Gender-differentiated preference for sweetpotato traits and their drivers among smallholder farmers: Implications for breeding

Abstract

To improve sweetpotato (Ipomoea batatas L.) productivity, several improved high-yielding varieties have been developed by breeders. However, many farmers still grow low-yielding landraces. Farmers choose varieties to grow based on their preference for the attributes of those varieties. Varietal preferences have been shown to differ between males and females. This study assessed farmer preferences for sweetpotato traits and the factors that drive the choice of most preferred traits. It used a uniquely large data set collected through personal interviews with male and female sweetpotato growers. The study employed multinomial probit regression to examine the drivers of trait preference. It finds a higher preference for production-oriented traits among farmers in general and especially older ones. This is, however, lower among more educated farmers who mainly prefer risk-averting traits, and those growing local varieties who mainly prefer quality traits. Hence, alongside production-oriented traits, other traits critical for the acceptance of new varieties by farmers in their respective contexts should not be ignored.

Abbreviations

IIA

independence from irrelevant alternatives

MNP

multinomial probit

OFSP

orange-fleshed sweetpotato

SPHI

sweetpotato profit and health initiative

SPVD

sweetpotato virus disease

VAD

vitamin a deficiency

1 INTRODUCTION

1.1 Background

Sweetpotato (Ipomoea batatas L.) is a starchy tuberous crop from the family Convolvulaceae. On a worldwide scale, sweetpotato's economic importance among all food crops is only exceeded by cereals such as wheat and rice (Odondo et al., 2014). Sweetpotato is also an important food security crop, often crucial during famine due to its excellent drought tolerance and ability to grow on marginal lands (Mukhopadhyay et al., 2011). Projections by the International Food Policy Research Institute indicate that among the root and tuber crops, the value of sweetpotato is estimated to increase by USD 23 billion between 2017 and 2050 (International Food Policy Research Institute [IFPRI], 2022). Over the same period, growth in global demand for sweetpotato is estimated to increase by 20 metric tons. According to Food and Agricultural Organization Statistical Databases (FAOSTAT, 2019), Uganda is the fifth largest sweetpotato producer after China, Nigeria, Tanzania, and Indonesia.

However, Uganda's sweetpotato productivity is constrained by several factors, including sweetpotato virus disease (SPVD), sweetpotato weevils, and vine shortages caused by prolonged dry seasons, among others (Namanda et al., 2019; Okonya et al., 2014). Sweetpotato breeding objectives in the country have over time been largely based on yield enhancement with a strong focus on virus resistance (Yada et al., 2013). Also, given the currently high prevalence of vitamin A deficiency (VAD) in Uganda, the sweetpotato breeding program has directed significant investments toward developing vitamin A-rich orange-fleshed sweetpotato (OFSP) to combat VAD, especially through projects such as the Sweetpotato Profit and Health Initiative (SPHI) (Low et al., 2017).

Over the last decade, several improved sweetpotato varieties, both OFSP and non-OFSP, have been officially released by the National Sweetpotato Program in Uganda (Ssemakula et al., 2013). By 2017, it was estimated that eight new OFSP varieties bred in Uganda or introduced from other countries were released under the SPHI for use by farmers (Low et al., 2022). Most recent data indicates that, overall, there are 27 released orange and non-orange sweetpotato varieties in Uganda, all largely locally bred and a few registered landraces (Sseruwu et al., 2020). Nevertheless, many farmers continue to grow a large number of local landraces (also known as farmer varieties), most of which are low yielders, do not contain much vitamin A, and are less resistant to pests and virus diseases (Mwanga et al., 2011). Zawedde et al. (2014) estimated that a farmer in Uganda maintains an average of four varieties in his/her sweetpotato plot in a given season, and that there were more than 3000 varieties of sweetpotato in Uganda, many of them local landraces.

Preference for sweetpotato traits among farmers in Uganda is quite diverse and depends on several factors (Mwanga et al., 2021). These factors include the agronomic, technical, and socio-cultural context of the modes of production and processing (Zawedde et al., 2014). Among other factors, gender plays a key role in explaining the differences in trait preferences among farmers (Christinck et al., 2017). Polar and Demont (2022), for instance, indicate that tend to have a preference for quality traits while men prefer agronomic traits. Thiele et al. (2021) argue that a lack of attention to gender differences in trait preference has played a role in the low uptake of improved varieties of root and tuber crops including sweetpotato. The gender-differentiated trait preference emanates from the differences in the roles played by males and females in various crop production or postharvest activities. Specifically, women's preferences tend to focus on production and use/consumption-related traits, whereas men's trait preferences relate more to production and marketing (Weltziem et al., 2019). To tailor breeding programs for a diversity of users, Polar and Demont (2022) recommend the need for a gendered analysis of trait preferences. This study therefore explored male and female farmers’ sweetpotato trait preferences. It examined the factors affecting trait preferences of sweetpotato growers. Specifically, the study assessed the effect of gender, personal, household, and farm-specific variables on farmers’ choice of the most preferred sweetpotato traits.

Core Ideas

• While desirable, production-oriented sweetpotato traits like yield are not always the most farmer-desired traits.

• Farmers' preference for sweetpotato traits is heterogenous and varies based on age, education level, and variety grown.

• Breeding should take a more pluralistic approach with regard to the combination of traits prioritized.

1.2 Study context

The study focused on sweetpotato because it is a major staple crop in Uganda with a per capita consumption of 83 kg/year, by the vast majority of rural households, making it a major food security crop (Bashaasha et al., 1995; Ingabire & Vasanthakaalam, 2011; Yada, 2009; Yanggen & Nagujja, 2006). Uganda is the secondary center of diversity in sweetpotato varieties with more than 3000 known genotypes (Zawedde et al., 2014). In addition, sweetpotato growers, most of whom are female and smallholders in nature, face multiple constraints (both biotic and abiotic) that they have to pay attention to, in addition to consumption traits, in the selection of the varieties to plant. Thus, sweetpotato and Uganda provide an interesting case to study.

We conducted the study within the framework of the demand-driven sweetpotato breeding project implemented by the International Potato Center (CIP) (see Okello et al. (2022) for a description of the project). The project was designed to investigate farmers’ preferences for sweetpotato varietal traits. The study respondents were sweetpotato growers in Mpigi district in Central Uganda. The region is a hotspot for SPVD and has been the focus of breeding and field testing of clones bred for SPVD resistance. Varieties released specifically for the central region are therefore resistant to SPVD while also having other essential traits especially high-yielding ability.

To generate information about trait preferences, we asked farmers to choose their single most preferred trait from any of the three broad categories of desirable traits described as (1) production-oriented traits, (2) risk-averting traits, and (3) quality traits. Notably, preferences for crop traits vary between farmers based on gender and other socio-economic factors under which farmers operate (Christinck et al., 2017).

2 STUDY METHODS

2.1 Theoretical framework

This study has a theoretical grounding in Lancaster's model of consumer choice (Lancaster, 1966) and the random utility models (McFadden, 1974). Lancaster postulated that consumers derive satisfaction or utility maximization not only from the good itself but also from the attributes or characteristics of the good (Okello et al., 2015). They strive to attain a product with the attributes they most desire subject to a budget constraint. These attributes are different for male and female consumers (Mwanga et al., 2021). This implies that the utility of a good is derived from traits or attributes of a good rather than the good itself (Lancaster, 1966). Lancaster's approach provides a rationale for dividing a product into several attributes. Thus, male and female consumers’ preferences for a product can be studied in terms of their preferences for each attribute, which in this case are the sweetpotato traits. Given the possibility of making errors in measuring people's subjective views on the values of the attributes, this study also adopted random utility theory (Bhat, 2002; Lancaster, 1966).

With the underlying assumption of utility maximization, random utility models are specified based on maximum likelihood estimation (Bhat, 2002). The multinomial probit (MNP) model for a categorical dependent variable with unordered outcomes was fitted. The MNP relaxes the assumption of independence from irrelevant alternatives (IIA), thus allowing the potential interdependence of trait preferences (Greene, 2018). It is based on the assumption that the values of the dependent variable are irrelevant. In addition, under the probit model specification, the random terms, often called error terms, are assumed to be independent, identical, and standard normal.

2.2 Conceptual framework

The conceptual framework for this study is derived based on the above theories. The primary focus of this framework is to identify the underlying factors that influence choice behavior for sweetpotato traits by male and female farmers. In this case, where the decision maker is a farmer, a choice decision can be viewed as a decision-making process linked to factors such as farmer, household, and farm-specific characteristics, as presented in Figure 1. Given that the most preferred sweetpotato trait is a discrete choice good comprising several traits with the potential to meet several objectives, the decision maker's problem is the choice of a sweetpotato trait that best maximizes his/her utility from a list of desirable sweetpotato traits.

FIGURE 1

Conceptual framework of most preferred sweetpotato trait choice. SPVD, sweetpotato virus disease.

Following Lancaster (1966), the utility derived from sweetpotato as a good is perceived to be determined by a set of alternative traits of sweetpotato. The decision maker (i.e., the male or female farmer) is assumed to assign a utility value for each alternative by valuing and trading off the traits that are important in his/her choice decision. By so doing, the farmer is assumed to exhibit a utility-maximizing behavior in his/her choice of a most preferred trait. That is, the farmer chooses a trait with the highest positive utility value as most preferred. Since the sources of utility are strictly linked to the farmer, household, and farm-specific characteristics, these factors were included to explicitly account for the observed preference heterogeneity.

The economic model for the discrete choice framework for sweetpotato traits considers unobserved heterogeneity presented as follows. Each individual's choice set is assumed to have a finite set of J mutually exclusive and exhaustive alternative sweetpotato traits to choose from. For each choice situation, a sampled decision-maker or farmer is assumed to have full knowledge of the factors that influence his/her choice decision when asked to choose the most preferred sweetpotato trait from the competing J alternatives subject to the budget constraint.

Following the random utility theory, an individual n receives utility U from choosing an alternative trait j from a choice set with J alternative traits, if and only if this alternative generates at least as much utility as any other alternative trait.

2.3 Sampling procedure and data

This study used data collected from sweetpotato growers in Mpigi district in Central Uganda. This district was purposively selected because it is a leading sweetpotato-producing district and has hosted many varietal testing and evaluation breeding experiments in the past (Mwanga et al., 2011). Sampling was done in three stages using a multistage sampling technique. First, all the eight sub-counties in Mpigi district were selected. Then, the probability proportionate to size sampling method was used to randomly select 24 parishes and 48 villages from the selected parishes. In each selected village, the Local Council I chairperson and other knowledgeable village members, including three to five sweetpotato farmers, were engaged in the development of a sampling list of all sweetpotato growers in their villages. The focus was on male and female farmers who cultivated sweetpotato in at least one season of 2019. Next, 20 farmers in each village were randomly selected from the pre-generated sampling lists to participate in the quantitative survey.

A total of 377 female and 415 male farmers were interviewed. Data was collected through individual in-depth interviews in February 2020 using pretested questionnaires programmed in 1000minds decision-making and conjoint analysis software (Okello et al., 2022). The data collected included information about individual sweetpotato trait preferences of male and female farmers and their personal, household, as well as farm-specific characteristics. The questionnaire was written in English, and interviews were conducted by trained enumerators in the local language (Luganda) or English, depending on the respondent's preference.

2.4 Empirical and analytical methods

The study used univariate, bivariate, and multivariate analytical approaches to assess the effect of personal, farm household, and farm-specific characteristics on farmers’ choice of the most preferred sweetpotato traits. Univariate analysis involved computing the number of respondents (N), means, minimum, and maximum, as well as the standard deviation, which shows the deviation of the values of each variable from the mean. However, the bivariate analysis involved the computation of the Pearson correlation coefficients ($${r_{xy}}$$) between all variables—the dependent variable and independent variables. Multivariate analysis is elaborated below.

2.4.1 Multivariate analysis

To elicit preference for specific traits, each respondent was asked during the survey to choose his/her single most preferred sweetpotato trait. These choices were entered as dummy variables, with “1” if a trait is selected and “0” otherwise. After consultations with the breeders and social scientists, the traits were grouped into three broad categories presented in Table 1, namely: (i) production-oriented traits—these are agronomic traits that influence crop productivity; (ii) risk-averting traits—these traits that foster the crop's ability to withstand challenging environments, hence reducing the risk of loss to farmers; and (iii) quality traits—the organoleptic traits that assure good eating quality and outward/visual/sensory appeal of the roots to consumers. The dependent variable is therefore categorical, relating to which category the respondent's most preferred variety falls under.

TABLE 1. Varietal traits used to compute the dependent variable in the regression equation.

Traits	Description
Production-oriented traits (include agronomic traits that influence crop productivity)
1. High root yield	More number of bags of roots per acre
2. High vine yield	More bags of vines per acre
3. Bulking ability	The ability of the roots to enlarge
4. Early maturity	Less number of days from planting of the vines to full maturity of the roots
Risk-averting traits (include traits that foster the crop's ability to withstand challenging environments, hence reducing the risk of loss to farmers)
5. Underground root storage longevity	The ability of the roots to remain marketable for more days after maturity when stored in soil
6. Disease resistance (SPVD)	Ability to produce yield under SPVD infestation
7. Stress tolerance (drought, poor soils)^a	The ability of the plants to survive during drought and establish in low fertility soils
8. Vine survival under stress	The ability of the vines to withstand stressful conditions like drought
Quality traits (include the organoleptic traits that assure good eating quality and outward/visual/sensory appeal of the roots to consumers)
9. Root sweetness	The sugariness of the root
10. Root mealiness	The dry mouth feel when eating a boiled root
11. Nutritional and health benefits	The ability of the root to provide vitamin A as indicated by the intensity of the orange color in the root flesh
12. Hard root texture	Firmness of the root after boiling
13. Root skin color	Color of the root skin
14. Root size	Size of the roots by appearance
15. Root shape	Shape of the roots by appearance
16. Root smoothness	Smoothness of the root skin texture

• Abbreviation: SPVD, sweetpotato virus disease.
• ^a Stress tolerance (drought, poor soils) = tolerance to stressful conditions caused by drought and poor soils.

To assess factors influencing the choice of most preferred sweetpotato traits, a MNP regression model was estimated using variables described in Table 2. Oftentimes, the MNP model is adopted to avoid the limitations of the multinomial logit model (MNL)–IIA (Wooldridge, 2019). The paper uses one, two, and three trait choice categories to denote production-oriented traits, risk-averting traits, and quality traits, respectively. Production-oriented trait choice (option 1) is chosen as the base category. In doing so, the utilities of other trait choice categories (risk-averting and quality traits) are compared to those of the base category. A farmer's decision is grounded on the utility derived from the alternative trait choice categories. Assuming that a farmer i is faced with j alternative choices, based on the utilities linear parameters, a MNP is represented as follows:

$$$$\begin{equation}{U_{ij}} = X_{ij}^{\prime}\;\beta + {\epsilon_{ij}},{\epsilon_{ij}} \sim N\;\left[ {0,\sum } \right]\end{equation}$$$$(1)

$$$$\begin{eqnarray}{\mathrm{\;}}{Y_{ij}} &=& \;1\;{\mathrm{if}}\;{U_{i1}} \le {U_{ij}}\;{\mathrm{for}}\;i\; = \;1,2,3, \ldots ,n;j\;\nonumber\\ &=& \;1,2,3, \ldots ,n;\;{\mathrm{and\;}}0\; = \;{\mathrm{otherwise}}\end{eqnarray}$$$$(2)

TABLE 2. Description and measurement of the study variables.

Variable name	Description	Hypothesis
Dependent variable
Farmer choice of most preferred sweetpotato trait	(Production-oriented traits = 1, risk-averting traits = 2, and quality traits = 3)
Explanatory variables
Farmer characteristics
Gender	Sex of the respondent (1 = male; 0 = female)	+
Age	Age of respondent in completed years (count)	+
Educy	Years of formal education completed by respondent (count)	+
Household specific characteristics
hHsize	Number of members in the household (count)	+
Chld5	Dummy for having children under 5 years in the household (1 = yes, 0 = no)	+
Farm specific characteristics
Pdnarea	Size of cultivable land area normally used to produce sweetpotato (acres)	+
Variety	Dummy for growing local sweetpotato varieties by the farmer (1 = local; 0 = improved)	+
Qtysold	Amount of sweetpotato normally sold per season (100-kg bags)	+
Nostorage	Dummy variable for nonuse of sweetpotato storage facilities (1 = do not store; 0 = store)	+
Pdnmtd	Dummy variable for using rainfed production method to produce sweetpotato (1 = rainfed; 0 = not rainfed)	+

$${Y_{ij}}$$ signifies the choice made by farmer i of choosing a trait in a certain category as their most preferred trait.$$\;{U_{ij}}$$ represents unobservable alternative utility j as observed by farmer i. $$X_{ij}^{\prime}$$ signifies a vector of explanatory variables featuring both i and j. $$\beta $$ refers to coefficient estimates of the regressed variables (farmer, household, and farm-specific characteristics), whereas $${\varepsilon _{ij}}$$ is the error term assumed to be normally distributed with mean zero and variance sigma. Hence, the anticipated probability of a farmer choosing any of the trait choice categories is illustrated as follows:

$$$$\begin{eqnarray}{p\hspace*{0.28em}\left(\hspace*{0.28em}{Y}_{i}=\hspace*{0.28em}1\right)} &=& {\hspace*{0.28em}p({U}_{i1}+{\mathrm{\varepsilon}}_{i1}>{U}_{i2}+{\mathrm{\varepsilon}}_{i2}\hspace*{0.28em}\mathrm{and}\hspace*{0.28em}{U}_{i1}} \nonumber\\ && {+\, {\mathrm{\varepsilon}}_{i1}>{U}_{i3}+{\mathrm{\varepsilon}}_{i3}}\end{eqnarray}$$$$(3)

$$$$\begin{eqnarray}{p\hspace*{0.28em}\left(\hspace*{0.28em}{Y}_{i}=\hspace*{0.28em}2\right)} &=& {\hspace*{0.28em}p({U}_{i2}+{\mathrm{\varepsilon}}_{i2}>{U}_{i1}+{\mathrm{\varepsilon}}_{i1}\hspace*{0.28em}\mathrm{and}\hspace*{0.28em}{U}_{i2}} \nonumber\\ && {+\, {\mathrm{\varepsilon}}_{i2}>{U}_{i3}+{\mathrm{\varepsilon}}_{i3}}\end{eqnarray}$$$$(4)

$$$$\begin{eqnarray}{p\hspace*{0.28em}\left(\hspace*{0.28em}{Y}_{i}=\hspace*{0.28em}3\right)} &=& {\hspace*{0.28em}p({U}_{i3}+{\mathrm{\varepsilon}}_{i3}>{U}_{i1}+{\mathrm{\varepsilon}}_{i1}\hspace*{0.28em}\mathrm{and}\hspace*{0.28em}{U}_{i3}} \nonumber\\ && {+\, {\mathrm{\varepsilon}}_{i3}>{U}_{i2}+{\mathrm{\varepsilon}}_{i2}}\end{eqnarray}$$$$(5)

Assuming that the choice categories are mutually exclusive, then $$\sum _{j-1}^{j}{p}_{\textit{ij}}=\hspace*{0.28em}1$$. Hence, for each farmer i, the probabilities add up to one.

3 RESULTS AND DISCUSSION

This section presents and discusses the results of the study. It presents choices of the most preferred sweetpotato traits by farmers. It also presents a summary of the personal, household, and farm-specific characteristics of respondents and their influence on the choice of most preferred traits based on bivariate and multivariate analyses.

3.1 Key preferred sweetpotato traits by study respondents

Table 3 presents a list of 16 traits identified among the most preferred sweetpotato traits based on the individual choices of male and female farmers and their proportional mean distribution. Using the t-test, differences among the group means of males and females were determined. It was found that preference for only one trait (stress tolerance [drought, poor soils]) statistically differs between males and females. Further, preference scores of different traits were compared within the gender categories. Results show that, among the production-oriented traits, high root yield was the most frequently selected production-oriented trait by both male and female respondents. It was selected by 46.9% and 45.3% of the female (n = 377) and male (n = 415) respondents, respectively. Pooled data also showed that the most frequently selected trait was high root yield, selected by 46.1% of the 792 respondents.

TABLE 3. Descriptive statistics of dependent variable categories, by gender of the respondent.

Variables	Female respondents (n = 377)				Male respondents (n = 415)				Pooled (N = 792)				Group differences
	Mean	Min	Max	SD	Mean	Min	Max	SD	Mean	Min	Max	SD	t-statistic	p-value
Production-oriented traits (1 = yes; 0 = no)
High root yield	0.469a	0	1	0.499	0.453a	0	1	0.498	0.461a	0	1	0.499	0.4643	0.6426
High vine yield	0.003	0	1	0.052	0.005	0	1	0.069	0.004	0	1	0.061	−0.4952	0.6206
Bulking ability	0.095c	0	1	0.294	0.092c	0	1	0.289	0.093d	0	1	0.291	0.1893	0.8499
Early maturity	0.016	0	1	0.125	0.017	0	1	0.129	0.016	0	1	0.127	−0.1052	0.9162
Risk-averting traits (1 = yes; 0 = no)
Underground root storage longevity	0.013	0	1	0.115	0.024	0	1	0.154	0.019	0	1	0.136	−1.1165	0.2645
Disease resistance (SPVD)	0.042	0	1	0.202	0.043	0	1	0.204	0.043	0	1	0.203	−0.0646	0.9485
Stress tolerance (drought, Poor soils)	0.095c	0	1	0.294	0.135b	0	1	0.342	0.116b	0	1	0.301	−1.7315*	0.0838
Vine survival under stress	0.119b	0	1	0.325	0.092c	0	1	0.289	0.105c	0	1	0.306	1.2752	0.2026
Quality traits (1 = yes; 0 = no)
Root sweetness	0.058d	0	1	0.235	0.055d	0	1	0.229	0.057e	0	1	0.232	0.1779	0.8588
Root mealiness	0.034	0	1	0.183	0.029	0	1	0.168	0.032	0	1	0.175	0.4470	0.6550
Nutritional and health benefits	0.034	0	1	0.183	0.034	0	1	0.181	0.034	0	1	0.182	0.0578	0.9539
Hard root texture	0.011	0	1	0.103	0.009	0	1	0.098	0.010	0	1	0.100	0.1364	0.8916
Root skin color	0.005	0	1	0.073	0.002	0	1	0.049	0.004	0	1	0.061	0.6618	0.5083
Root size	0.000	0	1	0.000	0.002	0	1	0.049	0.001	0	1	0.036	−0.9531	0.3408
Root shape	0.000	0	1	0.000	0.002	0	1	0.049	0.001	0	1	0.036	−0.9531	0.3429
Root smoothness	0.003	0	1	0.052	0.005	0	1	0.069	0.004	0	1	0.061	−0.4952	0.6206
Grouped-traits (1 = yes; 0 = no)
Production-oriented traits	0.567	0	1	0.496	0.549	0	1	0.498	0.558	0	1	0.496	0.5157	0.6062
Risk-averting traits	0.286	0	1	0.452	0.310	0	1	0.463	0.299	0	1	0.458	−0.7473	0.4551
Quality traits	0.145	0	1	0.353	0.139	0	1	0.347	0.142	0	1	0.349	0.2460	0.8057

• Note: Lowercase letters represent the five topmost preferred traits by the respective gender categories in chronological order. Bold values indicate the five top most preferred sweetpotato traits based on mean preference scores.
• Abbreviations: Max, maximum; Min, minimum; SD, standard deviation.
• *Significant at 10%.

For the risk-averting traits, the most frequently selected trait for female farmers was vine survival under stressful conditions, which was selected by 11.9% of the 377 respondents. Tolerance to stressful conditions caused by drought and poor soils was the more frequently selected trait for males than females. It was selected by 13.5% of the male respondents. For the pooled data, the most frequently selected trait was also tolerance to stressful conditions caused by drought and poor soils.

The most frequently selected trait among the quality traits by both female and male respondents was root sweetness. It was selected by 5.8% (n = 377) and 5.5% (n = 415) of the female and male respondents, respectively. Pooled data also showed that root sweetness was the most frequently selected quality trait by 5.7% of the 792 respondents.

The between-groups analysis indicates that, overall, the most frequently selected trait by the 377 female respondents was high root yield (46.9%). Similarly, the most frequently selected trait by the 415 male respondents was high root yield (45.3%).

3.2 Factors affecting the choice of most preferred sweetpotato traits by farmers in Uganda

This section presents summary statistics of the variables that are likely to affect the trait preferences of sweetpotato farmers based on literature and a priori expectations. It also presents bivariate analysis results as well as prediction results of MNP regression.

3.2.1 Descriptive statistics of the independent variables

A total of 792 sweetpotato growers took part in the survey. According to the results in Table 4, the majority of the respondents were males (n = 415) while the rest were females (n = 377). The age of the respondents ranged from 18 years to 80 years. The mean age was significantly different between the gender groups (p < 0.01). The average age of female respondents was 47 years, and their male counterparts had a mean age of 44 years. We also found a significant difference (p < 0.01) between male and female participants with respect to education. The education for male and female respondents was approximately 8 years and 7 years, respectively. Household sizes ranged from 1 to 30 members. Female respondents had, on average, larger household sizes of about seven members compared to males with about six members. The majority of the respondents in this study had children under 5 years of age. The proportion of female and male respondents with children under 5 years of age was 62% and 66%, respectively.

TABLE 4. Independent variables and their descriptive statistics based on gender.

Variables	Females					Males					Pooled					Group differences
	n	Mean	Min	Max	SD	n	Mean	Min	Max	SD	N	Mean	Min	Max	SD	t-statistic	p-value
Farmer characteristics
Age (years)	371	47.28	20	80	13.03	418	44.10	18	80	13.82	789	45.59	18	80	13.54	3.3184**	0.0009
Education (years)	372	6.84	0	16	3.24	418	7.89	0	18	3.37	790	7.40	0	18	3.35	−4.4703**	0.0000
Household characteristics
Household size (no. of people)	371	6.64	1	21	3.44	417	6.32	1	30	3.92	788	6.47	1	30	3.71	1.2398	0.2154
Children under 5 years (1 = yes)	372	0.65	0	1	0.47	418	0.68	0	1	0.46	790	0.67	0	1	0.47	−0.9959	0.3196
Farm characteristics
Area planted (acres)	372	0.64	0.025	5	0.61	417	0.97	0.025	18	1.40	789	0.81	0.025	18	1.11	−4.1255**	0.0000
Quantity sold (100-kg bags)	352	5.02	0	150	10.92	411	9.28	0	200	16.67	763	7.31	0	200	14.45	−4.0915**	0.0000
Variety grown (1 = local; 0 = improved)	308	0.37	0	1	0.48	349	0.36	0	1	0.48	657	0.37	0	1	0.48	−0.0991	0.9211
No storage (1 = do not store; 0 = store)	373	0.13	0	1	0.34	417	0.19	0	1	0.39	790	0.16	0	1	0.37	−2.2756*	0.0231
Production method (1 = rainfed; 0 = not rainfed)	370	0.66	0	1	0.47	413	0.55	0	1	0.49	783	0.61	0	1	0.48	3.1162**	0.0019

• Abbreviations: Max, maximum; Min, minimum; SD, standard deviation.
• * and ** denote significance at 5% level (p < 0.05) and 1% level (p < 0.01), respectively.

The average sweetpotato production area per season was significantly different between male and female respondents. Male farmers had a larger average sweetpotato production area (0.97 acres) per season compared to their female counterparts with only 0.64 acres. Similarly, the quantity sold of sweetpotato by the male respondents was significantly larger than that of the female respondents (p < 0.01). Male farmers sold, on average, nine 100-kg bags of sweetpotato compared to five 100-kg bags for their female counterparts. Only 29% of both male and female respondents were growing local sweetpotato varieties at the time of this study. We also found a significant difference in the percentage of female and male respondents who did not store sweetpotato (p < 0.05). More male respondents (18%) did not store sweetpotato compared to their female counterparts (12%). The percentage of females using the traditional rainfed production method (63%) was significantly larger (p < 0.01) than that of males (53%).

3.2.2 Bivariate analysis results

Pearson correlation coefficients for all the variables were computed at a 1% significance level using a correlation matrix. Results are presented in Table 5. The signs on the coefficients indicate the direction of association between the variables, and the coefficients show the magnitude of the relationship between the variables. The purpose of this analysis was to determine the degree of association between the variables.

TABLE 5. Pearson correlation coefficients between the variables.

Correlation	Production-oriented traits (1 = yes; 0 = no)	Risk-averting traits (1 = yes; 0 = no)	Quality traits (1 = yes; 0 = no)	Sex (1 = male; 0 = female)	InAge (years)	InEducation (years)	InHousehold size (no. of people)	Children under 5 years (1 = yes; 0 = no)	InArea planted (acres)	InQuantity sold (100-kg bags)	Variety grown (1 = local; 0 = improved)	No storage (1 = do not store; 0 = store)	Production method (1 = rainfed; 0 = not rainfed)
Production-oriented traits (1 = yes; 0 = no)	1.00
Risk-averting traits (1 = yes; 0 = no)	−0.72^*	1.00
Quality traits (1 = yes; 0 = no)	−0.47^*	−0.25^*	1.00
Gender (1 = male; 0 = female)	−0.01	0.02	−0.01	1.00
InAge (years)	0.08	−0.09^*	−0.00	−0.13^*	1.00
InEducation (years)	−0.06	0.09^*	−0.02	0.14^*	−0.17^*	1.00
InHousehold size (no. of people)	0.04	−0.07	0.03	−0.08	0.22^*	0.04	1.00
Children under 5 years (1 = yes; 0 = no)	0.03	−0.01	−0.02	0.03	−0.08	0.08	0.40^*	1.00
InArea planted (acres)	0.03	−0.01	−0.02	0.21^*	−0.03	0.14^*	0.15^*	0.15^*	1.00
InQuantity sold (100-kg bags)	0.06	−0.01	−0.06	0.18^*	−0.17^*	0.13^*	0.06	0.11^*	0.61^*	1.00
Variety grown (1 = local; 0 = improved)	−0.08	−0.02	0.15^*	−0.00	0.01	−0.12^*	−0.10^*	−0.08	−0.13^*	−0.17^*	1.00
No storage (1 = do not store; 0 = store)	0.15*	−0.07*	−0.11*	0.08	−0.01	0.07	−0.05	−0.01	0.08	0.19^*	−0.13^*	1.00
Production method (1 = rainfed; 0 = not rainfed)	−0.02	0.04	−0.02	−0.11^*	0.07	−0.07	−0.09^*	−0.03	−0.19^*	−0.30^*	0.15^*	0.01	1.00

• ^* Significant at 1% level (p < 0.01).

Based on the results in Table 5, most of the coefficients are very low, which indicates that the relationship between variables is very weak or nonexistent. Sweetpotato storage nonuse is significantly (p < 0.01) associated with the choice of production-oriented traits as the most preferred traits. It has a weak positive and statistically significant relationship of 0.15 with choice of production-oriented traits. Further, variables of age, education level, and sweetpotato storage nonuse were significantly (p < 0.01) associated with the choice of risk-averting traits as the most preferred traits. Age and sweetpotato storage nonuse had weak negative and significant relationships of 0.09 and 0.07, respectively, with the choice of risk-averting traits. On the other hand, there was a weak positive and significant association of 0.09 between farmers’ education level and choice of risk-averting traits.

It was also found that variables, such as variety grown and sweetpotato storage nonuse, were significantly associated with the choice of quality traits as most preferred traits. There was a weak positive and statistically significant association of 0.15 between farmers growing local varieties and choice of quality traits. However, sweetpotato storage nonuse had a weak negative and statistically significant association of 0.11 with the choice of quality traits.

Results also indicate a reverse relationship between production and risk-averting traits. This is likely because of the inherent correlation between yield/output and weather effects, which contribute to risk-related traits. For instance, agronomists indicate that when there is a shortage of rainfall during critical stages, such as root formation and filling in sweetpotato, yield is adversely affected. Specifically, a dry spell during root filling results in long slender roots that are not marketable usually referred to as pencil roots.

3.2.3 Multivariate analysis results

Using the choice of a production-oriented trait as the base outcome, the MNP regression model was estimated, and the coefficient results are presented in Table 6. To control for any possible cases of heteroscedasticity, robust standard errors were used. Also, because data does not usually conform exactly to the theory underlying the model during analysis, some diagnostic tests were performed before proceeding with the estimation of the MNP regression equation to ensure econometric adequacy (Greene, 2018). These tests and their results are elaborated below.

TABLE 6. Variance inflation factor (VIF) results.

Variables	VIF
Sex (1 = male; 0 = female)	1.09
InAge (years)	1.16
InEducation (years)	1.09
InHousehold size (no. of people)	1.38
Children under 5 years (1 = yes; 0 = no)	1.28
InQuantity sold (100-kg bags)	1.92
Variety grown (1 = local; 0 = improved)	1.08
No storage (1 = do not store; 0 = store)	1.08
Production method (1 = rainfed; 0 = not rainfed)	1.15

• Note: Mean VIF = 1.30.

Overall model fit

This was assessed using the Wald test. This test yielded a chi-square and p-value of 42.35 and 0.0025, respectively. These results indicate that the coefficients are not simultaneously equal to zero. Hence, statistically significant predictors in the model lead to better prediction and better model fit.

Test of multicollinearity

The variance inflation factor (VIF) was used to detect multicollinearity. Lower VIF values indicate an absence of multicollinearity. The VIFs of predictor variables in this study were all less than 5, as shown in Table 6. Thus, there was no multicollinearity problem in the data.

Determinants of the choice of risk-averting relative to production-oriented sweetpotato traits

Results from the analysis of factors affecting the likelihood of choosing risk-averting traits relative to production-oriented traits are presented in columns 2, 4, and 6 of Table 7. They show that education level and nonuse of underground storage affect the decision to choose risk-averting traits relative to production-oriented traits.

TABLE 7. Gender differentiated coefficient (Coef) estimates from multinomial probit regression on the choice of most preferred sweetpotato traits.

	Females (n = 298)		Males (n = 337)		Pooled (N = 635)
	Risk-averting traits	Quality traits	Risk-averting traits	Quality traits	Risk-averting traits	Quality traits
Explanatory variable	Coef (RSE)	Coef (RSE)	Coef (RSE)	Coef (RSE)	Coef (RSE)	Coef (RSE)
Farmer characteristics
Gender (1 = male; 0 = female)	–	–	–	–	0.155 (0.156)	0.081 (0.180)
InAge (years)	−0.356 (0.395)	−0.016 (0.481)	−0.403 (0.376)	−0.563 (0.427)	−0.374 (0.269)	−0.362 (0.312)
InEducation (years)	0.336 (0.231)	0.206 (0.267)	0.309 (0.215)	−0.115 (0.225)	0.335** (0.155)	0.032 (0.172)
Household characteristics
InHousehold size (no. of people)	−0.310 (0.240)	0.197 (0.302)	−0.076 (0.207)	0.106 (0.227)	−0.175 (0.152)	0.133 (0.182)
Children under 5 years (1 = yes; 0 = no)	0.279 (0.264)	0.057 (0.311)	−0.187 (0.255)	−0.408 (0.293)	0.040 (0.182)	−0.165 (0.208)
Farm characteristics
InArea planted (acres)	0.223 (0.538)	0.940* (0.544)	−0.069 (0.361)	0.039 (0.412)	0.010 (0.305)	0.321 (0.326)
InQuantity sold (100-kg bags)	−0.016 (0.148)	−0.220 (0.163)	−0.082 (0.115)	−0.138 (0.137)	−0.050 (0.090)	−0.155 (0.103)
Variety grown (1 = local; 0 = improved)	−0.169 (0.239)	0.685** (0.267)	0.101 (0.222)	0.392 (0.253)	−0.016 (0.161)	0.558*** (0.182)
No storage (1 = do not store; 0 = store)	−0.857** (0.400)	−0.836* (0.490)	−0.526* (0.282)	−0.830** (0.364)	−0.656*** (0.226)	−0.779*** (0.290)
Production method (1 = rainfed; 0 = not rainfed)	0.224 (0.248)	−0.496* (0.280)	0.148 (0.229)	0.228 (0.264)	0.192 (0.165)	−0.132 (0.191)
Constant	0.307 (1.728)	−1.873 (1.958)	0.772 (1.502)	1.468 (1.759)	0.399 (1.117)	0.111 (1.272)
	Wald χ^2 (18) = 34.42		Wald χ^2 (18) = 25.43		Wald χ^2 (18) = 25.43
	Prob > χ^2 = 0.0112		Prob > χ^2 = 0.1136		Prob > χ^2 = 0.1136
	Log pseudolikelihood = −268.67		Log pseudolikelihood = −315.77		Log pseudolikelihood = −315.77
Base category = Production-oriented traits

• Note: Robust standard errors (RSE) are given in parentheses. The bold values indicate coefficients that were statistically significant.
• *, **, and *** denote significance at 10% level (p < 0.10), 5% level (p < 0.05), and 1% level (p < 0.01), respectively.

Estimation results of the pooled model show that as years of formal education increase, the probability of choosing risk-averting traits instead of production-oriented traits increases. This is probably because more educated farmers are better able to access and process information on the abiotic and biotic stress factors and weigh the returns of growing more resilient sweetpotato varieties.

On the other hand, the non-storage of sweetpotato reduces the probability of choosing risk-averting traits relative to production-oriented traits in the pooled model. Similar results are valid for both male and female models.

Determinants of the choice of quality traits relative to production traits

Factors affecting the decision to choose quality traits relative to production-oriented traits are presented in columns 3, 5, and 7 of Table 7. Four farm-specific characteristics, namely, the size of the cultivable land area normally used to produce sweetpotato, the variety grown, sweetpotato storage nonuse, and the production method, affect the decision to choose quality traits as the most preferred sweetpotato traits relative to production-oriented traits.

Holding other factors constant, results in the pooled model show that growing local varieties increases the probability of choosing quality traits rather than production-oriented traits. This is likely because, over the years, local varieties have been passed down owing to specific desired sensory characteristics, mostly consumption and eating quality, along with traditional cooking recipes (Ficiciyan et al., 2018). These recipes require specific qualities in sweetpotato, which may not be in the newly bred varieties. For instance, sweetpotato preparation in Uganda involves cooking the roots stacked along with other foods often for a long period of time. This practice requires roots that remain hard/firm after cooking, a characteristic demanded by the majority of consumers (Mwanga et al., 2021). Hence, farmers may continue growing local varieties that have this trait (i.e., hardness after cooking) even when they have lower yield.

Pooled model results further show that the non-storage of sweetpotato reduces the probability of choosing quality traits relative to production-oriented traits. These results allow us to assess the effect of gender on the decision by farmers to choose quality and risk-averting traits relative to production-oriented traits. They show that while the direction of effect is positive, the coefficient of gender variable for both quality trait and risk-averting trait is statistically insignificant at any level of statistical confidence. These findings suggest that after controlling for other personal, household, and farm characteristics, the sex of the farmer has no effect on the decision to choose quality trait and risk-averting trait as the most important trait considered when selecting sweetpotato variety to plant.

Results, however, show that the non-storage of sweetpotato reduces the probability of choosing quality traits relative to production-oriented traits in both the male and female models. This effect is the same for female and male farmers, suggesting that, for both groups of farmers, non-storage of sweetpotato has a negative effect on the choice of quality and risk-averting traits relative to production traits. Other things constant, the likelihood of a female farmer who does not store sweetpotato roots choosing risk-averting and quality traits relative to production traits decreases by a highly significant amount. We find similar results for the choice of risk-averting and quality traits among male farmers who do not store sweetpotato roots. Farmers who do not store sweetpotato are more likely to sell most of their produce at harvest. This is especially likely to be the case for farmers with medium-sized to larger sized plots who grow mainly for the market. Okello et al. (2012) found similar results for kale farmers in peri-urban Nairobi who grow kale mainly for market rather than home consumption.

Results further show that the use of rain-fed production methods affects the probability of choosing quality traits relative to production-oriented traits among female farmers. Specifically, we find evidence, albeit weak statistically, that the use of rain-fed production reduces the likelihood of female farmers choosing quality traits as the most preferred relative to production traits. There is, however, no effect of the use of rainfed production on the choice of risk-averting traits as the most important relative to production traits. The findings relating to the choice of risk-averting traits are in line with a priori expectations. Mpigi district has a bimodal rainfall that is well distributed throughout the growing season, hence weather-related risks are minimal (Zake & Hauser, 2014).

4 SUMMARY, CONCLUSION, AND IMPLICATIONS

Low productivity of smallholder agriculture is one of the major challenges facing the agricultural sector in Uganda. Current efforts by crop breeders seek to improve productivity by developing high-yielding, pest-, and disease-resistant improved varieties. However, such efforts can only be beneficial if there is widespread adoption of the improved varieties by farmers. Breeders’ awareness of farmers’ trait preferences is important for the development of improved varieties that match farmers’ demands, hence the adoption of these varieties. This study, therefore, investigated male and female farmers’ preferences for sweetpotato traits and its drivers among sweetpotato growers in Uganda. It also examined the drivers of farmers’ prioritization of particular traits over others in the selection of sweetpotato varieties to plant.

The data used in this study was collected through individual in-depth interviews conducted with 377 female and 415 male sweetpotato growers in Mpigi district of Central Uganda, one of the leading sweetpotato-producing districts. Trait preferences by female and male farmers were assessed using descriptive statistics. A MNP regression model was then estimated to examine the effect of farmers’ personal (including gender), household, and farm-specific characteristics on the choice of their most preferred sweetpotato traits. The traits in this model were categorized into production-oriented traits, risk-averting traits, and quality traits.

Findings of descriptive statistics revealed that farmers’ choices of most preferred sweetpotato traits were heterogeneous. The most frequently selected trait by both female and male respondents was high root yield. Production-oriented traits were, in general, the most frequently selected traits. Results of MNP regression showed that, after controlling other personal, household, and farm characteristics, gender has no effect on the choice of risk-averting and quality traits. However, the probability of choosing risk-averting traits relative to production-oriented traits increases with more years of education but decreases with more years of age and sweetpotato storage nonuse. Further, the probability of choosing quality traits relative to production-oriented traits is higher for both female and male farmers growing local varieties but lower for both female and male farmers who do not store sweetpotato. Regression results further showed that variables of age, education level, variety grown, and nonuse of sweetpotato storage significantly influenced farmers’ choice of most preferred sweetpotato traits.

This study therefore concludes that farmers’ preference for sweetpotato varietal traits is heterogeneous and that, while desirable, yield is not always the most preferred trait. Quality traits and risk-averting traits such as those that relate tolerance of the crop toward drought (e.g., vine survival and drought tolerance) drive farmers’ choice of variety to grow. We further conclude that gender has no statistically significant influence on the choice of risk-averting and quality traits after controlling other variables. Our findings imply that breeding should take a more pluralistic approach with regard to the combination of traits prioritized. This is because of the heterogeneous nature of farmers’ trait preferences. The results in this study are limited to the central region only, hence a case study. As a result, there is a need for further research on trait preferences among sweetpotato growers in other regions of the country.

AUTHOR CONTRIBUTIONS

Josephine Namirimu: Conceptualization; data curation; formal analysis; investigation; methodology; software; validation; visualization; writing—original draft; writing—review and editing. Julius Juma Okello: Conceptualization; funding acquisition; investigation; methodology; project administration; resources; supervision; writing—review and editing. Andrew Muganga Kizito: Conceptualization; supervision; writing—review and editing. Agnes M. Ssekiboobo: Supervision; writing—review and editing.

ACKNOWLEDGMENTS

We acknowledge that we undertook this study as part of the CGIAR Research Program on Roots, Tubers, and Bananas (RTB) under the “Demand-driven sweetpotato breeding in Uganda” project implemented by the International Potato Center (CIP). The study was supported by the CGIAR's Excellence in Breeding Platform, the Bill and Melinda Gates Foundation through its investment (OPP1213329) awarded to the International Potato Center (SweetGAINS project), the United Kingdom's Foreign, Commonwealth & Development Office (FCDO), through Grant No. 300649 (Development and delivery of biofortified crops at scale) awarded to the International Potato Center, the United States Agency for International Development, and the Bill & Melinda Gates Foundation through its investment OPP1178942 (RTBFoods-Breeding RTB products for end-user preferences), coordinated by the French Agricultural Research Centre for International Development (CIRAD), Montpellier, France. We also acknowledge additional funding from market intelligence initiative work package 1 in the revision of the final manuscript versions. TWe acknowledge contribution of Hugo Campos, Deputy Director General, Research and Innovation at CIP in conceptualization and preparation of the project proposal. Also, we acknowledge the CIP and NaCRRI breeding team for their contribution to preparation of data collection tools. Finally, we acknowlege AbacusBio international limited for the software used in the data collection process.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.