Gendered investment differences among smallholder farmers: evidence from a microcredit programme in western kenya

The advent of microcredit programmes in sub-Saharan Africa provides opportunities for rural households to acquire agricultural inputs and consumer goods. This study analysed gender differences in investment behaviour and repayment performance using a unique dataset—the complete client database (21,386 clients) of a microcredit programme operating in Western Kenya. Products purchased via the microcredit programme include seeds, fertilisers, post-harvesting technologies (drying sheets, storage bags, and pesticides), chicken feed packages, and different solar panel products. A machine learning-based basket analysis identified combinations of products purchased by male and female clients. Our results showed that female farmers usually made smaller investments, had higher repayment rates, and purchased more post-harvesting technologies than male farmers. In addition, female farmers used their loans to purchase less expensive products, whereas male farmers usually purchased more fertiliser and expensive solar panel products. The basket analysis revealed that female farmers purchased multiple products simultaneously more often than male farmers did. Finally, households without mobile phones had low repayment capabilities. Collectively, our findings show that microcredit programmes serving smallholder farmers can capitalise on their business data to learn about their clients’ gendered investment preferences and repayment behaviour.


Introduction
The gender gap in smallholder agriculture is a primary concern for development and food security in Sub-Saharan Africa (SSA) (Gebre et al., 2019;Tavenner et al., 2020).While female farmers comprise the majority in many rural areas, their productive capacity is often constrained by gendered differences in access to resources, especially land and capital (Gawaya, 2008;Palacios-Lopez et al., 2017).Under these circumstances, female farmers often apply less fertiliser to smaller plots (Marenya & Barrett, 2009) and typically realise 20-30% lower yields than their male counterparts (World Bank, 2009, 2015).This 'gendered yield gap' is not only perpetuated by gendered divisions of labour in rural areas and persistent social norms but also by poorer market opportunities, restricted access to agricultural technologies, and limited access to credit (Gebre et al., 2019).
The advent of private microcredit initiatives in the agricultural sector in SSA enables resource-constrained smallholder farmers, including women, to access agricultural technologies and inputs on credit and improve their households' food security situation (Benjamin et al., 2020;Dossou et al., 2020;Miled & Rejeb, 2015).Understanding farmers' preferences and investment behaviour is critical for microcredit programmes, not only for the compilation of sought-after input packages but also from a loan repayment perspective.Loan default is a major concern for microcredit providers (Banerjee, 2013).Many previous studies have affirmed that female borrowers repay their Keiji Jindo contributed equally to this work.loans more regularly than male borrowers (Aggarwal et al., 2015;Armendariz & Labie, 2011;D'Espallier et al., 2011D'Espallier et al., , 2013)).Mechanisms underlying the decision-making and repayment performance of credit program participants have been described by D 'Espallier et al. (2011'Espallier et al. ( , 2013)), who found that the higher repayment rates of female borrowers were at least partly attributable to specific circumstances faced by women in developed countries.For example, female borrowers in rural areas often have limited mobility and restricted financial access, and thus they are often subjected to social pressure as an enforcement mechanism to prevent delayed repayment, especially when the number of female participants is high (D'Espallier et al., 2011;Mia et al., 2021).Additionally, women are more susceptible to the peer pressure associated with joint liability loans, which can be seen as an alternative form of coercive enforcement (Morduch, 1999).Another underlying mechanism that may partly explain the better repayment performance of women is that microcredit institutions often actively implement a more personalised, tailor-made approach adapted to the needs of women (D'Espallier et al., 2011).
It should be noted that microfinance has gradually moved away from the traditional approach of collective lending towards various forms of individual credit that rely on alternative mechanisms instead of joint liability to ensure loan repayment.Understanding the motivations of individual microcredit clients and the factors contributing to successful management by microfinance institutions (MFIs) is crucial to understanding the behaviours exhibited by microcredit clients (Dalla Pellegrina et al., 2021a;Naegels et al., 2022), particularly the relationship between investment behaviours and repayment performance.Several significant determinants and underlying mechanisms associated with the operational activities of microcredit exist.For instance, setting explicit and specific goals (e.g., saving) can improve the task performance of microcredit clients, and providing feedback is important for helping individuals trace their progress (Dalla Pellegrina et al., 2021b).To prevent borrower discouragement, it is important to change their negative perceptions about loans by promoting their communication with loan officers and providing transparent information on loan processes, rights, and obligations (Naegeles et al., 2022).Offering trainings about money management and entrepreneurship can contribute to repayment performance (Dalla Pellegrina et al., 2021a).Group meetings are seen as a means of fostering social capital, and a higher frequency of such meetings is linked to superior performance among microcredit groups (Dalla Pellegrina et al., 2021c).A pleasant atmosphere and supportive environment contribute to the motivation of learners (Williams & Williams, 2011).By cosigning loans, as loan guarantors, individuals can expand credit accessibility by increasing the pool of collateral, thereby shifting the risk from the bank to the guarantor (Dalla Pellegrina et al., 2014).
Therefore, this study explores gendered differences in smallholder farmers' input/product package preferences and repayment behaviour by building on a unique dataset of the client database (n = 21,836) from a private microcredit programme led by Agrics Ltd., a social agricultural organization in Western Kenya.Agrics provides credit for the purchase of different agricultural inputs (seeds and fertilisers), chicken feed packages, post-harvest technologies, and solar power technologies.Building on machine learning techniques, such as basket analysis (Tandon et al., 2016), we show that basic client data, such as gender, product preferences, loan size, down payments, and repayment rates, can already provide sufficient information for the development of a viable gender-sensitive business strategy.
This paper is organised as follows.Sections 2.1 to 2.3 provide details on the institutional set-up of the Agrics microcredit programme and the client database that constitutes the core data source analysed in this study.The information on rural household characteristics in our study area, with a specific focus on agriculture, is described in Section 2.4.
In section 3.1, we present the results of gender-based differences in repayment behaviour and input package preferences as found in the Agrics client database.Section 3.2 describes the results of our poverty analysis, with a focus on the investment behaviour of 'low-endowed' farmers, including female farmers who are the heads of their households (Tittonell et al., 2005) or of households lacking access to mobile phones (Wesolowski et al., 2012).

Origins of the Agrics microcredit programme
The formation of the social agricultural programme Agrics by the Dutch non-governmental organisation International Child Support (ICS) was informed by the results of a nudging experimental study by Duflo et al. (2008Duflo et al. ( , 2011)).This study examined the dynamics of farming households' liquidity positions in the agricultural season.Farmers participating in the Agric programme were mainly from the Western Kenya Province (Busia, Bungoma, Kakamega, and Vihiga Counties), with smaller numbers located in the Nyanza (Siaya County) and Rift Valley Provinces (Nandi, Trans Nzoia, and Uasin Gishu Counties).
Agrics was not the only microcredit programme operating in Western Kenya at the time of this study.Many smallholder farmers in the area were registered with microcredit organisations such as the One Acre Fund, SmepBank, Faulu, and Kenyan Women Microfinance Bank.In Kenya, the Microfinance Act was passed in 2006 and commenced its operations in 2008 (Wijesiri & Meoli, 2015).The legislation of the Microfinance Act provides directives for MFIs to strengthen corporate governance standards, safeguard depositors, comply with necessary capital prerequisites, promote competition for enhanced effectiveness, and carry out their activities with care and expertise.
Agrics started operating in the early 2010s, providing microcredit to farmers during the long rainy season.The Agrics microcredit scheme for agricultural inputs recruited new clients immediately after the harvest season, when farmers still had cash from crop sales.In 2017, the enrolment of farmers started in October and November, after the end of the 2016 long-rain season.This study analysed the Agrics client database for the 2017 long-rain season.

Organisation of the Agrics credit programme
Agrics farmers were enrolled before a new cropping season, with an advance payment of 10% of the total loan taken out.After evaluating whether there were still pending debts, a farmer was allowed to register as a client.Farmers were individually served by Agrics, which provided seeds, fertilisers, agrochemicals, chicken feed packages, post-harvest products, and solar panel power technologies.Farmers were organised into groups that were frequently visited by the company's extension workers.Each farmer group consisted of 15-20 members, with a group leader under the supervision of a community facilitator, who was a temporary employee of Agrics living in the same local community.After their enrolment was approved by Agrics, farmers were given a training course on saving skills, which are a key components of repayment performance.
During the cropping season, farmer group meetings were conducted weekly, with two main objectives: 1) collecting a small portion of the repayment and 2) sharing information with the farmers about Agrics microcredit schemes, field management, agricultural and post-harvesting technologies, and value chain participation.The Agrics community facilitator living in the same community as the farmer group members acted as a mediator between the client group and the executive officers of Agrics on agricultural matters.They revised together with the group leaders the progress of the repayment performances and gave feedback to the farmers.They communicated with Agrics office members not only about problems related to the microcredit programme but also to the farming practices and local issues at the community and individual farm levels (e.g., pest and disease problems, drought, and insecurity caused by local militia guerrillas).By the end of the cropping season (i.e., in August and September), the costs of the purchased Agrics products had to be repaid.Each Agrics client with a mobile phone (>95% of all clients) repaid their loan using the M-PESA mobile money platform.Group leaders were given prepaid minutes, commonly known as "airtime", on mobile phones to reach out to farmers.Farmers received both calls and short-message service (SMS) texts inviting them to attend trainings and meetings.Clients who did not own a mobile phone made their payments in cash directly to their group leader or community facilitator.Ideally, the repayment was expected to be completed by the end of September, and as part of the intervention, group leaders and community facilitators conducted home visits to motivate and convince farmers with outstanding debt to make their payments.Agrics clients appeared to represent Western Kenya's farming population concerning the plot sizes they cultivated and the maize yields they obtained.More information on Agrics was presented by Langeveld and Quist-Wessel (2015).

The Agrics client database
In this study, we used the Agrics client database (Table 1), which comprises 21,386 individual client records for farmers (7,842 male and 13,544 female farmers) from the highland to lowland counties in Western Kenya.The database included information on the items purchased on the loan, the total value of the loan, repayments, and the advance payments made for the next season (Table 1).The types of products bought by the farmers and the price of each product were also recorded.The following paragraphs describe these products.

Input packages and premium seeds
The packages of maize seeds and chemical fertilisers (diammonium phosphate [DAP] + calcium ammonium nitrate [CAN]) are available in different sizes from 0.25 acres onward.The maize seeds in the standard input packages sold by Agrics were products of Kenya Seed Company Ltd., which provided a wide range of varieties for different agroecological zones (highlands to lowlands).However, some clients preferred other, more expensive varieties, referred to as 'premium seeds', which required additional payments.Premium seeds were maize varieties from Western Seeds, SeedCo, Pannar, and Pioneer.

Other products
Agrics also offered post-harvest products, non-maize crop seeds (sorghum, common bean, and soybean), solar energy panels, and a chicken feed package.Post-harvest products included insecticides (1.70 United States Dollars (USD)), drying sheets of either nylon (Category 1: 25 USD) or canvas (Category 2: 35 USD), a durable hermetic bag with additional inner plastic lining (3.50 USD), and hermetic aluminium storage silos of two sizes (fitted for 360 and 720 kg of maize grain with a cost of 77.40 and 157.40 USD, respectively).Four categories of solar energy panels were available to farmers: Category 1, 8.50 USD; Category 2, 31 USD; Category 3, 37.85 USD; and Category 4, 74.90 USD.The chicken package (35 USD) included ten chicks, feed, and wires.

General characteristic of farm households in Western Kenya
According to the previous studies (Ojiem et al., 2014;Tittonell et al., 2005;Jindo et al., 2020), two main types of farm households were identified in this region based on agroecological indicators: 'maize mixed farm households', which were dominated by maize and legume crops in lowland regions; and 'highland perennial farm households', which grew food and cash crops such as tea and coffee at higher altitudes.Relatively droughtresistant sugarcane is often cultivated in the middle and lowlands of the Western Kenyan region as a cash crop.In the study area, there was one main cropping period during the long rain season (March to August) and a minor cropping period during the short-rain season (September to January).The potential yield of maize crops, when nutrients and water are not limited, is 8-10 t/ ha for the region under rainfed conditions, whereas actual yields on most smallholder farms in this region are approximately 1.9 t/ ha (van Ittersum et al., 2016).Rural poverty and food insecurity are more prevalent in lowland zones in this region, according to Kenyan government reports.The average farm size in this region is two acres, equivalent to 0.8 ha (Channa et al., 2019).Maize is the main staple crop used for food self-sufficiency (Marinus et al., 2021).The duration of households' food self-sufficiency varies between 4.6 and 10.7 months, depending on farm characteristics (Jindo et al., 2020).The number of livestock measured by tropical livestock units (TLU) ranges from 1.3 to 5.1 TLU (Jindo et al., 2020) in this region.Off-farm income is often used to finance farming activities (Marinus et al., 2022).
The blanket fertiliser recommendation for maize in Kenya is 50 kg/acre (123.55 kg/ha) of DAP at planting.Regarding fertiliser application of maize crops in Kenya, a blanket rate of 123.55 kg/ha, corresponding to 50 kg/acre of DAP, for seeding and CAN as top-dressing during the long-rain season from March to August is commonly used across the country.The recommended plant density was 53,333 plants/ ha, based on row distances of 0.75 m and 0.25 m and one seed per planting station.
Regarding mobile phone use, 93% of people in Kenya were mobile phone users by 2012 (Wijesiri & Meoli, 2015), and between 73% and 80% of the rural population habitually uses mobile banking because they lack bank accounts (Wijesiri & Meoli, 2015;Jin et al., 2019).Mobile banking systems in Kenya, such as the M-PESA system introduced in 2007, have greatly enhanced the spread of microcredit programmes targeting smallholder farmers and reducing the transaction costs of credit provision to people without bank accounts (Kandie & Islam, 2022).

Statistical analysis
Statistical analyses were conducted to identify the effects of gender and location on product preferences and repayment behaviour.Welch's t-test and the Mann-Whitney test were used to assess statistically significant differences (HSD) between male and female farmers under normal and nonnormal distributions.Anderson-Darling normality tests were applied to evaluate normal and non-normal distributions.Post hoc Tukey and Games-Howell tests were used for normal and non-normal distributions for different counties.A one-way ANOVA with a non-parametric Kruskal-Wallis test was used to analyse the county effects with unequal sample sizes.Mix-model interactions between county and gender were assessed using a linear mixed model, considering county as a fixed effect and gender as a random effect.Before analysing the mixed model, the county classification per observation was performed using the shapefiles of Kenyan official county boundaries.Mixed models were fitted using the R statistics package (R Studio Team, 2020).The effects of statistical analysis were considered significant at p ≤ 0.05.
Market basket analysis, which was developed in the 1990s (Agrawal & Shafaer, 1996, 1997), was deployed to assess behavioural elements of microcredit clients and to identify associations between different products in a set and frequent transaction pattern.The associate rule mining approach, an unsupervised machine learning method, has been used extensively in the science fields of agriculture and ecology (Aronne et al., 2012;Pitchayadejanant & Nakpathom, 2018;Tandon et al., 2016).
A basket analysis was performed using the 19 Agrics products shown in (Supplementary Information File 1) in order to determine the associations between binary or categorical variables in the large dataset of the Agrics database.Additionally, the numbers of female and male clients were included in the basket analysis to understand their relationship with the products.Support and confidence are key measures that express confidence in the association's rules.Support represents the popularity of a product in all transactions.The higher the support, the more frequent the item set.Confidence can be interpreted as the likelihood of one client purchasing two products (products 1 and 2).This parameter was calculated as the number of transactions that included products 1 and 2 divided by the number of transactions that included product 1.The higher the confidence level, the greater the likelihood that the second item will be purchased.A confidence value of 0.65 was chosen for the threshold based on the classical rule mining procedure (Tandon et al., 2016).Rules of associate analysis used as a proxy for purchasing behaviour were generated from the frequent itemset using an algorithm that guarantees that the support of the rule is greater than or equal to the minimum support (= 0.005).Overall, the rule with a confidence value of 0.65 means that 65% of clients jointly buy products 1 and 2, while 5% of support value (= 0.005) shows that this combination covers 5% of transactions in the database.R packages Arules and ArulesViz with the default 'Apriori' algorithm were used for this analysis, with support and confidence set at 0.005 and 0.65, respectively (Hahsler et al., 2005).The algorithm computes a lift value that determines the strength of the association pattern between two products.The larger the lift value, the stronger the link between the two products.Rules with a high support value are preferred, as they will likely apply to many future transactions.As an additional step for obtaining a better visualisation of the result, redundant rules were taken away by using R packages of 'Arules'.Based on Bayardo et al. (2000), a rule was considered a 'redundant rule' if more general rules with the same or higher confidence exist.

Results
We analysed the data on products purchased and repayments by comparing male and female participants in the microcredit programme.In the Discussion section, we elaborate on the gendered differences we found.

Gendered differences
Analysis of the Agrics dataset revealed that each of the following were higher in male than female farmers (Table 2): total credit taken out per client, repayment balances, plot sizes in the client packages, and advance payments for the next year, which were called 'down payments' on the loans (p < 0.005).Female farmers showed significantly higher repayment performance (p < 0.005).
Figure 1 depicts the cumulative distribution of the plot size of the farming input packages purchased, along with the maize seed usage and the fertiliser types applied at sowing and as top-dressing (Fig. 1).Approximately half of the farmers bought 'half-acre packages' (for 0.202 ha) consisting of 25 kg DAP and 25 kg CAN fertilisers and 5 kg maize seeds and covering 0.202 ha.Female farmers proportionally took out more '0.25-to0.5-acre packages' (equivalent to 0.101 to 0.202 ha), whereas male clients tended towards larger loan packages, covering 1 to 15 acres (0.202 to 6.07 ha). Figure 2 shows mosaic visualisations of the observed frequencies of microcredit product uptake by gender (Fig. 2).A significantly higher proportion of female farmers took out premium seeds, post-harvest products, and solar panels on loan than male farmers (p < 0.005).Table 3 shows the gendered differences in monetary terms and the amounts borrowed per product.It appears that male farmers invested significantly more in fertilisers and maize seeds.In contrast, female farmers invested significantly more in post-harvest products and solar panels (p < 0.005).No gendered difference was found in purchases of the chicken production package.
Investments in solar panel products revealed another significant difference between male and female clients in the microcredit programme (Fig. 3).While more female farmers purchased relatively cheaper solar products from Category 1 (8.50 USD) and Category 2 (31 USD), male farmers purchased more expensive solar products from Categories 3 and 4 (37.85USD and 79.40 USD, respectively).
A so-called market basket analysis, also known as 'association rule mining' is a data mining method that allows us to investigate an extensive transactional database to identify which items were most frequently purchased jointly.In our study, this approach revealed differences in the types and numbers of items purchased by farmers in the microcredit programme (Fig. 4).Some items do not appear in Fig. 4 due to the low frequency and the failure to satisfy the minimum value of the support constraint (=0.005).The result generated 56 rules along with the lift (between 1.025 and 26.984) using the default algorithm (Supplementary Information File 2).After additional treatment to remove redundant rules based on the criteria of Bayardo et al. (2000), 27 rules remained.The premium seeds ('Premium') and the post-harvest goods of storage silos ('Post_Harvest_Silo') appeared most frequently (Supplementary Information File 2).Regarding the comparison between different products, the product of the storage silo, seen as the 'Post Harvest Silo' in Fig. 4, was strongly associated with other products.This reflects that the clients who purchased this product, equivalent to 3.7% of our dataset, concomitantly ordered another product such as a post-harvest drying sheet (C2), insecticide, or a 0.5-acre input package, as shown in rules 1, 2, and 3 (Supplemental Information File 2).Note that the product 'Premium_Seeds' in the middle of Fig. 4 is linked to many rules (shown as arrows), implying that this product was often purchased with other products such as 'Post Harvest Silo', 'Package 0.5-acre', and 'Post Harvest Insecticide'.Regarding gender differences, Fig. 4 shows that the female farmers were linked to a wide range of products and were involved in 42 rules, while male farmers were isolated in the bottom right corner of the figure with only 2 association rules.The large number of arrows (rules) linked to female farmers suggests that female farmers bought a larger number of products than male farmers.For example, rule 16 in the Supplemental Information showed that all female farmers who ordered a 0.5-acre package and drying sheet (C2) concurrently purchased storage silos (support = 0.00616, confidence = 1, lift = 26.9).Overall, this result represents the difference in purchasing behaviour between male and female farmers and confirms the finding in Fig. 2 that a significantly larger proportion of female farmers purchased post-harvest, premium seeds and solar panels.

Poverty
Female-headed households and households without mobile phones are more frequently classified as low-endowed or 'poor' households (Franke et al., 2016;Tittonell et al., 2005;Wesolowski et al., 2012).Although our study could not reveal whether female-headed households were disproportionally excluded from the microcredit programme, we found that male and female clients with no mobile phones made smaller investments (Table 4).Their repayment rates were far below average (80.4%;see Table 4), suggesting problems with repaying even small amounts.We found significant gender differences in both the total cost of investment (loan) and the repayment rate among clients without mobile phones.Male farmers without mobile phones had a slightly higher repayment rate (significance of < 0.1) than The green colour represents the proportion of the number of clients who purchased the product, while the red colour represents the proportion of the clients who didn't purchase.Significant differences (Chi-square test) between male and female farmers are indicated with ** (P <0.05) their female counterparts.By contrast, the gender of the household head did not cause any significant differences in these variables (Table 4).

Discussion
Microcredit programmes are often seen to address Sustainable Development Goals 1, 2, and 5 simultaneously: no poverty, zero hunger, and gender equality (Rambaud et al., 2022).The growing number of client-based social enterprises providing credit to smallholder farmers in Africa suggests that farming households see these programmes as an opportunity to tackle food insecurity by using agricultural technologies (Burke & Lobell, 2017;Nakano & Magezi, 2020) as well as an opportunity to reduce poverty (Aggarwal et al., 2015).Our analyses of the client database of a social enterprise operating in Western Kenya confirm that microloans can support smallholder farmers' agricultural production, enabling them to access agricultural inputs such as fertilisers and seeds.They also show that by extending financial services to resource-poor farmers, we can learn about farmers' needs and gender differences in financial and investment behaviour.Below, we discuss these gender differences and their implications for microcredit programmes targeting smallholder farmers, especially those aiming to address gender inequality in rural areas of Sub-Saharan Africa.

Gendered financial behaviour: borrowing, repayment, and mobile phone access
Our study demonstrates that a higher proportion of women (63%) than men participated in the Agrics programme over the study period, which aligns with other research (Aggarwal et al., 2015;D'Espallier et al., 2011).Aggarwal et al. (2015) found that 67% of borrowers in 1,019 global MFIs were female.They suggest targeting women clients as they are perceived to be more trustworthy, and lending to them has a greater social impact.Furthermore, it is important to note that the legal status of an MFI (e.g., non-governmental organisations (NGO)) is strongly associated with the proportion of female participants (D'Espallier et al., 2011(D'Espallier et al., , 2017)).In terms of the financial management of MFIs, small loans are known to be less profitable, which may result in a reduction in the number of women in microcredit programmes in more profit-oriented MFIs (Agier & Szafarz, 2013).
Fig. 4 The outcome of basket analysis for the interaction between gender and multiple items offered in microcredit programme.In this graph, the items grouped around a circle represent an itemset, and the arrows indicate the relationship rules.The size of the circle represents the level of confidence associated with the rule, and the colour is the level of lift with the range of 1.195-1.886.
(The larger the circle and the darker the grey, the better).The support value is in the range of 0.005-0.325As the analysed customer database was set up for financial monitoring of clients, product purchasing, and distribution, we have no detailed information on the household composition of Agrics clients, including the number of household members, or on various aspects of their farming management, such as the labour availability, farm yield, or food self-sufficiency.Nevertheless, our analyses revealed gender differences in the financial behaviour of Agrics clients.
First, female participants in the microcredit programme tended to take out smaller loans than male participants, which is in line with other works (D'Espallier et al., 2011(D'Espallier et al., , 2013)).While this is undoubtedly a reflection of the fact that women in rural areas tend to have lower access to land (i.e., smaller fields) compared to men, as well as lower financial power, it may also reflect gendered differences in risk-taking and customer behaviour; women may not like to risk nonrepayment.In our analysis, we observed both higher repayment rates for female farmers and higher down payments for microloans for the next season among male farmers (Table 2), which agrees with the finding of D'Espallier et al. ( 2011) that female clients have higher repayment and carry fewer provision expenses than male clients.It appears that women and men deploy different strategies to secure their continued participation in the microcredit programme.From an institutional management perspective, MFIs with a higher proportion of female borrowers exhibit lower portfolio-atrisk and lower write-off rates.This suggests that women are considered to be reliable credit risks for MFIs (D'Espallier et al., 2011).
Second, in line with the findings of a study of over 700 MFIs (Adegbite & Machethe, 2020), our analysis of the Agrics client database shows that the repayment rate of female farmers is significantly higher than that of male farmers (p < 0.05).Other studies have reported that the higher repayment rate of women is at least partly attributable to their greater trustworthiness (Shahriar et al., 2020) and greater willingness to pay back microloans compared to men.Moreover, because female-headed households-despite their tendency to be low-endowed (Tittonell et al., 2005)-have high repayment rates, they also have an increased capacity to meet their financial obligations (Table 4).Thus gender, rather than the household's socioeconomic status, appears to be a better indicator of a client's capacity to repay the microloan.In this regard, the findings of Kodongo and Kendi (2013) agree with our present results.Discussing a microcredit use case in Kenya, they found that the willingness of women to repay was related to the limited credit opportunities for women as compared to men, implying that female clients are eager to ensure their continued access to the limited financial services they are offered.
Active participation in group gatherings can assist borrowers in forming networks to exchange information about business prospects and build trust-based relationships beyond their immediate family (Larance, 2001).As mentioned above in the Materials and Methods section, every group of Agrics members held a weekly meeting in the village where the members lived.The main purpose of the weekly meeting was to track the progress of the individual repayment, share opinions about the difficulties and successful plans, and explicitly plan for the repayment and farming practices in the coming weeks.Della Pellegrina et al. (2021c) found that more frequent participation in such group meetings improved repayment performance.In our present analysis of the Agrics microcredit programme, local members reported that the female farmers attended the weekly meetings more frequently than males, probably because male farmers more often had additional work outside the farm compared to their female counterparts.Another possible reason could be that females were more willing to attend meetings for the purpose of information sharing (Dalla Pellegrina et al., 2021c).
Third, the study found that female farmers without mobile phones had significantly lower repayment rates than male farmers in the Agrics microcredit programme.Mobile phone ownership is not only a wealth indicator but also a communication tool, allowing the microcredit programme to remind clients of their repayment obligations.This observation is supported by reports that many rural households in Kenya have been lifted out of poverty through the M-Pesa mobile banking service (Gichuki & Kamau, 2021).In addition, farmers in our study received SMS text messages reminding them to attend training and meetings.This finding aligns with those of Suri and Jack (2016), who showed that phone access was essential for repayment and poverty alleviation in Kenyan rural areas.

Gendered investment behaviour and livelihoods
Microcredit is not a common source of finance for agricultural input purchases among farmers in Sub-Saharan Africa (Adjognon et al., 2017).Western Kenya is probably different because of the presence of two large-scale microcredit providers-One Acre Fund and Agrics.Since the Agrics microcredit programme extends beyond a standard package of fertiliser and seeds, our study analysed not only financial behaviour, but also farmers' gendered preferences.The products available through the Agrics programme address pertinent problems in African rural areas, such as postharvest crop losses (Sheahan & Barrett, 2017), access to electricity (not least for mobile phone charging), and the need for additional on-farm income generation (especially for women).Before discussing farmers' preferences for different products, we should note that female farmers in our study group tended to purchase more but less expensive items (Table 3 and Fig. 3 and 4), including cheaper solar energy products and post-harvesting technologies such as storage bags.While this investment-spreading behaviour of female farmers may be challenging to interpret, the gendered nature of farmers' investment behaviour suggests that female farmers have considerable autonomy in their investment decision-making; women in male-headed households showed the same investment behaviour as women in femaleheaded households.

Post-harvest products
Reducing post-harvest losses is a crucial step towards achieving food and nutrition security in sub-Saharan Africa (Tefera, 2012;Affogrnon et al., 2015;Mutungi et al., 2023).
Early harvested maize often contains a higher proportion of mouldy grain, increasing the risk of spoilage (Chegere, 2018).Research conducted in Kenya demonstrated that the commonly used polypropylene bags led to maize storage losses of up to 30% after 9 months (Likhayo et al., 2018).
In contrast, the hermetic bags sold by Agrics resulted in only 0.5-1.8%losses after 9 months (Huss et al., 2021).Improved storage bags were initially widespread in West and Central Africa, and then were later (i.e., in 2015) introduced in Kenya (Channa et al., 2019).However, the higher cost of hermetic bags, compared to traditional woven bags with no insect or aflatoxin protection, poses an economic challenge for smallholder farmers (Channa et al., 2019).Similar economic constraints were observed for aluminium silos, which proved profitable for capacities exceeding 0.5 tons (Affognon et al., 2015).Subsidies through government programs have been suggested to reach more farmers in need of these new products (Channa et al., 2019).
In Western Kenya, where our study was conducted, the major issue during storage is losses caused by maize weevils and grain borers (De Groote et al., 2023).Farmers in Kenya gradually started adopting hermetic bags to combat these pests.Those who were already aware of the advantages of the hermetic bags showed a higher willingness to pay for them compared to those unfamiliar with the technology (Channa et al., 2019).Although relatively few farmers in our study used their microloans for post-harvest products (Fig. 2), we found that female farmers, who typically have primary responsibility for household food security and postharvest storage management, purchased significantly more of these products compared to male farmers (Table 3) (Badstue et al., 2020).
Given the limited resources and climate variations affecting smallholder farmers in Kenya, reducing harvest losses is considered a critical component to address food security challenges (Chegere, 2018).Additionally, storing maize for sale during periods of higher market prices allows farmers to increase their expenditure on health, education, and other domestic needs (Affognon et al., 2015).Hermetic bags hold the potential to improve profitability provided that farmers utilize them to preserve maize for a duration of 4 months per season, a process which can generally be repeated four times, as the bags remain intact for a span of 4 years (De Groote et al., 2023).
Research on gender-related post-harvest loss is currently limited and warrants greater attention (Lelea et al., 2022).Subsequent investigations will be needed to examine the ramifications of post-harvest technologies facilitated by microcredit programmes and trainings on product handling, and, ultimately, to elucidate the influence of such technologies on household status indicators such as self-sufficiency duration, as well as on the management of post-harvest tasks including the reduction of labour hours contributed by women.

Maize variety preferences
While the agricultural input packages of Agrics provide for both short-and long-duration hybrid maize varieties of Kenya Seed Company Ltd., many farmers preferred to purchase premium seeds from other seed companies, such as SeedCo's Duma 43 variety.Male farmers invested significantly more in premium seeds than female farmers (Table 3).This gender difference may be related to male and female farmers' different production orientations and resource endowments.With larger package sizes of chemical fertilisers and maize seeds (Table 2 and Fig. 1), male farmers may be more oriented towards selling (surplus) maize, whereas women may prioritise household maize self-sufficiency.The preference of farmers for specific short-duration varieties implies their awareness of the benefits associated with growing these varieties, which are available at a premium price through the microcredit programme.In addition to gendered production orientations and an awareness of the benefits of short-season varieties, differences in performance (emergence, yields, and disease resistance) among different short-duration hybrid varieties may also be necessary for understanding farmer preferences.Investing in higherquality seeds is a relatively small investment (as compared to fertiliser), and hybrid varieties tend to be more responsive to fertiliser inputs, which are also part of the package.

Other products
While the growing need for electricity in rural areas may not appear to have a gender dimension, our study found that significantly more female farmers purchased solar products with their microloans than male farmers.However, the chicken package was not as popular among female farmers because feeding large chickens to raise income is labour-and capital-intensive and requires secure market access.In addition, many rural households have poultry for home consumption and local sales.

Microcredit institution
Agrics, a social enterprise, provides smallholder farmers with various agricultural and non-agricultural inputs.Some of their activities and institutional structures could partly support the enhancement of repayment performance, especially for female farmers.For example, Agric provides initial training about saving skills to all enrolled clients.In addition, clients are offered individual loans rather than group loans.As discussed above, Agrics employees attend weekly meetings and provide feedback to farmers (Langeveld & Quist-Wessel, 2015).In addition, the types of products offered by Agrics are stratified to fit the needs of different clients, as seen in our study by the wide range of package sizes of fertiliser and maize seeds as well as the different types of solar panels.Agrics extension officers support farming practices and instruct farmers on the application of products.In the case of farmers with debt, Agrics members visit them to provide advice and encourage repayment rather than adopting strict enforcement methods such as asset seizure (Goetz & Gupta, 1996;Rahman, 1999Rahman, , 2004)).
The productivity of MFIs is a topic of frequent debate in developing countries, particularly when it comes to policies regarding social outreach for economically disadvantaged clients (Mia et al., 2019;Remer & Katilakoski, 2021).These debates often involve trade-offs between outreach/poverty alleviation and institutional operational self-sufficiency (OSS) (Mia et al., 2019;Remer & Katilakoski, 2021).In cases where an institution has a significant number of female clients with lower portfolios-at-risk and lower write-off rates, rates of OSS could decline (D'Espallier et al., 2011).To enhance institutional OSS, various suggestions have been proposed in the existing literature: 1) Implementation of advanced technologies to reduce transaction and administration costs (Wijesiri & Meoli, 2015); 2) Increasing the recruitment of women board members, as they tend to be more inclined to generate higher operational revenues, thus enhancing OSS (Mia et al., 2019).Based on our findings, it seems that if microcredit institutions were to offer a wider range of affordable products aligned with the specific needs and norms of women in rural livelihoods (such as agriculture, education, sanitation, health, and energy), this targeted product availability could lead to increased investment by female clients.And subsequently, there is the potential for a modest improvement in institutional productivity, although the magnitude of this increase is likely to be relatively small.

Input selection and decision-making dynamics
Gender-related disparities in input selection can be influenced by various factors, including input availability, perceived reliability, and their impact on family economics.Despite the intention of microfinance programmes to be gender-neutral, they may yield divergent outcomes for distinct groups such as low-endowed or female smallholder farmers.The effectiveness of decision-making elements during credit scheme repayment can differ depending on client characteristics.For example, males may possess a distinct understanding of 'Challenge' and 'Feedback' (elements of effective decision-making defined in the Goalsetting theory) (Brandts et al., 2021).Males may exhibit a propensity for riskier challenges and immediate success, whereas females may display heightened sensitivity to negative feedback, leading to risk aversion and a focus on long-term effects (e.g., post-harvest goods and solar panels) that benefit the family unit beyond monetary purchases, in contrast to income derived from maize yields.Nevertheless, these differences in input selection do not imply that female farmers derive lesser benefits from the type of microloans studied here.Evidence exists to support the notion that the selected inputs contributed to an increase in household incomes, and no indications of disparities in programme satisfaction were observed.Enhancing credit programmes necessitates attention to the available product portfolio, with tailored offerings better targeting specific stakeholders (e.g., female farmers), ultimately augmenting efficiency within the realm of smallholder agriculture.
Women's access to and control over agricultural resources plays a vital role in household food security (Kihiu & Amuakwa-Mensah, 2021), especially by avoiding the potential loss of maize grain post-harvest.The gendered nature of farmers' investment behaviour suggests that microcredit programmes may effectively contribute to women's empowerment and female farmers' uptake of new technologies if they develop various investment packages that include relatively cheap items in response to the preferences of female farmers.
Reconsideration of the role of technologies as vehicles is important for challenging the current gender inequalities in SSA countries (Huyer, 2016, Huyer & Carr, 2017).Moreover, matching microcredit schemes and gender norms in the local context is critical, especially in Sub-Saharan countries (Johnson, 2004).Our results suggest that a reorientation of microcredit programmes associated with new technology products towards the specific needs, roles, and potentials of female farmers is both necessary and feasible, such as through the provision of post-harvest products and solar panels that match the demands of farming women based on their behavioural norms in households.Proposing a microcredit scheme instead of a subsidy program, as suggested by Channa et al. (2019), to offer post-harvest products to female clients, along with training programs and technical assistance on post-harvest handling (Chegere, 2018), holds considerable promise as a strategic approach to bolster food security.Furthermore, this approach could have significant political implications.
Further studies should be conducted to evaluate the impact of microcredit programmes on the gender gap across several years, given that previous studies have questioned the durability of the positive impact of microfinance on practitioners in terms of female empowerment in developing countries (van Rooyen et al., 2012;Banerjee et al., 2015).In addition, the influence of cultural differences in this region-including those related to ethnicity, religion, and education level-on gender inequality in farming should be considered (Diiro et al., 2018;Labeyrie et al., 2016).
Exploring the significance of female roles within microcredit institutions, including positions such as board members, loan officers, and extension members, is an intriguing research area that could contribute to enhancing female repayment performance (Otiti et al., 2021;Mia et al., 2021).By examining the various female roles within the institutions themselves, such research would extend the knowledge gained by prior client-side examinations.

Conclusion
This study found that there is a considerable gender effect on the use of microcredit programmes in Western Kenya.While female farmers were found to purchase a more comprehensive range of relatively less expensive products, including those related to post-harvest technology, and to have a higher repayment rate, male farmers tended to buy higher-priced packages of chemical fertiliser and hybrid maize seeds, as well as more expensive solar panels.As a consequence, male farmers took larger loans than female farmers.Furthermore, the analysis showed that male and female farmers may deploy different strategies towards microcredit programmes; while women had higher repayment rates, male farmers tended to maintain their relationship with the programme by making larger down payments on new loans.Such different investment preferences and financial practices/strategies may have wider applicability in smallholder farming areas.
Microcredit for farming inputs and household goods constitutes a powerful strategy to empower smallholder farmers.This research has shown that such empowerment is more cost-effective when targeted at female farmers, who can repay their loans better than their male counterparts despite their limited endowments and financial power.To strengthen women's empowerment and the viability of microcredit programmes in rural Africa, microfinance providers should consider a stronger focus on the goods and services preferred by female farmers.Based on our study findings, one potential strategy could be to provide a range of affordable products related to livelihood necessities such as agriculture, sanitation, energy, and health.Reorienting the design of credit programmes for female farmers with fewer endowments will support them in obtaining inputs, increasing yields, properly storing their harvested products, and consequently improving food security.

Fig. 1
Fig. 1 Cumulative distribution of order plot size (ha) for packages of fertilizers and maize cultivar seeds with different gender categories (Female: solid line and Male: dashed lines)

Fig. 2
Fig. 2 Area-proportional plots of observed frequencies of gendered product preferences for different Agrics products (top: female farmers, bottom: male farmers) for different Agrics products (top left: Premium Seeds; top middle: Non-maize seeds; top right: Post-Harvest products; bottom left: Chicken package; bottom right: Solar panel).

Fig. 3
Fig. 3 Distribution of the number of clients (left: female farmers and right: male farmers) for four different categories of solar panel products by microcredit programme.(Category 1: 8.5 US Dollars (USD); Category 2: 31USD; Category 3: 37.85 USD; Category 4: 74.9 USD)

Table 1
Contents

Table 3
Median, Means, Standard deviations (SD)of loan items of male and female farmers in the package of the chemical fertilisers (Di-ammonium Phosphate and Calcium Ammonium Nitrate) and maize standard hybrid seeds from Kenya Seed Company; Chemical fertilisers and premium maize seeds; Only premium maize seeds; Post-harvest goods; Non-maize crop seeds (sorghum, soybean, and common beans); Chicken package, and Solar panel.The unit of all values represents the US dollar (USD).Asterisk mark represents a significant difference between male and female farmers (P < 0.05)

Table 4
Mean and Standard deviation (SD) of repayment rate (%),