In Zambia, smallholder agriculture forms the backbone of rural livelihoods and national food security, yet it remains predominantly traditional and low in productivity. The majority of farmers rely on rudimentary tools such as hand hoes for land preparation, depend on family labor, and utilize minimal modern inputs—a situation that reflects structural inefficiencies within the sector (Mason et al., 2020; Sibhatu et al., 2022). Maize is the principal crop cultivated nationwide, playing a critical role as both a staple food and a source of income (Silva et al., 2023; Workman, 2018). However, Zambia’s agricultural sector is increasingly vulnerable to multifaceted challenges, including threats from climate change, land degradation, diminishing soil fertility, and prevalent pests and diseases. For instance, during the 2017/18 farming season, maize production in Zambia amounted to 2.4 million tons, marking a decline of 34% from the previous year and 20% from the 5-year crop record. This production level was 14% lower than the country’s consumption needs (Sihlobo, 2018). Additionally, during the 2023/24 farming season, the government declared a national disaster due to a severe drought that drastically impacted food production and electricity generation. This drought, the worst since 1981, led to the destruction of crops, including maize along with livestock losses, exacerbating poverty and food insecurity for more than 9.8 million people (WFP, 2024).

Among the various initiatives aimed at addressing Zambia’s agricultural vulnerabilities, the MAB project stands out for its targeted strategies designed to transition smallholder farmers from subsistence to commercially viable agriculture. These two key approaches include: (i) forming common interest groups, such as self-help groups and savings groups, to enable community members to collaborate in improving the cultivation and marketing of various commodities, and (ii) establishing farmer-training centers (demo sites) to facilitate capacity building and skills transfer. Participation in these savings groups is voluntary, with each group limited to a maximum of 30 members. Notably, all MAB groups adhere to a constitution governing their operations, including aspects such as leadership selection, term of office, loan interest rates, and loan management. These groups determine their meeting schedules and frequencies within a month.

Given the challenges associated with accessing formal credit, RUSACCOs in project areas have the potential to empower women participants in maize production and address fluctuating production levels, which are believed to contribute to Zambia’s classification as the third most hungry nation globally in 2015 (GHI, 2016) and among the most hungry nations in 2020 and 2023.

The study primarily focused on engaging smallholder farmers in the four districts where the MAB project operates, gathering data between July and September 2022. Ethical clearance for the study was obtained from Prince G Consultancy and Academy, with all participants providing voluntary consent. Maize production quantities, costs, and input prices were captured for the 2021/2022 farming season, while the recall period for other key variables is detailed in section 3.3.

Employing a multi-stage sampling technique aimed at mitigating program placement bias, two agricultural villages with comparable socio-economic conditions and infrastructure were purposively chosen in each district—Chisamba, Kapiri-Mposhi, Kabwe, and Mumbwa. Within the villages, two groups were delineated based on MAB activities, separated by approximately 35–40 kilometers: areas with active MAB participation were categorized as “vibrant RUSACCOs areas,” while those lacking such activities were labeled “inactive RUSACCOs areas.” This approach, recommended by Tambo and Wünscher (2018), aimed to address program placement bias. The sample size was determined following Cochran (1977) criteria for known populations (N = 3,600) presented in Equation 1 at a 95% confidence interval, resulting in an expected sample size of 450 participants. However, to account for program placement bias, 300 from vibrant RUSACCO areas and 150 from inactive RUSACCO area were selected and are inclusive of both participants and non-participants as shown in Figure 2. The final analysis utilized 436 observations due to missing data for 14 respondents.

Where 𝑛 = Sample size, 𝑁 = Population size, 𝑍 = Z-value (1.96), 𝑝 = Estimated proportion of an attribute present in the population (50%) and 𝐸 = Margin of error (5%).

Data collection employed a structured and pretested questionnaire administered by experienced enumerators to minimize measurement error. The enumerators primarily targeted the household head and their spouse for data collection to ensure comprehensive and accurate responses. Since most households cultivate a single plot of land, the agricultural input and output data collected specifically pertain to the household’s maize plot.

Triangulation was achieved through key informant interviews (KIIs) with various stakeholders, including project staff, agronomists, training center community facilitators, and lead farmers. Recognizing the susceptibility of EE estimates to measurement errors, a rigorous data cleaning protocol was implemented. Outliers were identified and verified against field notes, and cross-validation with original survey forms ensured transcription accuracy. Range checks were applied to key variables, and enumerators received extensive training to standardize data collection techniques. Robustness checks confirmed that the results were stable even after excluding potentially erroneous observations. To mitigate potential data recall bias, information from the immediate past farming season was collected, aligning with recommendations by Tarrant et al. (1993) and Connelly et al. (2000). Additionally, questions were framed with specific seasonal references, and wherever feasible, data were triangulated with secondary sources to mitigate recall bias.

Summary statistics of the sample are presented in Table 1. It is important to note that family labor was quantified by aggregating the self-reported hours spent by household members on maize farming activities, including land preparation, planting, weeding, and harvesting. To value this labor, we used the prevailing regional wage rate for agricultural labor. This approach ensures that the opportunity cost of family labor is reflected in the analysis and aligns with common practices in agricultural productivity research.

The main explanatory variable of interest, WP in RUSACCO, is defined as a dummy variable measured at the household level. It takes the value of 1 if at least one adult woman in the household has participated in RUSACCO for two consecutive years, including the survey year, and 0 otherwise.

For the potential pathway analysis, three variables were considered: AgI, women’s empowerment and, labor-land ratio. AgI is captured as a dummy variable, with one representing a household where an adult member used digital tools to support agricultural processes in the survey year, and 0 otherwise. On the other hand, women’s empowerment is measured in terms of the proportion of short- and medium-term productive assets in monetary value owned by women or jointly by male and female household members within a household. This concept of asset ownership is commonly utilized in the literature to assess the economic situation of women within households. To mitigate potential issues of reverse causality, only assets such as agricultural equipment (hoes, saws, wheelbarrows, sprayers, etc.) and vehicles (bikes, motorbikes, trucks, etc.) were considered, excluding highly durable assets like land or buildings. Thus, a larger proportion of such assets owned by females or jointly owned by male and female household members indicates a higher degree of women’s empowerment (Sekabira and Qaim, 2017a). Lastly, labor-land ratio is the quotient of the labor-hours by land area which reveals the amount of labor hours per hectare.

Finally, EE serves as the outcome variable estimated through stochastic frontier analysis (SFA).

To achieve the main objective of this study, which is to evaluate the effects and underlying mechanisms of women’s participation in RUSACCOs on economic efficiency, we employ a rigorous and multifaceted empirical strategy. First, we apply a sample selection SFA, combining propensity score matching (PSM) and stochastic cost frontier (SCP) techniques, to assess the relationship between WP and EE. The use of these advanced econometric techniques ensures reliable and robust results, providing deeper insights into this critical relationship. Subsequently, we explore potential pathways using conditional mixed process (CMP) modeling and structural equation modeling (SEM). To enhance the robustness of our findings, we employ the pooled and matched sample. Additionally, emerging themes from key informant interviews were incorporated to corroborate and enrich the empirical results, providing a more nuanced understanding of the findings.

Figure 3 illustrates that the proportion of WP in RUSACCOs is approximately 63%. In more vibrant areas, the proportion of WP households is about 10% higher compared to inactive areas. Conversely, inactive areas exhibit lower levels of WP, with around 10% more households lacking WP than their vibrant counterparts. Despite the pooled WP exceeding the 50% threshold, key informants highlighted that the persistent prevalence of credit constraints among participants necessitates targeted efforts to enhance WP. They pointed out that while the overall participation levels are promising, many women still encounter significant barriers to fully accessing and benefiting from credit services. “Sometimes we attend the meetings, but getting a loan is still difficult because we lack what they ask for as security,” explained one woman farmer, reflecting the structural barriers that limit financial inclusion. Additionally, key informants unanimously emphasized that addressing these challenges requires tailored interventions, such as improving access to financial resources, enhancing financial literacy, and ensuring that RUSACCOs adopt inclusive and equitable practices. These insights align with the study’s findings and existing literature, underscoring the necessity of fostering a more enabling environment to maximize the benefits of WP in RUSACCOs (Kumar et al., 2021; Rickard, 2022).

The first stage of the sample selection process involves conducting a probit regression to analyze the factors influencing WP. For robustness checking, we present results from both the pooled and the matched samples. It is crucial to validate the quality of the matching to ensure that covariates are balanced, common support is established, and bias reduction is achieved, as illustrated in Figures 4, 5. The successful balancing of covariates and the reduction of bias through PSM confirm the robustness of our matching process and provide confidence in the validity of our results. A balanced distribution of covariates indicates that the matching process effectively mitigated any systematic differences between the two groups, thereby enhancing the comparability of treatment effects (Caliendo and Kopeinig, 2008; Leuven and Sianesi, 2018).

Following the confirmation of satisfactory matching quality, as recommended by Caliendo and Kopeinig (2008), we present the factors influencing WP in RUSACCOs in Table 2 for both the matched subsample and the pooled sample. As matching adjusts for bias arising from observed characteristics, model diagnostics at the bottom of Table 2 indicate that the matched sample offers improved classification accuracy relative to the pooled sample. Nevertheless, the pooled results reinforce the robustness of the findings, demonstrating consistency in the estimated effects even in the absence of matching controls.

Regarding the determinants of WP in RUSACCOs, results from both models indicate that households headed by older individuals are less likely to exhibit WP. One possible explanation is that older household heads may place greater caregiving responsibilities on women, particularly in supporting elderly family members, thereby limiting women’s time and mobility to participate in cooperative activities (Glazebrook et al., 2020). Additionally, in both models, attaining basic education and training positively influences WP. This finding aligns with the agricultural literature, which highlights the role of education in equipping individuals with the skills and knowledge necessary to adopt innovative financing mechanisms like RUSACCOs (Gouët and Van Paassen, 2012). Education likely enhances women’s understanding of the benefits and operational dynamics of these cooperatives, fostering their participation. Furthermore, larger households are more likely to have WP in RUSACCOs. This could be due to greater labor division in larger households, which allows women to dedicate more time to cooperative involvement. It may also reflect the need for additional income sources in larger families, motivating women to seek financial opportunities through RUSACCO participation.

For the matched sample, we find that longer distances to markets have less influence on WP. This finding may be attributed to the role markets play as hubs for information exchange among rural women, offering insights into innovative agricultural practices and financing opportunities. However, for women who already access alternative sources of information, such as cooperatives or community networks, proximity to markets may become a less critical factor influencing their participation (Kropf et al., 2007; Tadesse and Bahiigwa, 2015).

Finally, prior knowledge of RUSACCO operations emerges as a significant determinant of WP, reinforcing its relevance as an instrumental variable. This indicates that awareness and understanding of the structure, functions, and potential benefits of RUSACCOs substantially enhance women’s likelihood of participation. Consequently, interventions that prioritize the dissemination of accurate and accessible information about cooperative operations may be essential in fostering greater women’s engagement in RUSACCOs.

Table 3 presents both the association and the estimated causal effect of WP on EE, using three different estimation strategies to ensure robustness and clarity in interpretation. However, it is important to acknowledge that while pre-matching estimates provide initial insights, they may be biased due to differences in observable characteristics between treatment and control groups. Post-matching estimates reduce this bias by improving comparability; however, they may still be subject to unobserved heterogeneity. This potential bias is reflected in the discrepancies observed in the estimated mean, minimum, and maximum values of EE reported at the bottom of the table. Given these limitations, greater confidence is placed in the estimates derived from the sample selection framework, which accounts for both observed and unobserved factors influencing selection into WP.

In the frontier model, we find that all inputs and the quantity of maize output produced positively and significantly influence the cost of production. Notably, across all model specifications, inefficiency accounts for variation in production costs, as evidenced by a lambda value greater than one (Adom and Adams, 2020; Bravo-Ureta et al., 2007). Regarding our primary inquiry, we find that WP in RUSACCOs in both the pooled and matched samples negatively and significantly influences economic inefficiency. This is further confirmed by the mean EE values of the sample selection for households with and without WP, showing an approximate 21% efficiency gap in favor of households with WP. This gap is higher than than found in Appiah-Twumasi et al. (2022) and Mustafa et al. (2024). The efficiency gains observed among women participating in RUSACCOs are likely linked to improved financial planning and input acquisition. This is reinforced by qualitative insights; one participant explained that “because I can now save and borrow from the cooperative, I plan better for each farming season.” Another highlighted that: “cooperative membership allowed me to avoid disruptions during planting—I do not have to wait anymore for someone to lend me money.” These reflections underscore how financial stability rather than technical support contributes to the significant efficiency gap observed in favor of households with WP.

In terms of the distribution of EE between households with WP and their counterparts, Figure 6 presents histograms and kernel density distributions to illustrate regions of concentration. Peaks in the distributions signify areas where the data are more concentrated, while troughs represent regions with less concentration. Analyzing the kernel density plots, it is important to note that a wider curve indicates higher variability, while a narrower curve suggests lower variability. Peaks in the density plot correspond to modes in the data distribution.

In general, households without WP exhibit a symmetric EE distribution while that of households with WP is skewed toward the 100%. While there is relatively more variability among households without FP, their counterparts are more concentrated around 90%. Households with WP display a unimodal distribution that is notably higher compared to the bimodal distribution among their counterparts. This suggests a higher level of consistency in TE among households with WP than those without. Importantly, the analysis acknowledges a methodological limitation linked to the valuation of family labor. Although the use of the prevailing regional wage rate is a standard practice in agricultural efficiency studies, it introduces variability due to regional wage dispersion and informal labor dynamics. Additionally, reliance on self-reported hours for family labor inputs may result in recall bias, particularly in cases where farmers do not systematically record working hours. While mitigation efforts (e.g., time-bound recall protocols) were employed during data collection, this limitation warrants consideration in interpreting efficiency estimates.

While direct WP can trigger improvements in EE, it is also important to investigate other pathways in alignment with the implementation of the MAB program. To comprehensively examine the potential pathways underlying the WP-EE link, Table 4 presents the estimates of the CMP model. The negative and statistically significant correlation coefficient between the error terms, Rho, indicates the presence of negative selection bias. This validates the appropriateness of using CMP model to estimate the mediating effects of AgI reliably. Furthermore, it suggests that farmers’ decisions to adopt AgI are influenced by unobserved factors, such as innate farming talent or risk tolerance. Given the limited access to diverse extension services, farmers often mitigate risks by incorporating innovative practices, particularly in accessing information and financial solutions.

In the first equation of Table 4, WP in RUSACCOs is found to significantly enhance the likelihood of AgI. This is consistent with the understanding that access to financial services, particularly savings and credit, can ease liquidity constraints and enable women to invest in technologies that improve their farming practices (Amponsah et al., 2023; Mwalupaso et al., 2025). Through improved financial stability, women are more able to acquire mobile phones, radios, or even basic agri-informational services, laying the groundwork for digital adoption. On the other hand, the second equation focuses on the role of AgI in influencing EE, with results indicating a positive and statistically significant relationship. This confirms the mediating role of digital engagement between cooperative participation and improved efficiency outcomes. Such findings reinforce the argument that digital technologies, when made accessible and relevant, serve as catalysts for enhancing farm-level decision-making, market responsiveness, and resource optimization (Agyekumhene et al., 2018; Fabregas et al., 2019; Jenny, 2008).

However, these interpretations must be considered in light of certain methodological constraints. First, the measure of AgI is based on a binary variable indicating whether any adult in the household used digital tools. This approach, while pragmatic, arguably compresses a spectrum of digital behaviors into a single indicator. It does not account for the frequency, purpose, or complexity of tool usage, dimensions that can shape the impact of AgI on efficiency. For instance, using a phone to check rainfall updates occasionally is not equivalent to regular engagement with market platforms or weather forecasting tools (Tadesse and Bahiigwa, 2015). The measure thus captures adoption, but not depth or quality of usage. Additionally, the variable may conceal intra-household disparities. In households where men dominate tool access or decision-making, the recorded use might not reflect women’s empowerment or their ability to benefit directly from informatization.

Despite these limitations, qualitative feedback from women participants provides compelling evidence of the instrumental role that WP via digital access plays. One woman emphasized that “having access to savings and loans means I can buy inputs on time and not wait until the rains have started.” Another remarked that “after accessing credit, I now use my radio and phone to make my own farming plans which is helpful.” These reflections demonstrate how both access to financial services and digital tools are empowering women to make timely and independent agricultural decisions, thereby improving efficiency. Taken together, the findings suggest that RUSACCOs can serve not only as financial intermediaries but also as critical enablers of digital inclusion—provided that interventions recognize the diverse realities of women farmers.

Although digital tools like mobile phones and radios have penetrated many rural areas, their effective use still depends on literacy, digital skills, and infrastructure. Thus, scaling AgI as a mechanism for improving economic efficiency requires addressing educational and infrastructural barriers, especially among women. Recognizing these constraints is crucial for designing inclusive policies that leverage RUSACCOs not only as financial institutions but also as platforms for digital empowerment.

Beyond understanding the mediating effect of AgI, it is crucial to examine the pathway through which AgI influences EE. AgI has the potential to trigger various processes that improve EE. To explore this, we employed the CMP model with three equations consistent with the empirical strategy and the results are presented in Table 5.

The first equation of Table 5 reveals that AgI positively and significantly influences women’s empowerment, which some scholars also use as a proxy for gender equality consistent with the finding in Sekabira and Qaim (2017a). As households adopt digital tools for agricultural activities, women gain greater access to productive assets, such as financial resources, knowledge, and technology. This empowerment enhances their capacity to contribute meaningfully to agricultural processes.

Women’s empowerment, in turn, is strongly associated with an improved labor-land ratio consistent with Rasheed et al. (2020) and Vemireddy and Choudhary (2021). As women become more empowered, their active participation in agricultural production increases, leading to a better allocation of family labor relative to land use. This relationship highlights how gender equality and resource accessibility, facilitated by AgI, drive structural improvements in farm management practices.

Finally, our analysis establishes a robust link between the labor-land ratio and EE in maize production. A more balanced labor-land ratio, driven by increased women involvement, supports cost-efficient farming practices, improving overall EE. These findings affirm that WP in RUSACCOs enhances EE through a chain of effects: AgI fosters women’s empowerment, which improves the labor-land ratio, ultimately boosting EE. This is also confirmed by our robustness checking estimates from the SEM provided in Table A1.

The findings of this study have significant policy implications for agricultural value chains. Promoting WP in RUSACCOs can enhance agricultural informatization, gender equality, and women’s empowerment in agriculture. By ensuring equitable access to agricultural assets, these policies support a more balanced and sustainable agricultural value chain. This empowerment not only improves individual productivity but also strengthens the overall resilience of the agricultural sector.

With only 6 years remaining until the deadline for the SDGs, it is crucial to maximize the contributions of women, who are often marginalized within the agricultural sector. Economically, WP could facilitate poverty alleviation through more efficiently produced maize, consistent with the SDGs (Adam et al., 2020; Tambo et al., 2021). For example, policy-relevant calculations demonstrate that if households without WP adopt it, they could close the efficiency gap and potentially save USD 65.64 annually in production costs (Table 6). This translates to daily savings of USD 0.18, which, given the new international poverty line of $2.15 per day, could result in an 8% reduction in extreme poverty. This impact is particularly significant in developing countries like Zambia.

Moreover, closing the efficiency gap could facilitate the production of more than 487 kg of maize per household, equating to a daily gain of 1,336 grams. Considering the average global daily food supply quantity of maize over the last 5 years, as reported by the Food and Agriculture Organization (FAO), of 49.23 grams per day, these gains would significantly improve maize supply, especially in Africa where maize consumption is high.

Given the implications for poverty reduction, governments and development organizations can create a more inclusive and efficient agricultural value chain by implementing six key policies and practices that empower women and enhance economic productivity.

In summary, fostering WP in RUSACCOs not only boosts agricultural productivity but also contributes to broader socio-economic development. Targeted policies and interventions that promote gender-inclusive financial and agricultural practices are essential for enhancing the efficiency and sustainability of agricultural value chains, ultimately supporting poverty alleviation and food security.