of complexity than Figure 1; however, value chain diagrams in the literature tend to be more complex than this. Typically, they are composed of several vertical supply chains, which illustrate the journey of a product from conception to consumption. The same product is often supplied through different channels in the same country, resulting in several vertical chains in the same value chain.

OPEN IN VIEWER

Figure 2

Example of a value chain: Green beans in Kenya. SME, small and medium-sized enterprise. Source: Webber and Labaste (2010).

The functions performed at each link can be read horizontally: for example, in Figure 2, importers can be ‘wholesale importers’ or ‘distributors’, and end markets may consist of ‘supermarkets’ or ‘wholesale markets’. The entire value chain is therefore a matrix of functions and channels.

Value chains are built around specific agricultural products. There is no common definition of agricultural products. The FAO FAOSTAT database includes the following items: primary crops (e.g., rice, wheat), processed crops (e.g., cotton, oil, wine), live animals (e.g., goats, pigs), primary livestock products (e.g., milk, meat, eggs) and processed livestock (e.g., butter, cheese, silk). Forestry products include fuel wood, industrial wood and processed woods (e.g., paper, pulp). Fishery products include all fish varieties, aquatic plants, aquatic animal products (e.g., pearls, corals), aquatic animals (e.g., frogs, crocodiles, crustaceans, molluscs) and mammals (e.g., whales, seals).

Webber and Labaste (2010) and Haggblade and colleagues (2012) provided several examples of value chains including sorghum beer in Botswana, dairy products in Kenya, domestic catfish in Kenya, floriculture in Uganda, green beans in Kenya, cashews in Mozambique, pineapples in Ghana, coffee in Rwanda, cotton in Zambia and avocados in Kenya.

The core stages along the value chain are production, aggregation, processing and distribution (wholesale and retail). Embedded in each of these stages are several value chain interventions. In Figure 3 below, we present the value chain stages that all products went through (Morioka & Nicholas, 2014).

OPEN IN VIEWER

Figure 3

Value chain processes.

In value chain analysis, the following strategies are identified to improve farmers' incomes: vertical coordination, horizontal coordination and upgrading. These can be interpreted as fundamental mechanisms that are expected to operate in isolation or in conjunction with other value chain interventions. All these mechanisms add value to products along the chain and/or allow the actors to capture a higher fraction of the added value.

One finds the mechanisms articulated by different terminologies. Multiple terms used to identify the same mechanism (e.g., ‘functional upgrading’, ‘vertical integration’, or ‘value chain deepening’) could affect the clarity with which such discussions occur. Despite the multiple differences in terminologies, we identify the following mechanisms as vital to operation within value chain interventions:

Vertical coordination. This refers to establishing links between market agents along the value chain running from production to final consumption. New links can be established between farmers and buyers that can offer more remunerative prices (contract production). For example, farmers may sign profitable supply contracts with supermarket chains. Furthermore, farmers can undertake operations related to connected links (vertical integration), such as producers taking charge of the first stages of processing the product before selling it, thus adding considerable value to the end product.

Horizontal coordination. Farmers at one particular node of the value chain may aggregate and coordinate activities through farmer-based organisations or co-operatives. There are many potential benefits arising from coordination, such as an increase in bargaining power when negotiating prices, the exploitation of scale economies (e.g., through collective purchase of machinery), risk pooling, and other reductions in transaction costs.

Upgrading. Activities along the chain can be improved in several ways, and authors often distinguish ‘product’ from ‘process’ and ‘functional’ upgrading.

The distinction between these types of strategies is not always implicit and overlaps are possible. For example, horizontal integration (the organisation of farmer-based organisations) may lead to vertical integration (producers' farmer-based organisations taking charge of processing activities along the chain). Product upgrading (e.g., organic farming) may lead to vertical contract production as farmers gain access to specialised buyers. In general, a strategy targeting a particular link in the chain will have ripple effects on other links, and value chain interventions will often encompass a combination of the strategies above.

More recently, strategies for gender-sensitive and gender-responsive value chain interventions have been developed by several agencies and international NGOs. These strategies, as discussed below, aim to provide consolidated support for women to access the benefits offered by value chain interventions.

First, the gendered nature of tasks in rural economies often pushes women toward less remunerative nodes in the value chain, such as agricultural labour, petty trading and subsistence farming. There is scope to strengthen the value chain positions currently occupied by women and to enable their participation in more remunerative positions. However, to do this, it is first essential to identify and address the barriers faced by women who are willing to participate in the agricultural value chain.

Second, women are less likely to adopt new technology in their activities, primarily due to the very reasons that prevent their participation in profitable nodes in the value chain.

To this end, the FAO guidelines for sustainable value chains in particular list the following factors: poor access to information and training, limited participation and decision-making power within household and communities, limited access to financial services and other productive assets and resources, and excessive work burdens. These factors are in turn embedded in reinforcing cultural norms and institutions. Hence, interventions aiming to make the services accessible for women is required.

We can identify three principal ways in which agencies can promote gender-responsive value chain interventions. By gender-responsive, we mean interventions that in their design and implementation employ gender considerations.

Promote women-led nodes in value chains. Rural economies usually operate via gender-specific roles. Analysis of any value chain suggests that some links are entirely occupied by male farmers, while others are dominated by women. For example, women may oversee the processing of a staple crop, or retailing. An intervention may therefore decide to strengthen the node in the chain which is dominated by women to achieve the maximum benefit for them.

Promote women's participation in male-dominated value chains. An intervention may decide to promote a particularly remunerative node of the value chain (e.g., linking farmer-based organisations to buyers connected to export markets). However, this link may be dominated by male farmers. The intervention can therefore promote women's participation in farmer-based organisations and support their presence in leadership roles in a way that shares the benefits of the intervention across gender groups. The intervention may attempt to specifically promote the inclusion of women in highly profitable value chains or more profitable nodes of the value chain that are traditionally controlled by men.

Address discriminatory constraints to accessing value chains. Selecting value chain activities that are dominated by women and promoting their participation in project activities may be insufficient. Fundamental constraints must be identified that prevent women from participating in or enjoying the benefits of project activities. The value chain intervention may include specific complementary interventions that are oriented toward increasing women's participation in markets and in decision-making capacities. For example, a project might include loans specifically targeted at women, or make provisions for childcare to enable women's effective participation in the activities.

The gender-responsive value chain interventions that we will consider in our review are:

Value chain interventions explicitly targeting women;

Chain empowerment matrix

The chain empowerment matrix (Figure 4) is a useful framework

OPEN IN VIEWER

Figure 4

Chain empowerment matrix.

To explain further, farmers can undertake different activities such as drying or fermentation of their crops (post-harvest activities), or grading, processing, transporting and trading. Vertical integration/coordination is understood as involvement in various activities in the chain. Farmers concerned with management of the chain are usually involved in decision-making processes and control over management issues. Farmers are in control of decisions such as sale quantities and prices. Involvement in multiple chain management issues is known as horizontal integration/coordination (Senders et al., 2012).

Participants' ability to conduct value chain activities (e.g., processing, post-harvesting) more efficiently and profitably could result in chain empowerment. It reduces transaction costs by integrating activities that were previously handled or controlled by others. Participants gain control over the value chain processes through improved negotiation capacities, and build stronger relationships with other stakeholders in the value chain (Morioka & Nicholas, 2014).

Four dimensions of gender empowerment in value chains are mentioned below (Table 2). For this review, we focus on economic empowerment.

It is important to note that globally, approximately 75% of women farmers play active roles as producers, traders, processors, labourers and entrepreneurs; however, they are disadvantaged as their roles are largely unrecognised (Gurung et al., 2015). This is due to the many constraints they face, such as limited access to land, irrigation, productive inputs, financial credit, property rights, new technologies and markets (Agarwal, 2018; Doss & SOFA Team, 2011; Gurung et al., 2015; Mayoux & Mackie, 2008; Mutua et al., 2014; Rubin et al., 2010; Ugwu, 2019).

The evidence suggests that gender differences and inequalities exist at all levels of the value chain, and are the main barriers to women's participation in agricultural activities (Gurung et al., 2015; Ugwu, 2019). Other barriers include: (1) customs, beliefs and attitudes that confine women to the domestic sphere; (2) women's workload and time poverty; and (3) laws that affect access to resources, employment and education. It has been further elaborated that women's participation in markets is restricted due to a lack of accessible transportation, as well as social norms that prohibit them from travelling (Mayoux & Mackie, 2008; Quisumbing et al., 2013; Ugwu, 2019). The intended outcomes of some value chain interventions entail addressing of these constraints – particularly by increasing women's free time, mobility, income and decision-making.

For the ToC, we combine the existing approaches and focus on interrelated dimensions, access to productive resources, power, and agency. The same approach was adopted by FAO.

OPEN IN VIEWER

Figure 5

Theory of change. FBOs, farmer-based organisations.

The left side of the figure lists the value chain interventions:

Vertical coordination (contract farming [production]);

This ToC explains the inputs and activities required to attain the desired intermediate outcomes and, ultimately, women's economic empowerment. The casual pathways in Figure 5 can be followed from left to right. In the proposed ToC, we illustrated how access to inputs (e.g., fertilisers, credit, loans, information, training) can lead to the intended outcomes.

The value chain interventions may provide or promote women's contributions and aim for their economic empowerment (e.g., processing or storage technologies that add value to the product, which we would consider under ‘post-harvest management’ and ‘processing development’ interventions). The approach of ‘power with’ – such as the establishment and strengthening of farmers' groups, collectives, and more female farmers' groups – helps to improve income, decision-making, bargaining, women's participation, and economic empowerment.

Embedded in each of the stages are several value chain interventions; our ToC explains these, as well as how they may lead to women's economic empowerment. We assume that women's participation in the value chain as actors, partners and co-owners offers opportunities for economic benefit and empowerment.

There are various stages of value chain processes and involving them all in this ToC risked overcomplication. In the protocol, we listed classifications of interventions and intermediate outcomes. The current ToC is tentative, and one aim of our work is to build one that is theory-informed (or middle-range).

To achieve the outcomes – considering empowerment as an outcome – we have some assumptions. The first is that women have access to resources such as irrigation, productive inputs, financial services and new technologies. It has been also assumed that women actively participate and have access to knowledge and training. There is evidence that women engaged in farming are often exploited by other actors in the value chain (buyers and suppliers) due to their high levels of illiteracy, weak bargaining power, and lack of knowledge and access to market information (Gurung et al., 2015). We assumed that value chain interventions helped to address the above-mentioned constraints to achieve the intended outcomes.

The review will include the value chain interventions which engage women in agriculture value chain processes in different capacities. The interventions aim to increase women's rights and economic power. The interventions targeted the community to create an enabling environment to engage women farmers. It can also focus on land access and ownership, changes in policies, and social norms (Doss & SOFA Team, 2011).

A review of value chain strategies by UN agencies (Stamm & Von Drachenfels, 2011) and international NGOs (Webber & Labaste, 2010) led to the identification of the ‘value chain’ interventions in Table 1, which are classified by the main strategy above. The goal of value chain interventions is improvement in farmers' incomes, particularly small farmers. This should be achieved through better integration in markets, access to more remunerative prices, the development of new and more profitable products, and the adoption of functions that add value to production, such as processing or storage.

Types of participants

Included reviews examined women of all ages from low- and middle-income countries

Types of interventions

The interventions of interest under this review are those aiming to empower women through value chain development and market engagement implemented in the agricultural sector. These interventions aim to increase women's market participation in the value chain by removing barriers to participation, or by changing social norms in organisations, communities, and the family to facilitate participation.

The interventions may include the following:

Inclusive market systems development

Types of outcome measures

The outcomes of interest for this review include measures that result in women's economic empowerment and the various dimensions embedded in this (Supporting Information: Appendix E). The listed indicators reflect the main areas of empowerment within agricultural value chains and markets.

Not all studies will report impacts in terms of empowerment outcomes. We also consider other outcomes along the project causal chain (Table 3); these do not characterise empowerment but are often preconditions to it. In particular, we consider:

Economic outcomes, such as incomes, prices and market sales; and

Types of studies

To answer Objective 1, we used an experimental or quasi-experimental design. Eligible designs included those in which the authors use a control or comparison group, and in which one of the following is true:

Participants are randomly assigned (using a process of random allocation, such as a random number generation);

No restriction will be placed on the duration of follow-up.

The evidence incorporated under Objective 2 is broader, to answer that we included impact evaluations, as well as studies collecting and analysing qualitative evidence. Fulfilling Objective 3 involved the examination of plausible pathways, barriers, and facilitators of included studies across contexts, from the evidence included in the review, and a conceptual framework will be articulated.

Electronic searches

We used the strategies to identify completed and on-going potential studies mentioned in Supporting Information: Appendix A. The search included the following databases: Scopus, Web of Science Core Collection (Social Sciences Citation Index [SSCI], Science Citation Index Expanded [SCI-EXPANDED, Conference Proceedings Citation Index – Science [CPCI-S], Conference Proceedings Citation Index – Social Science & Humanities [CPCI-SSH], and Emerging Sources Citation Index [ESCI]), International Bibliography of the Social Sciences, EconLit, Business Source Premier, APA PsycInfo, Cochrane Central Register of Controlled Trials, Cochrane, Database of Systematic Reviews, CAB Abstracts and Sociological Abstracts.

Supporting Information: Appendix A also presents an example of the keyword strings used for publication databases and search engines, with terms for interventions, regions, and methodologies. We also conducted a machine-learning assisted search in EPPI-Reviewer beta version (Open Alex). Open Alex data set, like Google Scholar, is a comprehensive repository of research articles containing 250 million bibliographic records. It searches not only academic databases but also grey literature sources.

We also conducted a grey literature search from organisation websites such as: CGIAR; the International Fund for Agricultural Development; AgriProFocus; the Bill & Melinda Gates Foundation; Donor Committee for Enterprise Development; the UN Food and Agriculture Organisation; the International Labour Organisation; the Netherlands Development Organisation; USAID; the Swiss Agency for Development and Cooperation; the International Food Policy Research Institute; World Agroforestry; the International Livestock Research Institute; the Foreign, Commonwealth & Development Office; the British Library for Development Studies; AGRIS; IMMANA grant database; the 3ie impact evaluation database; Innovations for Poverty Action; J-PAL; the World Bank IEG evaluations; the USAID Development Data Library; Experience Clearinghouse; the proceedings of the Agriculture, Nutrition and Health Academy conference; the proceedings of the Centre for the Study of African Economies Conference; the proceedings of the North East Universities Development Consortium Conference; and the World Bank Economic Review.

Screening and study selection

The screening for inclusion or exclusion was undertaken in two stages by two independent researchers in EPPI-Reviewer. The first stage title and abstract screening were performed using EPPI-Reviewer's machine learning software. The second stage involved full text screening. The screening was performed with the inclusion criteria for this review and a third-party arbitrator in case of disagreement (SKM). The tool was developed based on PICOS for screening the studies to decide on eligible studies (Supporting Information: Appendix B).

Data extraction and management

Assessment of risk of bias in included reviews

Confidence in the findings of all studies included in the review was assessed using a critical appraisal tool for primary studies developed by the Campbell Collaboration Secretariat. The tool covers both quantitative and qualitative studies (see Supporting Information: Appendix D: Critical appraisal tool, along with a description of how the tool was developed). Critical appraisal assessment was completed by two reviewers.

The tool contains critical dimensions of the evaluation. Each of these is marked as high, medium, or low. The overall score uses the ‘weakest link in the chain’ principle. Hence, the confidence in study findings can only be as high as the lowest rating given to the six critical items in an effectiveness study, and nine critical items in a qualitative/process evaluation.

Data analysis

Data synthesis

In this review, we computed the standardised mean difference (SMD) from the available information found in primary studies such as means, regression coefficients, or chi-squared tests from analysis of variance. We computed the OR from available information on proportions and frequencies. The effects size calculation performed based on the effect size coding tool developed for this review, and, in a minority of instances the Campbell Online Effect Size Calculator.

Meta-analysis of effect sizes for each outcome were conducted using Stata. A weighted mean effect size for each outcome was reported under a random-effects model. Overall effect sizes for primary outcomes were reported in the common metric of odds ratios to communicate with policymakers and practitioners.

Sensitivity analysis

The sensitivity analysis was performed by removing studies from the meta-analysis one by one to see if the results of the meta-analysis are sensitive to any single study. We will also examine the sensitivity of findings by risk of bias (low risk, some concerns, and high risk).

Treatment of qualitative research

In this review, we adopted an approach that combined the qualitative data with the quantitative meta-analysis, within the framework of a theory-based systematic review (White, 2018). In the framework, the unit of analysis is the intervention, rather than an individual study. Different studies may contribute findings at various stages of the casual chain. For example, process evaluations and qualitative studies provide more evidence on implementation issues than most effectiveness studies (such as a lack of women's involvement in the programme and why), which can help to explain both the size of, and variations in, effect sizes.

The framework is shown in Table 4. Quantitative data are indicated as Qt and qualitative as Ql. Quantitative data refer to both effect sizes and factual quantitative data, such as participation rates.

Table 4 shows the theory-based systematic review TBSR framework, which is used for both horizontal and vertical synthesis (White, 2018). In Table 5 an abbreviated version of the row headings from Table 4 are pivoted to become column headings. The data in Table 4 are subject to vertical, horizontal, and total synthesis.

Vertical synthesis involves summarising the evidence across all cases, which is the way in which systematic reviews are usually performed, especially for quantitative analysis of effects. In the case of qualitative data, vertical synthesis is a thematic analysis, in which common themes are identified across studies.

Horizontal synthesis summarises across a case – which may be performed through in narrative reviews; however, the difference here is that the data for an intervention may be drawn from more than one study. The overall synthesis combines both, though it may well contain separate overall synthesis by subgroup. The overall synthesis approach, drawing on both horizontal and vertical synthesis, ‘tells the story’ of whether the intervention works, for whom, under what circumstances, and why.

Characteristics of the included studies

The studies included:

4 mixed-methods studies;

At the coding stage, studies were excluded that did not meet the methodological inclusion criteria (valid control group), did not evaluate a value chain intervention, or did not analyse women's economic empowerment outcomes. In some cases, we were not able to obtain the full text of the identified studies.

Geographical area

The majority of the studies included in the analysis are from Sub-Saharan Africa, accounting for 57% (67 studies). Approximately 30% (35 studies) are from South Asia, 8% (10 studies) from Latin America and the Caribbean, 5% (6 studies) from East Asia and the Pacific, 3% (3 studies) from South America, 2% (2 studies) from the Middle East and North Africa, and 1% (1 study) from Europe and Central Asia. The majority of the studies (63%) are from lower-middle income countries (74 studies), followed by low-income countries (31%, 37 studies). Only 8% (10 studies) are from upper-middle income countries (Figure 7).

OPEN IN VIEWER

Figure 7

Geographical representation of the included studies.

The distribution of studies across various countries shows a notable concentration in several key countries. The majority of the representation comes from India with 17% (20 studies), followed by Uganda and Kenya with 10% each (12 studies each). Following this, 6% (seven studies) represent Ghana and Bangladesh, and 5% (six studies) represent Tanzania, Ethiopia, and Nepal. Nigeria and Nicaragua each account for 4% (five studies). Similarly, China, Mozambique, Zambia, and Burkina Faso each represent approximately 3% (four studies). There are three studies from Indonesia, Rwanda, and Sri Lanka. Only one or two studies each are from the following countries: Côte d'Ivoire, Burundi, Peru, Colombia, Malawi, Senegal, Mexico, Guatemala, the Philippines, Pakistan, Syria, Chile, Zimbabwe, the Solomon Islands, Morocco, the Dominican Republic, Tajikistan, the Republic of Congo, Thailand, Laos, Vietnam, Cambodia, and East Timor.

Study populations

Gender

A total of 64% (75 studies) included both genders, while 36% (43 studies) exclusively focused on interventions targeting women (Figure 8). There were 43 studies on interventions for women only, such as women's self-help groups (Anand et al., 2019; Desai & Joshi, 2013; Mohapatra, 2016; Raghunathan, 2019), Grameen Ghana's microcredit scheme (Al-Hassan et al., 2011), producer association farmers' centres (Desai, 2014), women in aquaculture (Panda et al., 2012), IMC Microcredit Programme (Pellegrina, 2021), BRAC Microfinance Programme (Namayengo, 2017), and the National Dairy Development Board programme (Papa et al., 2000).

OPEN IN VIEWER

Figure 8

Gender representation of included studies.

Effectiveness studies

The critical appraisal tool assess the confidence in findings of the studies based on key features including study design, attrition, power calculation and reporting. The tool contains critical dimensions, each of which is marked as ‘high confidence’, ‘medium confidence’, and ‘low confidence’. The overall score uses the ‘weakest link in the chain’ (or maxi-min) principle. Hence, the confidence in study findings can only be as high as the lowest rating given to the six critical items in effectiveness studies.

In this review, most of the effectiveness studies included mixed-methods studies (n = 46; 73%) are rated as having low confidence in their study findings, with only 22% (14 studies) rated as medium confidence, and 5% (three studies) rated as high confidence (see Figure 10). The studies score highly in terms of framing the evaluation questions, detailing the intervention and the outcomes, and attrition of study participants.

OPEN IN VIEWER

Figure 10

Critical appraisal of effectiveness studies.

Few studies are randomised controlled trials due to the difficulty of setting up experiments to evaluate these types of interventions; the vast majority use quasi-experimental methods such as matching, difference in difference, and instrumental variables. These types of studies often rely on existing data and do not perform power calculations beforehand to define the optimum sample size. Because of the predominance of quasi-experimental studies and the rarity of power calculations, the included studies obtain an average score of low confidence.

Process evaluations

In this review, there are a total of 55 process evaluations and 4 mixed method studies. According to our assessment tool, approximately 30% (18 studies) were rated as high-confidence in study findings, 37% (22 studies) were rated as medium confidence, and 32% (19 studies) were rated as low confidence (see Figure 11). The assessment tools considered all the items critical items and it follows the principle of the weakest chain principle.

OPEN IN VIEWER

Figure 11

Critical appraisal of process evaluation.

The studies score highly in terms of framing the evaluation questions, detailing the intervention and the outcomes, and mentioning the qualitative methodology and recruitment or sampling strategies.

The predominance of low rating overall is due to lower rating lack of reporting on researcher's assumption and ethical consideration.

Meta-analysis by outcome domains

Average impacts of the interventions were estimated through meta-analyses, firstly, by outcome domain and subsequently by outcome subdomain. To generate forest plots at the outcome level, we first calculated a single synthetic effect size for each study by taking the weighted average of effects across subcategories by outcome, as indicated in the methods section. The overall forest plot for outcome domains is shown in Figure 12.

OPEN IN VIEWER

Figure 12

Overall meta-analysis findings by outcome.

The findings indicate that the outcomes in nearly all cases favour the intervention, meaning that a positive effect is estimated; the magnitude of effects are of small to medium size. In spite of the small effect sizes and owing to the large numbers of studies we were able to find and incorporate in the analysis, most effects are statistically significant (i.e., the 95% CIs do not overlap the null effect line). Figure 13 presents individual forest plots for the outcomes shown in summary in Figure 12, indicating the heterogeneity in effects across the contributing studies.

OPEN IN VIEWER Figure 13 OPEN IN VIEWER

Thus, for knowledge and access, evidence from 17 studies provides the estimated pooled effect of 0.06 standard deviations improvement in the outcome (SMD = 0.06, 95% CI = 0.00, 0.13) with substantial between-study heterogeneity (I ² = 0.71%, Tau² = 0.09). Most of the remaining findings suggest medium-size effects: for market outcomes (SMD = 0.07, 95% CI = 0.01, 0.13; I ² = 85%, Tau² = 0.02; 33 estimates); economic outcomes (SMD = 0.10, 95% CI = 0.05, 0.14; I ² = 78%, Tau² = 0.01; 41 estimates); and empowerment outcomes (SMD = 0.10, 95% CI = 0.04, 0.16; I ² = 82%, Tau² = 0.02; 24 estimates). The effect for improved inputs, where the pooled effect is estimated of 0.15 standard deviations improvement in the outcome (SMD = 0.15, 95% CI = −0.01, 0.31), is not statistically significant and estimated with substantial estimated between-study heterogeneity (I ² = 98%, Tau² = 0.18; 29 estimates).

These results mean that value chain interventions are on the whole successful at improving economic and empowerment outcomes. All outcome domains are positively affected by the interventions, the average impacts are not large, and the impact on empowerment outcomes is significant in comparison to outcomes observed in other domains. However, these estimates are grouping together very different outcomes and projects that differ greatly in characteristics and implementation; they are therefore concealing a large amount of heterogeneity. A better understanding of the effectiveness of the interventions requires greater differentiation of the outcomes and of the interventions.

It is also noteworthy that, while effects on outcomes are positive in most cases, there is no beneficial effect on time use outcomes – that is, interventions do not act to reduce women's burden of working time or increase their available time for leisure. Indeed, effects were time-neutral, as there were neither significant reductions nor significant increases in time spent on agriculture and household chores (SMD = −0.01, 95% CI = −0.05, 0.02) with very little heterogeneity in findings across the studies (I ² = 24%, Tau² = 0.00; 11 estimates).

Meta-analysis by outcome subdomains

Further analysis revealed some potentially important differences in findings by outcome subdomains. Taking the knowledge and access outcome group (Figure 14), the meta-analysis found a moderate and significant increase in study participants trained (SMD = 0.16, 95% CI = 0.10, 0.22; I ² = 0%, Tau² = 0.00; 4 estimates), but there were smaller or null effects for the other outcomes, such as knowledge (SMD = −0.036, 95% CI = −0.326, 0.253; I ² = 72%, Tau² = 0.06; 4 estimates), access to credit (SMD = 0.016, 95% CI = −0.029, 0.061; I ² = 34%, Tau² = 0.00; 8 estimates) or a bank account (SMD = −0.060, 95% CI = −0.247, 0.127; I ² = 83%, Tau² = 0.02; 3 estimates), or a farmer collective (SMD = 0.081, 95% CI = −0.066, 0.227; I ² = 38%, Tau² = 0.01; 3 estimates).

OPEN IN VIEWER

Figure 14

Meta-analysis by knowledge and access outcome subdomains.

However, one study of gender transformative programming (Cole et al., 2020) found a large impact on gender attitudes. Overall, however, the studies are too few and the between-study heterogeneity is too high to conclude with sufficient confidence about impacts on knowledge and access outcomes.

The meta-analysis for input subdomains (Figure 15) suggested that the substantial heterogeneity in findings at the outcome domain level were indeed due to differences across the subdomains. For example, while other improved inputs (which include adoption of improved technologies) appear to have increased under the interventions (SMD = 0.271, 95% CI = −0.008, 0.550), the findings are not statistically significant at 5% confidence, and heterogeneity is very high (I ² = 99%, Tau² = 0.31; 16 estimates).

OPEN IN VIEWER

Figure 15

Meta-analysis by input subdomains.

Similarly, effects on access to land were small and inconsistent (SMD = 0.068, 95% CI = −0.074, 0.210; I ² = 91%, Tau² = 0.04; 9 estimates), as were effects on reduction input costs (SMD = −0.039, 95% CI = −0.113, 0.035; I ² = 73%, Tau² = 0.01; 8 estimates). In the case of labour inputs (SMD = −0.045, 95% CI = −0.228, 0.138; I ² = 0%, Tau² = 0.00; 3 estimates) and employment (SMD = 0.003, 95% CI = −0.068, 0.074; I ² = 73%, Tau² = 0.01; 7 estimates), the findings suggested no effects, which mirrors the estimated null effect on time use. In other words, it does not appear that farm households typically draw on more labour inputs, either internally through increasing women's work or externally by hiring labour, to support the changes in practices being promoted.

The findings for market outcome subdomains (Figure 16) suggested improvements, on average, in market access (SMD = 0.118, 95% CI = −0.011, 0.247; I ² = 82%, Tau² = 0.02; 6 estimates) and access to more sustainable incomes (SMD = 0.073, 95% CI = 0.009, 0.137; I ² = 88%, Tau² = 0.02; 31 estimates). However, the effect of interventions on more favourable prices were small and inconsistent (SMD = 0.045, 95% CI = −0.029, 0.119; I ² = 72%, Tau² = 0.01; 11 estimates). Only three studies measured changes in bargaining or quality standards.

OPEN IN VIEWER

Figure 16

Meta-analysis by market outcome subdomains.

The pooled intervention effects for economic outcome subdomains were also estimated (Figure 17). The meta-analysis indicated that the interventions improved productivity by a medium and highly significant effect (SMD = 0.137, 95% CI = 0.069, 0.205; I ² = 87%, Tau² = 0.02; 20 estimates). The interventions also tended to improve income and expenditure itself, albeit with high heterogeneity (SMD = 0.063, 95% CI = −0.002, 0.129; I ² = 85%, Tau² = 0.01; 27 estimates). The effects on asset holdings (SMD = 0.047, 95% CI = −0.002, 0.097; I ² = 67%, Tau² = 0.01; 19 estimates) and savings (SMD = 0.027, 95% CI = −0.029, 0.119; I ² = 47%, Tau² = 0.05; 3 estimates) were on average small and inconsistent.

OPEN IN VIEWER

Figure 17

Meta-analysis by economic outcome subdomains.

Finally, we turn to the domains most closely related to women's empowerment. Figure 18 presents the time use subdomains, which indicate no effects, on average, on leisure (SMD = −0.014, 95% CI = −0.073, 0.044; I ² = 0%, Tau² = 0.00; 3 estimates) or workloads (SMD = −0.011, 95% CI = −0.059, 0.037, I ² = 36%, Tau² = 0.00; 10 estimates). There is small heterogeneity in effects on workloads, as a single study did find a moderate-sized and significant increase in workload for women (Cavatassi et al., 2019) as a result of a fishery community development project.

OPEN IN VIEWER

Figure 18

Meta-analysis by time use subdomains.

However, the intervention effects on other measures of women's empowerment were more positive, although they tended to be of small magnitude (Figure 19). Thus, there were positive intervention effects on women's decision-making (SMD = 0.059, 95% CI = 0.005, 0.114, I ² = 72%, Tau² = 0.01; 17 estimates) and empowerment (e.g., money management) (SMD = 0.091, 95% CI = −0.001, 0.183, I ² = 73%, Tau² = 0.01; 7 estimates).

OPEN IN VIEWER

Figure 19

Impact of interventions on women's empowerment.

While the meta-analysis of the few studies measuring women's leadership also suggested moderately positive impacts (SMD = 0.213, 95% CI = 0.126, 0.300, I ² = 0%, Tau² = 0.00; 3 estimates), the story was less clear for participation, where the average effect was of a similar magnitude but not statistically significant due to great heterogeneity in the estimates (SMD = 0.227, 95% CI = −0.094, 0.548, I ² = 99%, Tau² = 0.22; 9 estimates). The meta-analysis suggested there may be small effects on mobility on average (SMD = 0.058, 95% CI = −0.060, 0.177, I ² = 73%, Tau² = 0.01; 3 estimates). These were due to the positive effect found in a single study of the Indian NDP-I programme on needing permission to visit friends, doctors, attend village meetings or purchase household items, but the findings were not consistent across other studies.

Publication bias assessment

We assessed whether there might be small study effects indicative of publication bias, which would inter alia, because the estimated pooled effects tend to be upwards-biased. The publication bias assessments, carried out by outcome domain and subdomain, did not suggest that small study effects were present for the studies incorporated in this review (Figure 20). The funnel graphs were symmetrical at all values of the standard error, and the Egger regression lines were estimated to pass through, or near to, the origin for outcome domain (p > 0.66) and subdomain (p > 0.61).

OPEN IN VIEWER

Figure 20

Funnel graphs by outcome domain and outcome subdomain.

Moderator analysis

In the previous section we analysed the impacts of value chain interventions implemented in different countries and by different projects. Results, however, may differ in different contexts and by type of intervention. In this section we discuss the results of a moderator analysis (Supporting Information: Appendix F) in which we disaggregate the impact observed in the previous section by geographic area of implementation, by project type, and by whether the intervention was targeted at women.

In our first moderator analysis we consider five broad geographic areas: East Asia and the Pacific, Europe and Central Asia, Latin America and the Caribbean, South Asia, and Sub-Saharan Africa. The majority of available studies come from South Asia and Sub-Saharan Africa, while few report results from Europe and Latin America. The overall effects in South Asia and Sub-Saharan Africa are therefore estimated more precisely than others, and the results of our cross-area comparisons should be interpreted with some caution.

Only those projects implemented in South Asia and Sub-Saharan Africa show consistently positive results across all outcome categories. Results appear to be higher on average in Sub-Saharan Africa than in South Asia with respect to knowledge, input use, and market access. Overall impacts on empowerment outcomes and economic outcomes, however, are nearly identical in South Asia and Sub-Saharan Africa.

In our second moderator analysis, we consider separately targeted and untargeted interventions. Some interventions were specifically designed to provide services to women and to address constraints preventing them from accessing profitable markets. We compare the results of these interventions to those obtained by value chain interventions that were not specifically targeted to women.

Targeted and untargeted interventions do not appear to produce different effects on knowledge, input use, market access, and economic benefits. However, the inspection of forest plots suggests that targeted interventions have a slightly larger impact on women's time use and empowerment indicators. Targeted interventions appear to improve women's non-economic empowerment without creating larger economic benefits. Possibly, targeted interventions improve women's ability to make decisions, mobility, and participation through changes in gender attitudes, rather than through changes in economic power within the household. There are empowerment benefits of the interventions that are not directly brought about by an improvement in women's economic conditions.

Our third moderator analysis considered different types of value chain interventions operating under different mechanisms. Some interventions, such as contract farming, promote vertical integration and greater access to profitable markets. Other interventions, such as those promoting farmer-based organisations, promote horizontal integration and more equitable access to markets. Finally, other interventions promote production or processing upgrading by creating access to products and services that are more profitable.

We considered the differential impacts of these types of interventions on all outcome categories, but we were not able to find any significant differences or meaningful patterns. The different intervention categories appear to produce very similar results on all outcome accounts. It is possible that a different classification of the interventions could produce different results, but there is an inherent difficulty in conducting this type of analysis, as the sample of available studies cannot be sliced into highly specific intervention types without losing precision and statistical power. Our current results suggest that the outcomes of value chain interventions are independent of the specific type of intervention implemented.

Meta-regression analysis

In the previous section we analysed the differences in study results by geographic area, targeting, and project type, one factor at a time. In this section we consider the differential impact of all factors at the same time using meta-regression analysis (Table 8). We do not conduct a separate analysis for each outcome group, and we present the result of an omnibus analysis considering all outcomes at the same time. The factors included in the regression model fit the data reasonably well, and the model explains 36% of the observed data variance. Indicators of heterogeneity, however, point to a large variability in results.

The results are in agreement with the analysis conducted in the previous section and do not provide additional relevant information. They confirm that interventions implemented in South Asia and Sub-Saharan Africa are on average more successful than those in other areas of the world. They also confirm that interventions explicitly targeted to women do not seem to produce larger economic benefits than non-targeted interventions. Targeted and non-targeted interventions are likely to be very similar, and therefore tend to produce similar results. Many outcomes are reported at the household level, and it is difficult to say whether women benefit more from targeted interventions. It is also possible that targeted interventions occur in more challenging settings where improvements are more difficult to produce. Targeted interventions however appear to produce non-economic empowerment impacts.

Finally, the results show that processing and storage facility interventions, and interventions supporting horizontal integration of producer groups, tend to achieve better results, while interventions promoting financial services, processing techniques, inclusive market system development, and market structure appear to perform worse than the average project. However, given the small number of studies in each category and the variety of mechanisms at play, it is difficult to discern any meaningful pattern in favour of any specific type of intervention from this analysis. We were able to identify a large number of studies because we adopted a broad definition of value chain interventions, but when we disaggregate the studies by intervention type, the number of studies becomes smaller and it is difficult to discern patterns from random variation.

Settings and recruitment/referral mechanisms

The interventions ranged from acquiring certifications, fair trade, co-operatives and associations, self-help groups, farmer research groups and contract farming. They took place primarily in community, institutional, and market settings where the participation and involvement of farmers (primarily women) was gauged through various indices. The majority of studies were set in rural areas, though some towns and urban areas also constituted part of the study.

In most cases, the study area was selected based on a representative sampling done in the area, for which data were provided by the co-operatives, associations, NGOs and experts working on the ground with the intended population. In some studies, sampled organisations were selected from an available list of co-operatives in the chosen geographical area (district, town, village) provided by a regional office (Dohmwirth & Hanisch, 2019; Kent, 2018).

While respondents were recruited based on their voluntary interest, various criteria and recruitment factors were taken into account to ensure a robust study, such as a single gender co-operative or mixed-gender co-operative, number of members in a given group, minimum number of years constituting the members' experience, landowners, and so on (Dohmwirth & Hanisch, 2019; Holmes & Imai, 2019; Lukuyu et al., 2020; Papa et al., 2000). This enabled comparability among cases and in the identification of regions or groups that had higher and/or sustained involvement in certain agricultural activities or processes of the value chain (Othman et al., 2021).

Depending on the circumstances, key informants and stakeholders were identified through purposive sampling as well as snowball sampling. While in some cases, experts and officials on the ground helped to identify potential study participants (Altenbuchner et al., 2018; Asingwire, 2011; Oumer et al., 2014), in other situations, traditional social and religious leaders and elders in the community enabled recruitment (Porter et al., 2014).

Participatory rural appraisal survey results also enable participant recruitment into studies (Oumer et al., 2014). Focus group discussions in process evaluation reports are quite important for data analysis, and allowed authors to construct a sampling frame with good representation of actors from various backgrounds. For instance, in the case of a women's collective action group in Northern Kenya (Coppock, 2013), a focus group discussion made it possible to study 16 settlements in which all groups voluntarily participated in intensive interaction.

Barriers and facilitators to participation

The data suggest that intervention participation was voluntary. In most cases, household and community settings facilitated and sustained the participation for respondents. The main barrier was identified as access to institutional provisions and resources.

The following indicators are argued to have facilitated participation: 1.

Women's membership in groups such as associations (Cuellar-Gomez, 2008; Islam, 2008; Keelson & Nanekum, 2021), co-operatives (Bacon, 2010; Holmes & Imai, 2019; Papa et al., 2000; Pineda, 2019), self-help groups (Arakal & Roy, 2015; Desai & Joshi, 2013; Kumar, 2021; Madheswaran & Dharmadhikary, 2001) facilitated their access to networks, market linkages, peer support, and in general, provided a more inclusive environment to motivate them.

In qualitative synthesis, analysis of barriers and facilitators reveal that the same factor has the potential to be both a barrier as well as the facilitator. For instance, out-migration of men was identified both as a barrier as well as a facilitator for women's emancipation in value chain processes. As the facilitator, the absence of men gave women a chance to take more decisions and have a larger role in market settings; however, it also increased the chore burdens of women, who reported juggling domestic chores and caring for the family. Similarly, although the gendered nature of cropping and cultivation practices empowered women to partake in specific steps of agricultural production, it also restricted their opportunities to go beyond the limited and stereotyped spatial scope of practices and engagement.

Some other identified barriers to women's participation in the value chain processes are discussed below: 1.

Stereotypes conditioned by social norms and culture, which dictate that women ought to perform the primary role of caregiver first. As per the traditional norm, it was considered inappropriate for a woman to be present in the public sphere or hold membership of associations (Cuellar-Gomez, 2008; Lyon, 2008).

Barriers and facilitators to achieving outcomes

Causal processes

Three dominant causal pathways are identified in either facilitating or limiting intervention outcomes. These are related to lack of awareness of rights, illiteracy, out-migration of male members, usage of religious ideology to rationalise gender roles, and the effect of formalisation/professionalisation of the trades in discussion.

Causal processes aimed at gender equity in these reports emphasise the role of capacity building for women's empowerment broadly. Diversified income sources and increased income for women resulted not only in better household living standards and financial agency but also the appreciation and support of husbands and men in general. Changing climatic conditions, loss of biodiversity, and loss of traditional practices due to intense commercialisation were also brought forth.

In addition, we highlight evidence of women's exclusion from the co-operative level and the public sphere, where most associations are led by men – thereby negatively affecting women's freedom to participate or speak their minds:

Lack of access to knowledge and illiteracy, in general, rendered access to services, resources and emancipatory avenues difficult for women. However, a lack of awareness in certain cases led to natural resource degradation – such as in the case of commercialisation of argan in Morocco (Perry, 2020) or organic cotton cultivation in India (wherein farmers resisted participating in the intervention due to the perceived threat of losing indigenous knowledge and practices, further resulting in loss of biodiversity) (Altenbuchner et al., 2018; Mitra, 2019).

How do value chain interventions work?

As emphasised earlier, value chains are a result of the matrix of functions and channels that contribute toward the outcomes and overall impact for the stakeholders involved. In this review, we find that horizontal integration works in favour of the efficacy of any value chain. The qualitative synthesis found that interventions promoting horizontal integration through farmer-based organisations, self-help groups and co-operatives tend to support in promotion of women's participation in, and access to economic benefits from, the interventions.

Why do value chain interventions work or not work?

Understanding the causal pathway is useful in responding to this question (Figure 21). These pathways show how women's engagement from production to distribution influence their economic empowerment. The concept of economic empowerment broadly encompasses access to resources, availability of finance, and agency of the participants. McKague and Siddiquee (2014) state that constraints in successful implementation as well as uptake of value chain interventions not only arise from business enabling environment but also physical infrastructure, human capabilities and social norms.

OPEN IN VIEWER

Figure 21

Revised theory of change.

Three dominant causal pathways are identified in either facilitating or limiting the intervention outcomes. These are related to unawareness of ones' rights, illiteracy, out-migration of male members and effect of formalisation/professionalisation of the trades in discussion (e.g., Elias, 2017). Causal processes aimed at gender equity in these reports emphasise upon the role of capacity building for women empowerment broadly. While diversified income sources and increased income for women resulted not only in better household living standards, financial agency, but it also garnered appreciation and support of women husbands and male folk in general. Changing climatic conditions, loss of biodiversity and loss of traditional practices caused due to intense commercialisation were also brought forth.

McKague and Siddiquee (2014, p. 47) argue that for value chain interventions to work, it is important that a ‘sufficient number of value chain actors find it in their self-interest to implement and maintain them’, it has to be context driven. The authors in their work identify the unique challenges that emerge in a given place due to its specific market structure, prevalent social relations, business enabling environment, and customer demands.

The ToC in this review analyses the elements of value chain interventions at the micro and macro level to offer an overview the intermediate and final outcomes in shaping the economic empowerment of women participants in the value chain(s) as the desired impact. Macro level elements encompass the cultural, institutional, geographic, and economic characteristic, that is, the normative structures (norms, values, cultures) along with the prevalent economic and business ecosystem. The micro level elements include business capabilities, resources availability, financial literacy, cognitive capabilities, and so forth.

These are addressed by inputs such as training on improved production practices, access to finance, providing market linkages and capacity building measures including infrastructure investments. Activities such as providing training to farmers and other businesses, facilitate access to finance for farmers and other businesses, linking farmers and other businesses to high-value markets and investment in infrastructure, facilitate in achieving the outputs of both increased productions well as improved efficiency of the value chain overall.

Intermediate outcomes such as income generation from the chain with enhanced mobility for women are of significance in sustained economic empowerment for women. But this is closely related to and shaped by women's access and ownership of assets. Her agency, finance, utilisation of and access to resources therefore are three main elements in understanding the gendered participation in value chain and collective enterprises. Economic empowerment for women can be best achieved when women have the power to make their own decisions and choices, based on their own knowledge and understanding, to improve their social status.

To this end, unpaid household labour, mostly done by women, limits women's time and energy to earn money through value chains (Stoian et al., 2018). The factor of time poverty is reported as one of the main reasons for hampering women's participation and empowerment in value chain(s). This further dampens their control over assets than men, reducing their decision-making power and capacity to engage in more profitable nodes of value chains (Quisumbing et al., 2015).

Most of the evidence in this review is included from African regions. Those interventions implemented in Sub-Saharan Africa and South Asia are consistently more successful for all outcomes considered.