Low-cost soybean input bundles impact women farmers’ subsistence livelihood traps: evidence from Ghana

The Soybean Uptake and Network Survey was administered to a random sample of 832 smallholder male and female farmers in northern Ghana to explore gender and other factors related to soybean production. We investigated the effect of receiving a Soybean Success Kit (i.e., certified seed, fertilizer, inoculum) on soybean yield and income from soybean, controlling for factors such as gender. This analysis includes farmers who 1) resided in districts where Kits were distributed, 2) planted soybean in the past 12 months, and 3) for whom we had complete information for district and gender (n = 371). When results were disaggregated by gender among Kit recipients, average soybean yield (ASY) for males was 108% and average soybean income (ASI) was 97% of that for females. When results were disaggregated by gender among Kit non-recipients, ASY for males was 142% and ASI was 147% of that for females. When results for males were disaggregated by whether the respondent received a Kit, ASY for male Kit recipients was 113% and ASI was 112% of that for male non-recipients. When results for females were disaggregated by whether the respondent received a Kit, ASY for female Kit recipients was 148% and ASI was 170% of that for female non-recipients. These results suggest that providing smallholder female farmers with access to low-cost (˂USD6) input bundles to which they customarily have little or no access can help eliminate the gender gap in agricultural productivity. These results may be applicable to other sub-Saharan Africa countries, where targeting smallholder female farmers as input bundle beneficiaries may positively impact agricultural productivity.


Introduction
The U.S. Agency for International Development reports that "by providing women farmers with the same access to land, new technologies and capital that men have, we could increase crop yields by as much as 30 percent" (USAID, 2019), which is of evermore importance to food security as the world's population is expected to reach 9.7 billion by 2050 (UN, 2019). Low agricultural performance in sub-Saharan Africa (SSA) is of major concern as this region's population is expected to double in that same time period (ibid.). Leading development organizations hold that one reason that agriculture is underperforming in SSA is because smallholder female farmers lack equal access to productive plots, labor, high-quality agricultural inputs (e.g., certified seed, high phosphate fertilizer), credit, technical training, extension services, and technologies known to boost productivity (CGIAR/ LEAD, 2016;Johnson et al., 2016;Quisumbing et al., 2014;UN, 2019;USAID, 2019;World Bank, 2014). Providing smallholder female farmers with equal access to such assets is pivotal to closing the gender gap in agricultural productivity, strengthening livelihoods among both male and female farmers, and improving food security for rural farming communities (Ali et al., 2020;Doss, 2018;Fisher & Kandiwa, 2014).
Previous studies on smallholder farming systems in Ghana and other SSA countries have found that smallholder farmers' access to land, agricultural inputs, and technical training tends to be highly gendered in favor of men, such that -even within the same households -males tend have more access to arable land, training, and important inputs, such as inorganic fertilizer and high-quality certified seed (O'Sullivan et al., 2014;Sheahan & Barrett, 2017; see also Lambrecht et al., 2018), factors that help explain the gender gap in agricultural productivity. And in many such smallholder farming households, husbands and adult sons customarily serve as gatekeepers not only of women's access to farm plots, but restrain women's plot sizes, women's access to labor and inputs, women's decisionmaking regarding what crops to grow on their own plots (i.e., subsistence crops versus high-value cash crops), and women's decision-making and control over their earning from their production of food and cash crops (Kasanga et al., 2018;Lambrecht, 2016;Rodgers & Akram-Lodhi, 2019;Vercillo, 2020;Yokying & Lambrecht, 2020).
Research from previous decades on smallholder farming systems in Ghana and other SSA countries often dichotomized crops into "men's crops" (i.e., high-value cash crops) and "women's crops" (i.e., lower-value subsistence crops). Commenting on this in 2001, Doss stated that the "standard explanation for the division of crops by gender is that women are responsible for feeding the family and prefer to grow subsistence crops" while men "are responsible for providing cash income and to this end grow cash and export crops" (p. 2001). Although Doss makes the critical distinction that, "It is difficult to tell, however, whether women grow lower-value subsistence crops because they have different priorities or because they have less access to land, inputs, credit, information, or markets" (ibid.), a preponderance of evidence over the intervening years across Ghana and SSA suggest the latter (Britwum & Akorsu, 2016;Deere et al., 2013;Kilic et al., 2015;Mbanya, 2011;MGCD, 2016;Quisumbing et al., 2014;Quisumbing et al., 2021;Rodgers & Akram-Lodhi, 2019;Vercillo, 2020).
Indeed, as Meinzen-Dick has put it, "a dense web of laws, policies, programs, and customs" place female farmers in SSA "at a significant disadvantage" as compared to their male counterparts (Meinzen-Dick, 2019). Such gender-biased structural barriers can have an unintentional multiplier effect on inhibiting female farmers' agricultural productivity when they prioritize the production of highvalue cash crops among male farmers and the production of subsistence crops among female farmers. In fact, it is well-documented that female farmers routinely 1) lack equitable access to agricultural inputs (particularly that is sold in smaller, more affordable quantities), cash-in-hand or credit to purchase inputs, and technical training, 2) are socially obligated to perform many types of labor on their husbands' farm plots (e.g., planting, fertilizing, weeding, harvesting, and threshing) that takes their labor away from their own plots, 3) are overlooked by governmental policy-makers and subsidy programs, and 4) are negatively impacted by inheritance practices that are strongly biased in favor of males (Britwum & Akorsu, 2016;Deere et al., 2013;Fisher & Kandiwa, 2014;Kilic et al., 2015;Mbanya, 2011;MGCD, 2016;Vercillo, 2020). Such factors can inadvertently confine female farmers to subsistence livelihoods that fosters economic insecurity rather than resiliency.
For example, Zambia's Farm Input Subsidy Program (FISP) required that beneficiaries provide 50% of input costs upfront in order to qualify (MGCD, 2016). A gender analysis of this policy over five year by Zambia's Ministry of Gender and Child Development determined that the program's eligibility requirement disadvantaged female household heads because they lacked the ability to provide the input costs upfront that were necessary to qualify. This resulted in a profound gender gap over the five-year evaluation period, such that an average of 82.5% of beneficiary households were male-headed (MGCD, 2016). On the opposite end of the spectrum, a gender analysis of the impact of Malawi's FISP on modern maize adoption found that receiving both seed and fertilizer coupons increased the probability of modern 1 3 maize adoption by 222% among female household heads, although it had no impact among male household heads (Fisher & Kandiwa, 2014, p. 108).
An analysis of the gendered division of labor in smallholder farming systems in northern Ghana found that females were socially obligated to perform many different types of labor on their husbands' plots (e.g., sowing, weeding, applying fertilizer, threshing), which impacted the time and labor they were able to devote to their own plots. Males deemed that the application of fertilizer to their plots was "women's work" because, "women were used to bending down to perform household and farm tasks such as sweeping and sowing. Applying fertilizer requires the same skill and so comes easily to women" (Britwum & Akorsu, 2016, p. 31). Females were "obliged to apply fertilizer on their husbands' farms before they did so for their own farms" (ibid., p. 32), and "have to work on their husbands' farms before theirs" (ibid., p. 29). Likewise, females in northern Ghana are responsible for and expected to prioritize hand-threshing their husbands' crops before their own crops (J. Appiagyei, personal communication, 2020), which is both a tremendous physical burden and time burden as mechanized threshing is extremely rare in this region.
In northern Ghana, limited access to agricultural inputs such as high-quality certified seed and high phosphate fertilizer, to credit, and to technical training and extension services are long-standing challenges for smallholder farmers, particularly female farmers (Britwum & Akorsu, 2016;Dogbe et al., 2013;Etwire et al., 2016;Ragsdale et al., 2018;Tasila Konja et al., 2019;Yoking & Lambrecht, 2020). For example, results from the baseline Women's Empowerment in Agriculture Index (WEAI) implemented in northern Ghana found that, as compared to females, males were 4.35 times more likely to have adequate empowerment to speak up in public to ask extension agents about agricultural practices, policies, or decisions that directly affected them and males were 20 times more likely to have adequate decisionmaking input into the purchase, sale, or transfer of assets (Ragsdale et al., 2018).
Although a relatively new crop as compared to maize, soybean is an important food crop and cash crop in Ghana and across SSA that has market and non-market benefits for smallholder farmers. For example, soybean is valued as an income-generating crop for both male and female farmers, soybean's high protein and oil content makes it a valuable food crop that -when grown in rotation with maize -can increase dietary diversity and nutritional status of households that cannot afford animal-sourced protein, soybean's nitrogen-fixing capability contributes to its value as a source of improving soil fertility, and soybean crop residue is an important livestock feed (Acevedo-Siaca & Goldsmith, 2020; Asodina et al., 2020Asodina et al., , 2021bAwuni et al., 2020;Debnath & Babu, 2020;Dogbe et al., 2013;Gbegbelegbe et al., 2019;Siamabele, 2021).
Although soybean is a crop grown by both genders in northern Ghana, productivity among female farmers lags behind that of male farmers. In order to better understand how an agricultural intervention might help address this productivity gap, we sought to explore whether a low-cost agricultural input bundle that purposefully targeted smallholder female farmers (without excluding male farmers) impacted soybean yield and income derived from crop sales among female farmers. In this paper, we present results from a survey that we used to investigate the effect of receiving a soybean input bundle in the previous cropping season on soybean yield and income from soybean sales (hereafter referred to as "soybean income") among smallholder male and female farmers in Ghana's Northern Region. For the present analysis, we focus on addressing the following research question: Does receiving a soybean input bundle reduce gender gaps in soybean yield and soybean income among this sample of smallholder male and female farmers?

Soybean Success Kits
In this paper, we focus on the Soybean Success Kit intervention (hereinafter referred to as a "Kit"), an input bundle initiative of the Feed the Future Soybean Innovation Lab (SIL). A consortium of researchers, extensionists, nongovernmental organizations, and private sector actors, SIL is focused on nutrition-sensitive agricultural development throughout SSA. SIL research areas relevant to the present paper include seed quality, gender impacts, and economic impacts. Because poor seed quality and low input use are persistent issues for smallholder farmers in Ghana's Northern Region (Awuni et al., 2020;Dogbe et al., 2013;Etwire et al., 2016;Mbanya, 2011), researchers affiliated with SIL's seed quality program designed the Kit intervention to directly transferred input bundles of certified soybean seed, high phosphate fertilizer, and inoculum to smallholder beneficiaries. In March 2015, these SIL researchers conducted agronomic training workshops and distributed 1,197 Kits to smallholder male and female farmers in three districts in Ghana's Northern Region (i.e., Chereponi, Saboba, and Tolon). These activities were conducted by SIL in partnership with Catholic Relief Services-Ghana (CRS-Ghana) and with assistance from local Ministry of Food and Agriculture (MoFA) extension agents.
The training workshops and distribution of Kits were conducted in each of nine villages (three per district) and these combined activities were completed in approximately two hours per site. As a prerequisite to receiving their free Kit, recipients attended their village's on-site workshop, 1 3 which was conducted in their local language. Each workshop focused on interactive trainings and hands-on demonstrations of best agronomic practices to increase soybean yield, such as how to apply inoculum to soybean seeds, proper row spacing, and proper planting density. For example, Kit recipients were instructed that the Kit input bundle was adequate for an area approximately 20 feet by 20 feet (i.e., 1/10th of an acre or 404 square meters). The agronomic information was provided to workshop attendees in feet and acres, as this is the standard measurement used by smallholder farmers in the villages where the Kits were distributed.
Each workshop was designed to encourage all attendees -and particularly female farmers -to both ask questions and to answer questions put to them by the SIL, CRS-Ghana, and MoFA trainers. As stated previously, findings from the WEAI conducted in villages in these districts indicated that female farmers differed significantly from their male counterparts in their comfort level with speaking up in public with extension agents about agronomic practices (Ragsdale et al., 2018). In order to help mitigate this potential barrier among female Kit recipients, each workshop team included at least one female trainer from SIL, CRS-Ghana, or MoFA who was well-versed in purposefully engaging female farmers.
Each Kit included 2.5 kilos of locally produced certified Jenguma soybean seed, 2 kilos of high phosphate fertilizer (5-25-18), inoculum, sugar (for use with the inoculum), and padded leather gloves (for use during harvesting). The inputs were placed into a heavy-duty polyester bag that was printed with simple graphics to illustrate the workshop's trainings on agronomic best practices for soybean (i.e., optimal plant spacing, etc.). The bag was also printed with simple graphics to illustrate the Kit's motto of "eat some, save some, sell some" (i.e., use a portion of your soybean yield to feed your family, save a portion of your soybean yield to ensure you have certified seed for the next cropping season, and sell a portion of your soybean yield in order to generate necessary cash income). Each bag was secured with a thick rubber band to ensure the Kits were easy for recipients to transport and store as needed. See Table 1 for a detailed breakdown of the Kit costs in USD and Fig. 1 for an image of the Kit graphics.
Each of the nine villages received 133 Kits per village, for a total distribution of 1,197 Kits. The distribution process in each village was directly observed and facilitated by SIL, CRS-Ghana, and MoFA immediately after each Kit workshop and recipients included approximately equal numbers of male and female farmers. As the Kits were distributed in March 2015, recipients would have planted their Kit soybean seed in June-July 2015 and harvested their Kit soybean crop in October-November 2015. As the SUNS was administered in June 2016, it is expected that Kit recipients would have considered the Kits in their responses to the survey. To our knowledge, no other governmental or non-governmental organization (NGO) was conducting a large-scale input bundle intervention in 2015 or 2016 in the nine villages included in the present study.

Materials and methods
In June 2016, researchers affiliated with SIL's gender impacts program conducted enumerator training workshops and administered the Soybean Uptake and Network Survey (SUNS) in the same nine villages in which the 1,197 Kits were distributed in March 2015. Eligibility to participate in the SUNS included that the respondent reported that they were a resident of that village, 18 years  . 1 Image of the graphics printed on the Soybean Success Kit bags to reinforce and remind Kit recipients of agronomic best practices for soybean (e.g., optimal plant spacing) and the Kit's motto of "Eat Some, Save Some, Sell Some" ◂ of age or older, and a key decisionmaker in their household (i.e., a household head). Informed consent was obtained from all respondents, and no identifying information was collected. All study procedures and personnel were approved by the Institutional Review Board (IRB) of Mississippi State University. Community mobilization, recruitment of enumerators/IRB-required compliancy, and three two-day enumerator training workshops (one per district) were conducted by SIL and CRS-Ghana. During each workshop, enumerators received training on the protection of human subjects, informed consent, the random sampling strategy, and other protocol issues. The SUNS enumeration was supervised on-site by SIL and CRS-Ghana in each village.
For the random sampling, over-sized maps of each village were produced using Google Earth maps and each village was divided into four or more quadrants (depending on the size of the village). The research team numbered each house in that quadrant, created a list of all the numbered houses, and randomly generate a list of houses to be interviewed by each pair of enumerators (one male and one female) assigned to that quadrant. When possible, male enumerators interviewed male respondents and female enumerators interviewed female respondents.
The SUNS collected information on demographics, soybean seed access, soybean cultivation, income generated from soybean sales, etc. In addition to fill-in-theblank responses, the SUNS included yes/no responses, list responses, and 4-point Likert-type scale responses (the latter which were dichotomized into "agree" and "disagree" in this paper). For example, respondents were asked, "What was the actual yield from your soybean plot?" and "What was the actual income from your soy plot?" Respondents were also asked to rank the five most important crops grown by their household as a food crop in the past 12 months, and to rank the five most important crops grown by their household as a cash crop in the past 12 months.
The SUNS also included the previously validated Household Hunger Scale (Ballard et al., 2011) to explore household-level food insecurity. This 6-item scale was created in partnership with the Food and Agriculture Organization of the United Nations (FAO), FHI 360, and USAID. The Household Hunger Scale is a cross-cultural tool designed to measure occurrence of three householdlevel "hunger events" during the past four weeks, including 1) there was no food to eat in the respondent's house due to lack of resources to get food, 2) the respondent or another household member went to sleep at night hungry because there was not enough food, and 3) the respondent or another household member went a whole day and night without eating anything because there was not enough food. If a respondent reported that any of the three hunger events occurred 1-2 times in the past four weeks, the household was categorized as experiencing "occasional" food insecurity. If a respondent reported that any of the three hunger events occurred 3-10 times in past four weeks, the household was categorized as experiencing "moderate" food insecurity. If a respondent reported that any of the three hunger events occurred 11 or more times in past four weeks, the household was categorized as experiencing "severe" food insecurity.

(a) Statistical analysis plan
We obtained descriptive statistics of demographics characteristics, ranked importance of crops for household consumption and for market, the Household Hunger Scale variables, and soy-related variables on the full sample using SPSS 26.0. For the present analysis, the data was disaggregated by gender in order to compare results for female farmers vis-àvis male farmers (of whom the majority were husband-wife dyads). We also conducted statistical analyses focusing on soybean yield and soybean income with two main objectives: 1) we assessed how the receipt of a Kit affected soybean yield and soybean income while controlling for other variables, and 2) we examined any differences in the effect of receiving a Kit for male and female farmers.
Soybean yield and soybean income were the dependent variables, and they both took non-negative values and their distributions were both heavily skewed to the right (see Fig. 2). Thus, we employed quasi-Poisson log-linear models, in which the mean of the dependent variable was associated with the independent variables via a log link function. The mathematical expression of the log-linear model as well as the interpretation of the regression coefficients are provided in the Appendix.
As soybean income was found to have some association with plot size and with whether the respondent received a Kit, we used the technique of inverse probability weighting to avoid potential for bias around this pattern (Robins et al., 1994). In this technique, each respondent who reported their soybean income was weighted by the inverse of the estimated probability that the respondent reported their soybean income. Essentially, this technique creates weighted copies of the respondents with reported soybean income to remove the bias introduced by missing data processes (Li et al., 2013). To calculate the estimated probabilities, we fitted a logistic regression model in which the response variable is whether the respondents reported their soybean income, and the covariates include (i) whether the respondent's soybean plot size was larger than one acre, and (ii) whether the respondent received a Kit. The details of calculating the probabilities are included in the Appendix.

(a) Demographics characteristics
The sample (N = 620) included an equal number of smallholder male and female farmers who were nearly equally distributed across the three districts of Chereponi (34%), Saboba (35.4%), and Tolon (33.3%) (see Table 2). The majority of respondents were married (94.7%), Muslim (75.8%), and reported that their household was headed by a married couple (94%). Although the majority of respondents had received no formal education, 89.9% of females reported that they had never attended school as compared to 77.5% of males. Household size ranged from 2-50 with a mean of 9.4 (SD = 5.3).

(b) Ranked importance of crops for food and cash disaggregated by gender
As Table 3 indicates, when respondents ranked the five most important crops grown by their household as a food crop in the past 12 months, maize was ranked first by both females (93.5%) and males (98.5%). Soybean was ranked third by females (42.3%) in terms of importance as a food crop, as compared to fifth by males (37%). When respondents ranked the five most important crops grown by their household as a cash crop in the past 12 months, soybean was ranked first by both females (73.7%) and males (77.1%).

(c) Household hunger scale results
Hunger Event 1. Seventy-four percent of respondents (459) reported no occurrence of hunger event 1 in the past four weeks and these results differed little when disaggregated by gender (see Table 4). However, 21.1% of respondents reported that hunger event 1 occurred 1-2 times in the past Fig. 2 Histogram of soybean yield and soybean income, where "yield" is in kilograms and "income" is in Ghanaian Cedis  (131) and 4.7% of respondents reported that hunger event 1 occurred 3-10 times in the past four weeks (29). Hunger Event 2. Seventy-five percent of respondents (459) reported no occurrence of hunger event 2 in the past four weeks and these results differed little when disaggregated by gender. However, 19.6% of respondents reported that hunger event 2 occurred 1-2 times in the past four weeks (120) and 5.4% of respondents reported that hunger event 2 occurred 3-10 times in the past four weeks (33). Hunger Event 3. Eighty-one percent of respondents (493) reported no occurrence of hunger event 3 in the past four weeks and these results differed little when disaggregated by gender. However, 14.5% of respondents reported that hunger event 3 occurred 1-2 times in the past four weeks (88) and 4.1% of respondents reported that hunger event 3 occurred 3-10 times in the past four weeks (25).

Soybean yield and soybean income
The remaining analyses focus on the subsample of 371 respondents who met the following requirements: 1) resided in one of the three districts where the Kit intervention was implemented (i.e., Chereponi, Saboba, or Tolon) and 2) planted soybean in the past 12 months. As described in the statistical analysis plan above, we investigated the effect of the Kit on soybean yield and soybean income among this subsample. With Kit recipience as our dependent variable, we controlled for other variables that were significantly associated with soybean yield and soybean income. For soybean yield, control variables  included soybean plot size, household size, district, and soybean seed variety. For soybean income, control variables included soybean plot size, soybean yield, and soybean seed variety.

(b) Analysis results
To assess the effect of Kit recipience, we fit the log-linear models for soybean yield and soybean income, which are called Model 1-Yield and Model 1-Income, respectively, with Kit recipience and the corresponding control variables. Tables 5 and 6 include the estimates of the regression coefficients with the p-values from these two models.
Controlling for the soybean plot size, household size, district, and seed varieties, the average soybean yield among Kit recipients was 123% of that among Kit non-recipients (with a p-value of 0.062), and the average soybean income among Kit recipients was 126% of that among Kit nonrecipients (with a p-value of 0.087).
To compare the effect of receiving a Kit on soybean yield and soybean income for males as compared to females, we fit another two log-linear models, called Model 2-Yield and Model 2-Income, with the above-mentioned control variables for each model, the receipt of Kit, gender and the interaction between Kit receipt and gender. The interpretation of the regression coefficients is provided in the Appendix. Tables 5 and 6 also report the estimates of the regression coefficients for these two models. This analysis produced four sets of primary results.
First, when average soybean yield and average soybean income were disaggregated by gender among Kit recipients, both results were nearly identical. Specifically, average soybean yield for males who received a Kit was 108% (with 95% confidence interval (CI) (87%, 134%)) of that for females who received a Kit. Similarly, average soybean income for males who received a Kit was 97% (with 95% CI (73%, 127%)) of that for females who received a Kit. Second, in sharp contrast, when average soybean yield and average soybean income were disaggregated by gender among Kit non-recipients, both results differed more substantively as compared to those for Kit recipients. Specifically, average soybean yield for males who did not receive a Kit was 142% (with 95% CI (98%, 205%)) of that for females who did not receive a Kit. Similarly, average soybean income for males who did not receive a Kit was 147% (with 95% CI (90%, 240%)) of that for females who did not receive a Kit.
Third, when males' average soybean yield and average soybean income were disaggregated by whether the  146%)) of that for males who did not receive a Kit. Similarly, average soybean income for males who received a Kit was 112% (with 95% CI (83%, 151%)) of that for males who did not receive a Kit. Fourth, in sharp contrast, when females' average soybean yield and average soybean income were disaggregated by whether the respondent received a Kit, the results indicate that receipt of a Kit among females had a statistically significant impact on the outcomes of interest at the significance level 0.05 (with a p-value of 0.031). Specifically, average soybean yield for females who received a Kit was 148% (with 95% CI (105%, 210%)) of that for females who did not receive a Kit. Similarly, average soybean income for females who received a Kit was 170% (with 95% CI (105%, 274%)) of that for females who did not receive a Kit. In Model 2-Yield, the effect of Kit recipience and gender on average soybean yield is the same, regardless of district, plot size, household size, or seed variety planted. To illustrate this effect, Fig. 3 plots the estimated average soybean yield by Kit recipience and gender among our subsample and Anidaso. Right plot: estimated soybean yield by Kit recipience and gender among a "snapshot" of farmers from Tolon District with a plot size of one acre, a yield of 160 kg, and planted Jenguma and Salintuya of farmers from Tolon District with a plot size of 1 acre and a household size of 8 members (the median value of each variable), and seed variety planted (the combination of which was correlated with, and thus controlled for, in terms of yield). As with soybean income, in Model 2-Income, the effect of Kit recipience and gender on the average soybean income is the same, regardless of district, plot size, household size, or seed variety planted. Figure 3 also similarly plots the estimated average soybean income by Kit recipience and gender among our subsample of farmers from Tolon District who had a plot size of 1 acre and a yield of 160 kg, and planted Jenguma and Salintuya (the combination of which was correlated with, and thus controlled for, in terms of income). Figure 4 includes the residual plots (Pearson residuals versus fitted values of the response) of Model 1 and Model 2 for soybean yield and soybean income. For either yield or income, the two models perform similarly: the majority of the Pearson residuals are within -3 and 3 (except for a few outliers), indicating an adequate fit. In addition, we calculate the pseudo R-squared measure (Heinzl & Mittlböck, 2003) for each model, reported in Tables 5 and 6. It shows that there is still a proportion of the variation in the dependent variables that cannot be explained by the current model.

Discussion
In this paper, we investigated the effect of receiving a Soybean Success Kit (i.e., a soybean input bundle) in the previous cropping season on soybean yield and soybean income among smallholder male and female farmers in Ghana's Northern Region. This investigation produced four sets of primary results, which are compared and contrasted below. In contract to results of Asodina et al. (2021a), but aligned with results of Fisher and Kandiwa (2014) we found that -receipt of a Kit closed the gender gap in agricultural productivity for female Kit recipients, but did not boost agricultural productivity for male Kit recipients.
What are potential explanations for this set of results? It may be that male Kit recipients were more familiar with growing soybean and, therefore, might have disregarded the agronomic guidelines provided during the Kit workshops conducted by SIL in partnership with CRS-Ghana and MoFA. Other potential explanations for these findings include that male Kit recipients may have mixed their certified Kit seeds with "saved" soybean seed (Sedivy, 2019) that was not as high-yielding and/or male Kit recipients may not have applied their Kit inoculum per agronomic guidelines provided during the workshops. It is also possible that male Kit recipients may have not followed plant-spacing and

Fig. 4 Pearson residuals versus fitted value of the response variable for different models
Low-cost soybean input bundles impact women farmers' subsistence livelihood traps: evidence… 1 3 sowing agronomic recommendations, or provided as much attention in the form of weeding, etc. Male Kit recipients may also have planted their Kit seeds on their least productive land, as their crop from Kit seeds would have comprised a smaller proportion of their total soybean crop as compared to female Kit recipients.
In contrast, female Kit recipients may have valued their Kit seeds more highly given that it is well-documented that smallholder female farmers in this region often lack access to vital inputs -such as certified seed, high phosphate fertilizer, and inoculum -which are essential to higher input crops like soybean (Awuni et al., 2020;Britwum & Akorsu, 2016;Dogbe et al., 2013;Etwire et al., 2016;Lee et al., 2019;Mbanya, 2011;Ragsdale et al., 2018;Sedivy, 2019). Likewise, female Kit recipients may have applied their Kit inoculum per the agronomic guidelines provided during the workshops, may have followed plant-spacing and sowing instructions, may have provided more attention to their soybean crop in the form of weeding, etc., and may have planted their soybean crop on their most productive land.
Observations from the training workshops lend credence to these two contrasting sets of potentialities. As the lead workshop trainer reported, "In the group sessions, men were adamant that they knew the correct spacing for soybeans, even though the spacing was far greater than we recommended. Women were less sure, so if they followed our advice, just planting at a proper density would have increased yield. Also, men always said they put two seeds per hole. We told them to put one seed per hole, and if women were putting one seed per hole, that would have also greatly increased yield. And women might have guarded their inoculum more closely and used it properly" (K. Clark, personal communication, 2020).
In terms of the second set of findings -that when results among Kit non-recipients were disaggregated by gender, average soybean yield for males was 142% of that for females and average soybean income for males was 147% of that for females -this result is consistent with other findings from northern Ghana that female farmers, in general, experience lower agricultural productivity as compared to their male counterparts (Britwum & Akorsu, 2016;Dogbe et al., 2013;Etwire et al., 2016;Ragsdale et al., 2018;Tasila Konja et al., 2019;Yoking & Lambrecht, 2020). Ragsdale and colleagues (2018) also identified the female farmers' lacked empowerment in accessing the expertise of agricultural extension agents. This may help explain the present gender gap in productivity identified in this result, as lack of agronomic guidance can be a critical barrier for farmers who may need guidance on soybean, as it is a relatively new crop as compared to maize. And finally, it is likely that the present sample of female Kit non-recipients lacked access to supply chain partners, such as credit and service providers, as has been widely observed among other smallholder female farmers in northern Ghana and across SSA (Britwum & Akorsu, 2016;Kilic et al., 2015;Lambrecht et al., 2018;O'Sullivan et al., 2014;Sheahan & Barrett, 2017).
In terms of the third set of findings -that when results for males were disaggregated by whether the respondent received a Kit, average soybean yield for male Kit recipients was 113% of that for male Kit non-recipients and average soybean income for male Kit recipients was 112% of that for male Kit non-recipients -this result suggests that the benefit to agricultural productivity of a male receiving a Kit was slight. This result suggests that females are likely to have farming plots that are of lower fertility to begin with and, as a result, they experience a bigger response curve when fertilizer is added to their soybean plots. As agronomic literature for this region has demonstrated, adding fertilizer to fertile plots often does not produce a yield response (Awuni et al., 2020;Lee et al., 2019;Sedivy, 2019). Another potential explanation for this result is that male farmers are rotating their soybean plots with maize -which men prioritize as a cash crop in northern Ghana (Britwum & Akorsu, 2016) -and this rotation may provide benefits to their plots. It may also be that male farmers in this sample already tended to have access to the inputs that the Kits provided, so yields did not increase substantively among male Kit recipients as compared to male Kit non-recipients.
And in terms of the fourth set of findings -that when results for females were disaggregated by whether the respondent received a Kit, average soybean yield for female Kit recipients was 148% of that for female non-recipients and average soybean income for female Kit non-recipients was 170% of that for female non-recipients -these results suggests that the benefit to agricultural productivity of a female farmer receiving a Kit was more profound than the benefit to agricultural productivity of a male farmer receiving a Kit. These results also suggest that, as discussed previously, most female Kit recipients were likely starting from a much lower place of plot fertility and access to inputs than their male counterparts and, therefore, experienced a 'bigger bump' in soybean productivity when they had the opportunity to use the Kit inputs such as inoculum and fertilizer. In fact, it is noteworthy that when comparing average soybean yield and average soybean income for female Kit recipients (148% and 170%, respectively) and for male Kit non-recipients (142% and 147%, respectively) to that for female Kit nonrecipients the percentages for female Kit recipients and male Kit non-recipients are quite similar when contrasted with that of female Kit non-recipients. Most importantly, these results highlight that female farmers agricultural productivity responded extremely well to a very modest and low-cost (˂USD6) "inputs plus extension" intervention, which may be one way to increase the agricultural productivity of female farmers by facilitating their production of a crop that is both a high-value cash crop (which can foster economic resiliency) 1 3 and a nutritious food crop (which can foster household food security and dietary diversity. At the opposite end of this spectrum to the third set of results, it may be that female Kit non-recipients in the present sample -due to their lack of access to inputs, agronomic training, and lower agricultural empowerment (e.g., ability to negotiate for highly productive farm plots) -have rarely had what the Kit provided to female Kit recipients. As a result, female Kit recipients experienced a "bigger bump" in soybean productivity when they had the opportunity to use the Kit inputs where they could really have an effect. As it is likely that male Kit recipients were adding their Kit inputs to plots that were already more fertile than that of female Kit recipients, male Kit recipients experienced less dramatic soybean yield increases (Awuni et al., 2020;Lee et al., 2019;Sedivy, 2019) and, ultimately, less increases in income due to the sale of their soybean crop. It is possible that the male and female Kit recipients did not do anything differently with their Kit inputs. Instead, it may be that -because their 'starting points' in terms of plot fertility were so different for male farmers and female farmers -that the soil in the females' plots had a huge response to the addition of the Kit's inoculum and fertilizer, while the soil in the males' plots was not as deficient and, therefore, had a much lower response to the Kit inputs.
Care is needed when generalizing the results, as our findings may not be representative of all smallholder male and female farmers who receive agricultural interventions. A potential limitation of this study is that, although it was conducted among a random sample of smallholder male and female farmers, this sample for the 2016 survey was drawn from a population of farmers from villages in which the WEAI + Wave I survey was administered in 2014 (Ragsdale et al., 2018) and the Soybean Success Kit intervention was implemented in 2015. And because the 2016 survey collected self-reported data, it may be that some farmers over-or under-estimated their soybean yield and/or soybean income when responding to those items. Despite these potential limitations, the study supplies a needed analysis given the scarcity of existing data on how agricultural interventions differentially impacted smallholder male and female farmers -particularly those from the same communities and from within the same households.
Tamimie and Goldsmith (2019) describe the "long jump" nature of Ghanaian smallholder adoption of soybean, as it is a relatively new crop (as compared to maize, for example) and a crop whose agronomic requirements can make adoption more complex for smallholder farmers. For example, while high phosphate fertilizer boosts both maize and soybean production, soybean also requires the additional input of inoculum to be most productive. And soybean production requires smallholder farmers to employ agronomic best practices that may be relatively new to them (e.g., narrow row planting). Further, access to necessary inputs and technical training can be notable barriers to soybean adoption for smallholder farmers in general, and smallholder female farmers in particular. The Kit intervention lowered these barriers to adoption and specifically targeted smallholder female farmers as beneficiaries by ensuring that Kits were equally available to both genders during the training workshops. As predicted, soybean yield and soybean income were higher among Kit recipients as compared to Kit non-recipients, which suggest that this low-cost ˂USD6) intervention was effective.
The Kit intervention results and other research in this region (Awuni et al., 2020;Lee et al., 2019;Sedivy, 2019;Tamimie & Goldsmith, 2019) suggest that barriers to soybean adoption among smallholder farmers are not technical but may lie elsewhere. Instead, the present results point back to input-related "long jump" barriers faced by many smallholder producers -particularly females farmers -throughout SSA. For example, in a study using cross-country, nationally representative agriculturally intensive data collected in six countries in SSA, Sheahan and Barrett found that "plots managed or owned by men, the vast majority of all plots, are statistically significantly more likely to receive inorganic fertilizer and in higher amounts" (2017, p. 22). The Kit intervention relieved smallholder female farmers from the need to acquire three non-technical components including highquality certified soybean seed, high phosphate fertilizer, and inoculum, thereby lowering "long jump" barriers to soybean adoption for this subsample of smallholder farmers. The Kits were pre-bundled and required no additional steps other than that each Kit recipient attend a technical training workshop on agronomic best practices for soybean cultivation. The Kits eliminated recipients' need for credit and input investment, which are recognized as particular constraints for smallholder female farmers throughout SSA (Lambrecht et al., 2018;O'Sullivan et al., 2014;Sheahan & Barrett, 2017). Therefore, Kit recipients only risked the labor that they could have deployed on other economic activities (e.g., producing a different crop, engaging in another income-generating activity).
The more nuanced features of the Kit intervention then raise the challenge with "long jumps" as to how to lower adoption barriers and relieve smallholder farmers from bearing all production risk. We found a number of examples among commercial crops in both developing and developed countries where farmers access bundles, become part of knowledge networks, and share risks with upstream input suppliers and/or downstream buyers. For example, one of many businesses working with the Soybean Innovation Lab employs a novel in-grower scheme (A. Goodman, personal communication, 2019; see also Felgenhauer & Wolter, 2008;Oya, 2012;Väth et al., 2019). The business provides land for 200 farmers (in-growers) free of charge for one year. These 1 3 in-growers then produce soybean alongside more experienced outgrowers (smallholders) and the farm itself. Upon graduating from the in-grower experience, the smallholder farmer then takes the knowledge gained back to her/his farm. The relationship continues, as the nucleus farmer provides outgrowers who would like to produce soybean with a bundle of all necessary inputs on credit. At harvest, farmers then pay back the production credit with grain as payment in kind.
Similar to the Kit intervention, this in-grower scheme reduces barriers to adoption by combining an input bundle, a supportive knowledge network, and a buyer willing to share the production risk. This system differs from traditional demonstration plot or lead farmer models as the in-grower scheme enhances profitable soybean production and reduces risks by centering on a bundled approach where farmers are ensured of access to and affordability of inputs, access to technical information through a network of connected growers, and access to post-harvest markets. A similar system of barrier reduction occurs in developed countries, as farmers often buy seed, fertilizer, and chemical bundles specifically prescribed for her/his farm.
In conclusion, the challenge is not whether smallholder farmers, especially female farmers, can grow a productive soybean crop. Rather, the present results suggest that a lowcost intervention that bundles inputs to which smallholder female farmers often have little access (e.g., certified seed, high phosphate fertilizer, inoculum) and includes agronomic training that intentionally engages female farmers matters when seeking to scale adoption of soybean as both a cash and food crop, and has implications for other crops. Lowcost interventions that purposefully target female farmers may help lift female farmers out of subsistence livelihood traps. However, as Odoul and colleagues note, "designing an intervention that aims at integrating women in high-value agricultural commodity markets, particularly in the maledominated commercialised value chains…requires multipronged approaches that involve understanding complex social issues surrounding intra-household resource allocation and gender relations" (2017, p. 238).
One such multipronged approach put forth by Vondolia and colleagues (2021) is to "target spending explicitly on complementary inputs" (p. 8242), which the Kits accomplished by including certified seed, inoculum, and fertilizer. Other experts also suggest that targeted input subsidy programs can improve impact (Benin et al., 2013;Houssou et al., 2019;Jayne et al., 2018;Pauw, 2021;Ricker-Gilbert et al., 2014). It is clear that, in the case of the Soybean Success Kit intervention, this low-cost input bundle subsidy had a positive impact on productivity among smallscale female farmers -who were explicitly targeted as beneficiaries (without excluding their male counterparts). The present results provide further evidence that specifically targeting female farmers as beneficiaries has the potential to reap large dividends in agricultural productivity across SSA, given that much of the agricultural output is produced by smallscale farmers and that female farmers have long been overlooked as beneficiaries and are often delegated to the roles of "subsistence crop producers" and/or "help-mates" to their husbands, who are defined as the "cash crop producers." Finally, the Kit intervention involved a self-contained set of inputs and agronomic guidance. The bundle serves as a powerful vector to both deliver technology as well as help reduce the risk for smallscale female farmers embarking on a relatively new crop like soybean. As farmers increasingly face challenges from a changing climate and sustainably managing their farms, self-contain technology bundles that adjust with changing conditions and markets become an appropriate technology for smallscale female farmers seeking to exit subsistence livelihood traps. Continued research on the impact of low-cost agricultural interventions on productivity and economic outcomes among smallscale female farmers across SSA is urgently needed, given that female farmers in this region often encounter higher barriers to agricultural productivity vis-à-vis male farmers within their own communities and even within their own households (Britwum & Akorsu, 2016;Etwire et al., 2016;Kilic et al., 2015;Lambrecht et al., 2018;O'Sullivan et al., 2014;Ragsdale et al., 2018;Sheahan & Barrett, 2017).

Availability of data and material Not applicable.
Code availability Not applicable.

Declarations
Ethics approval The study procedures were reviewed and granted exemption by the Institutional Review Board of Mississippi State University.
Consent to participate Informed consent was obtained from all individual participants included in the study and no identifying information was collected.
Consent for publication Informed consent included consent to publish the data in the form of briefs, reports, scientific conference and meeting presentations, and journal articles.

Conflicts of interest
The authors declared that they have no conflict of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.