Exploring the Role of Farmer-Led Jumpstarting Project on Adoption of Orange-Fleshed Sweet Potato in Nigeria: Implications on Productivity and Poverty

Abstract

To promote the adoption of the orange-fleshed sweet potato (OFSP) among the farmers, OFSP was introduced to the farmers through a program tagged “Jumpstarting Project” (JP), a West African project initiative of Harvestplus and its partners, which was implemented in Osun and Kwara States, in Nigeria. Using household-level data in Nigeria, this paper examined how the intervention of a farmer-led multiplication–dissemination Jumpstarting Project has sustained the adoption of OFSP after years of its implementation and the consequent impact of adoption on productivity and welfare of the farmers. Our main empirical findings are as follows. First, the implementation of the Jumpstarting Project in project communities of Osun and Kwara States shows that the adoption of OFSP has increased and has been sustained over the years. Furthermore, the result of the propensity score matching confirms possible spillover between project communities and control groups. Second, estimation from the endogenous treatment effect indicates that the estimated average treatment effects on the treated indicates a yield gain for the adopters of OFSP. Third, the result of the average treatment effects on the treated from the endogenous switching probit model shows that the probability of being poor in the households that adopted OFSP would be 5.3% more if these households did not adopt OFSP. Our results suggest that immediate replication and scaling up of the Jumpstarting Project in other states in Nigeria should be implemented. This study demonstrates that to ensure the diffusion and adoption of improved technologies such as OFSP, intervention projects that promote the adoption of these technologies have a greater role to play.

1. Introduction

Nigeria’s predominantly indigenous rural areas have a long-standing history of eating staple foods. According to Harvestplus [1], an average Nigerian consumes 56 g of staple foods daily, most of which are deficient in vitamin A. This has led to vitamin A deficiency (VAD), particularly in women and young children (3). According to [2,3], 11% of rural women and more than 70% of children under the age of five are affected by VAD, which causes blindness [4], stunts growth and cognitive development, and raises the risk of various illnesses that can be fatal, particularly for young children, pregnant women, and nursing mothers.

Although crucial for preventing illness, disability, and death in susceptible populations, interventions that focus on boosting micronutrient intake through a single mechanism like supplementation are ineffective if they are not combined with complementary processes like food fortification and dietary diversification [5]. The breeding of staple crops fortified with vitamin A, such as the orange-fleshed sweet potato (OFSP), was introduced to Nigeria as a result of a lack of vitamin A in conventional staple crops [6]. The antioxidant (cancer-fighting) beta-carotene found in abundance in orange-fleshed sweet potatoes is changed into vitamin A by the human body, which is necessary for a robust immune system, healthy skin, clear eyesight, and good eye health [7]. One dietary strategy with significant potential to reduce VAD is the introduction of the orange-fleshed sweet potato, which is drought-resistant, high-yielding, and easy to grow [3]. Thus, the cultivation and consumption of OFSP could increase farmers’ income and improve farmers’ health and welfare.

The orange-fleshed sweet potato has been encouraged to be produced and consumed as part of recent attempts to combat VAD in sub-Saharan Africa and improve farmers’ output, income, and welfare [8]. These initiatives have been coordinated by the International Potato Center (CIP), HarvestPlus, and other partners like the International Institute of Tropical Agriculture (IITA). This led to the implementation of numerous past and current projects in Nigeria, Ghana, Uganda, Mozambique, Rwanda, Ethiopia, Zambia, and Malawi [8]. These projects involve educating people on the nutritional advantages of growing and eating OFSP, and they frequently support farmers’ access to high-quality OFSP planting materials [8]. One such effort, called the Jumpstarting Project (JP), was implemented in a few West African nations (Nigeria, the Republic of Benin, and Ghana etc.) [6]. The project aimed to demonstrate several demand- and market-driven sustainable strategies for encouraging production and uptake, including OFSP consumption. This comes with the ultimate objective of enhancing health status, household income, and farmers’ welfare. The populations most at risk for VAD, including homes with vulnerable children and women, were the focus of the initiative. These are households that have children under the age of 5 with women who are pregnant or breastfeeding [6]. These households received information on the advantages of OFSP and were encouraged to use clean planting materials, thus encouraging the uptake of OFSP. To promote the adoption of OFSP, the Jumpstarting Project provided planting materials to farmer groups together with organized training on proper agronomic procedures [6].

Many decades of research have gone into the body of knowledge regarding the adoption of innovations [6,9]. A large section of the research [10,11,12,13,14] focuses on the influence of socioeconomic and institutional factors on adoption decisions. There has been less focus on how the adoption of technology such as OFSP has been sustained over the years through a disseminated program even after several years of project implementation. Thus, there is a dearth of knowledge on the sustainability outcomes of the Jumpstarting Project on the adoption of OFSP necessitating the need for this research.

Furthermore, productivity constitutes a major factor influencing farmers’ decision to adopt a newly introduced technology among the smallholder farmers. Specifically, the key question arising is whether the adoption of OFSP improved the productivity of smallholder farmers. According to the literature, farmers were motivated to adopt new agricultural technologies if they increased their yield and income [15,16]. However, there is a dearth of information on the productivity and income impact of OFSP adoption among the smallholder farmers in Jumpstarting Project-implemented areas in Nigeria, also necessitating the need for this research.

In light of this, research on smallholder production and household (HH) welfare has spanned a wide range of topics over the years. An increasing amount of relevant research [17,18,19] assesses the effect of interventions on poverty and farmer productivity based on the assumption that higher farmer productivity has both direct and indirect effects on reducing poverty. The results of most projects that target smallholders indicate that higher output and better household welfare are positively correlated [20,21,22,23,24,25]. For OFSP farmers in Nigeria, to date and to the best of the knowledge of the researcher, studies on OFSP adoption, productivity, and household (HH) poverty status under a dissemination project in Nigeria are scarce, leading to the dearth of knowledge.

While the literature on agricultural interventions aimed at enhancing crop adoption and improving livelihoods among smallholder farmers in developing countries is growing, there exists a notable research gap regarding the specific impact and effectiveness of farmer-led jumpstarting projects focusing on orange-fleshed sweet potatoes (OFSP) in Nigeria. This research focuses on understanding how such projects influence the adoption rates of OFSP among smallholder farmers and the subsequent implications on agricultural productivity and poverty alleviation. First, there is a lack of empirical studies that specifically investigate the effectiveness of farmer-led jumpstarting projects in promoting the adoption of OFSP among smallholder farmers in Nigeria except for [6]. Existing literature [6,9,10,11,12,13,14] often generalizes findings from similar interventions without focusing on specific crop adoption dynamics. Second, little is known about the factors influencing smallholder farmers’ decisions to adopt OFSP through farmer-led jumpstarting projects. Understanding these factors is crucial for designing effective strategies that enhance adoption rates and sustainability of OFSP cultivation. Third, while there is anecdotal evidence [17,18,19] suggesting that OFSP cultivation can improve agricultural productivity and contribute to poverty alleviation, rigorous empirical studies are lacking to substantiate these claims within the context of farmer-led jumpstarting projects in Nigeria. Addressing the research gap on the role of farmer-led jumpstarting projects on OFSP adoption among smallholder farmers in Nigeria is essential for advancing agricultural development goals. This study aims to provide empirical evidence and insights that can guide policy formulation, program design, and resource allocation to enhance the impact of agricultural interventions on food security and poverty alleviation in Nigeria. This study benefitted from the theoretical framework proposed by [19] and, as such, enriched the previous study of [19].

The objectives of this study were to examine the sustainability outcomes of the Jumpstarting Project on the adoption of orange-fleshed sweet potatoes by smallholder farmers in Nigeria; examine the effects of the adoption of orange-fleshed sweet potatoes on yield and income of the farmers; and investigate the effects of adoption of the orange-fleshed sweet potato on the poverty status of the smallholder farmers in Nigeria.

The motivation for this research was centered on the fact that the essential goal of improving production and raising vitamin A intake in response to impending threats to global food security and malnourishment serves as the impetus for this investigation. The idea behind this study is that, particularly in the case of eliminating VAD in Nigeria, initiatives to support technology adoption and risk mitigation in crop production are effective ways to support farmers and increase the sustainability of regional food systems. The findings from this study will be useful for planning public investment in improved agricultural technologies in Nigeria and other developing countries. This study will also complement the existing studies that investigated the success of improved agricultural technologies offer suggestions for replication and future improvements and act as a springboard to undertake further detailed and comprehensive studies in Nigeria. Thus, stakeholders, research institutes, private agencies, nongovernmental organizations (NGOs), and international organizations will find the result of this study useful for policy-making and future breeding of biofortified crop varieties, especially orange-fleshed sweet potato (OFSP).

The rest of the paper is organized as follows: the literature review section followed by the methodology section, and then the results and discussion section. We present the conclusion and policy recommendations in other sections of the paper.

2. Literature Review

2.1. Components and Activities of Jumpstarting Project in Nigeria

White and yellow sweet potato types are the most widely grown in the country. In addition to providing other nutrients, including riboflavin, copper, pantothenic acid, and folic acid, their roots are a significant source of vitamins C, B6, K, and E [26]. Contrary to the OFSP varieties, these white and yellow varieties have little to no beta-carotene [27]. Several initiatives were implemented to increase the cultivation of OFSP in Ghana and Nigeria due to the limited adoption of OFSP types in these countries and the desire to encourage its consumption while also combating VAD. The Jumpstarting Project aimed to increase farmers’ and their families’ knowledge of and exposure to OFSP cultivars. The purpose of the project was to demonstrate several demand- and market-driven sustainable and inclusive strategies for boosting the production and consumption of OFSP, with the ultimate objective of raising household income and improving health [6]. The population most at risk for VAD, including homes with vulnerable women and children, was the focus of the initiative. These are households with children under the age of five and pregnant or breastfeeding mothers.

The Jumpstarting Project in Nigeria was implemented in the states of Osun and Kwara. For market linkage initiatives, it collaborated with the Home-Grown School Feeding and Health Program (O-Meals), the Partnership for Child Development, and Agricultural Development Programs (ADPs) [6]. These campaigns targeted farmers, consumers, and vine multipliers. The National Root Crops Research Institute (NRCRI) was in charge of managing the pre-basic seed supply, providing germplasm cleanup, and supporting seed and production systems. Before the start of the growing season (to promote vine sales) and throughout the harvest season (to promote root sales), the project used intense market sensitization activities carried out in neighborhood markets. These included plays, dances, displays of the OFSP’s historical roots, cooking demonstrations, the distribution of apparel bearing the OFSP logo (t-shirts, hats, scarves, umbrellas), and anecdotes and lectures by local leaders praising the nutritional benefits of the OFSP [6]. For interested members of the public to inspect, sample, and purchase OFSP roots and products, the audience was taught about the advantages of OFSP and where to find vines by promotional messaging. Together with collaborators from the selected West African countries, recipes were created, tested, and tailored to suit local tastes and preferences. In addition to these demand-generating market events, OFSP vines were provided to farmer groups as free samples along with thorough training on the fundamental agronomic procedures. Additionally, farmers received training on the value of employing high-quality sweet potato planting materials and connections to the project’s decentralized vine multipliers. The multipliers were placed specifically inside the surrounding areas to improve farmers’ access to vines [6]. In order to expand their market and income opportunities as well as give them a compelling reason to keep growing the crop, the farmers were also connected to a private sector buyer in the school feeding program in Nigeria [6]. These marketing campaigns specifically targeted Ghana and Nigeria because of the high prevalence of VAD [28]. Figure 1 illustrates the various linkages between the breeder, NRCRI, and participating HHs during the Jumpstarting Project. After the project implementation period expired, any remaining OFSP varieties multiplication and dissemination efforts were effectively turned over to independent farmer groups.

2.2. Conceptual Framework

The model for this study was designed to show the effect of breeders’ activities and socioeconomic characteristics of sweet potato farmers on the adoption of the orange-fleshed sweet potato and the respective outcomes of the adoption on farmers’ welfare and productivity. The conceptual framework is presented in Figure 2. The model is divided into four components.

The first component is the antecedent factors, which include the situation of the study area as it existed even before the introduction of OFSP. These factors include deficiency in vitamin A intake, expensive supplementation programs, death of children under the age of 5, death of lactating/pregnant women, low productivity of traditional white and yellow sweet potatoes, and farming household poverty. For example, many underage deaths resulting from a lack of vitamin A intake have been reported in different sub-Saharan African countries, including Nigeria. Low yield and income from the production of white and yellow sweet potatoes have limited income for the farmers, thus resulting in household poverty. These factors influenced the breeders and their partners, including the International Potato Center (CIP), Harvestplus, and the International Institute of Tropical Agriculture (IITA), to develop vitamin-A-fortified sweet potato varieties that could also increase farmers’ productivity and welfare. The breeders and their partners promoted the adoption of the orange-fleshed sweet potato through the program tagged “Jumpstarting Project”, which was subsequently implemented in West African countries, including Nigeria.

The second component is the independent variables. The socioeconomic characteristics of the farmers include age, gender, marital status, educational level, household size, farming experience, membership in the association, extension contact, having children under the age of 5, and being an indigene of the community. These factors are expected to influence farmer’s decision to adopt the orange-fleshed sweet potato through participation in the Jumpstarting Project. Gender, age household size, having children under the age of 5, and being an indigene of the community will influence participation in an intervention program like the Jumpstarting Project, which will subsequently influence the adoption of the OFSP.

The third component is the intervening variables. These variables, including government policy and managerial ability, might influence the dependent variables. For example, government policy in support of the implementation of the Jumpstarting Project could influence farmers’ decision to adopt the orange-fleshed sweet potato. The intervening variables (managerial ability) could also affect some of the outcome variables (yield and income). For example, farmers’ managerial ability in the use of resources could affect the expected yield of OFSP.

The fourth component is the outcomes of the adoption of OFSP. The expected outcomes of the adoption of OFSP include increased vitamin A intake, increased yield, increased income, and improved welfare (poverty status). Increased vitamin A intake is expected among adopting farmers and consumers. Smallholder farmers consume part of what they produce, thus increasing their vitamin A intake. Likewise, the consumers increase their vitamin A intake through purchasing OFSP. The adoption of OFSP is expected to lead to an increase in yield and income, which is expected to lead to improvement in the poverty status of the farmers who adopted the technology.

However, some of the independent variables (socioeconomic characteristics of the farmers) will most likely affect the outcome variables. Farmers’ socioeconomic characteristics such as age, gender, marital status, educational level, household size, farming experience, membership in associations, extension contact influence the use and combination of farm resources to produce agricultural crops such as OFSP, thus influencing its yield and income realized by the farmers.

2.3. Econometric Framework

The approach used in this study follows the approach of [19]. After adjusting for multiple external variables, the assumption was made that the correlation between the adoption and program participation serves as a reliable indicator of the influence of the Jumpstarting Project. First, the effect of the project is evaluated via the following model (1) estimated with ordinary least squares (OLS) following the approach of [19]:

ADOPT = αi + yTreated + δnon-treated + β_ix +u_i

(1)

where ADOPT, measured as the share (%) of sweet potato area allocated to OFSP, is regressed on the treated and non-treated (control group is the excluded category) and a vector of covariates (x) of HH characteristics. The variables of the HH characteristics are defined in

Table A1. Definitions of dependent and explanatory variables in regression models.

Variable	Description	Adopters		Non-Adopters		Pooled
		Mean	SD	Mean	SD	Mean	SD
Dependent Variables
Adoption	Adoption of OFSP = 1; 0 = otherwise	0.35	0.18	0.65	0.28	0.100	0.36
Adoption intensity	Area of land allocated to OFSP	2.1	1.1	--	--	2.1	1.1
Independent variables
Gender	Gender of HH head (1 = male, 0 = otherwise)	0.6696	0.384	0.6298	0.483	0.6438	0.309
Age	Age of HH head (years)	47.17	11.85	50.97	10.63	49.64	11.2
Marital status	Marital status measured as dummy (married = 1, otherwise = 0);	0.6250	0.453	0.6875	0.274	0.6656	0.293
Education level	1 = no formal education, 2 = primary education, 3 = secondary education, 4 = tertiary education	0.8929	0.473	0.8413	0.428	0.8594	0.573
Household size	No. of people in household (number)	7.72	2.32	7.52	2.58	7.59	2.49
Farming experience	No. of years in sweet potato production (years)	23.68	l2.19	23.51	8.66	23.57	10.02
Farm size	Area of land farmed (ha)	2.1	1.1	1.8	1.2	2.1	1.1
Risk aversion	HH head’s willingness to take risk (1 = willing to take risk; 0 = otherwise	0.100	0.56	0.3798	0.l97	0.70	0.56
Farmer association	Dummy, =1 if HH head is member of the local farmer association, 0 otherwise	0.8393	0.294	0.6154	0.243	69.39	0.328
Extension access	Dummy, =1 if HH head had access, 0 otherwise	0.6607	0.48	0.3702	0.237	0.4719	0.238
Number of years stayed in the community	HH head’s number of years stayed in community in years	29.95	11.24	29.54	12.07	29.68	11.77
Access to climate information	Dummy, =1 if the HH had access to climate information, 0 otherwise	0.5357	0.268	0.2260	0.189	0.3344	0.252
JP project participation	Participant (1 = yes, 0 = otherwise)	0.25	0.14	0.65	0.37	0.90	0.43
Indigene of community	Indigene of community (Yes = 1, 0 = otherwise)	0.8125	37.38	0.7115	28.20	0.7469	29.33
Visit to the agricultural field office	Visit to the agricultural field office (Yes = 1, 0 = otherwise)	0.5089	24.19	0.2163	12.75	0.3187	17.36
King J	Used King J OFSP vine (1 = yes, 0 = otherwise)	0.3214	13.83	--	--	0.3214	13.83
Mothers’ Delight	Used Mothers’ Delight OFSP vine (1 = yes, 0 = otherwise)	0.6786	12.94	--	--	0.6786	12.94
Children under five (5)	Children under five (5) (Yes = 1, 0 = otherwise)	0.7143	23.12	0.3269	17.38	0.4625	18.29

. The formulation and specification of variable ADOPT follow [19].

Model (2) is a variant of model (1) where ADOPT is regressed on HH location in project communities (PC) [19]:

ADOPT = α_ii + πPC + β_iix + u_ii

(2)

where PC = 1 for both treated and non-treated groups, and PC = 0 for the control group (the excluded category).

The specifications of the unrestricted models were re-estimated without including the vector of covariates (restricted). A comparison was made between these two specifications following the approach of [19], and the preferred model was chosen based on the results of an F-test. The terms α, β and u (error term) are parameters to be estimated. In model (1), y measures the effect of treated HHs and δ on non-treated neighbors. In model (2), π measures the effect of the Jumpstarting Project on all HHs from project communities.

In models (1) and (2), the dependent variable, which is the percentage of sweet potato area allocated to OFSP (ADOPT), ranges from 0 to 1. Historically, when attempting to estimate fractional outcome variables with boundary observations, conventional statistical models like the two-limit Tobit have encountered conceptual challenges and demonstrated biased estimates [29]. In this context, fractional regression methods have been suggested and employed as a viable approach to addressing bias arising from model misspecification [19,30]. These methods have gained support and acceptance in handling such issues. Various alternative methods for fractional regression (FR) estimation are available, including nonlinear least squares (NLS), quasi-maximum likelihood (QML), and beta-based maximum likelihood (ML) approaches. Alternative specifications of the functional form of the conditional mean, E(ay|x), include the cauchit, logit, probit, log log, and cloglog specifications [19,29]. When faced with these choices, it becomes crucial to choose a model that best fits the data being analyzed. One method to achieve this is by employing likelihood tests to assess the goodness of fit for the various alternative models. In this analysis, the QML estimation method was chosen, along with the probit specification, owing to several distinct advantages highlighted by [31].

As noted by [19], impact evaluation work faces another significant challenge when utilizing cross-sectional data, namely, biases arising from observable and unobservable variables. To address such biases, the ideal scenario would involve observing a group both in the treated and untreated states at a specific point in time. However, this is often not feasible. Hence, it becomes necessary to construct a counterfactual in order to attribute any changes in the indicator of interest to the intervention [32]. The conventional approach to creating a reliable counterfactual is for researchers to randomly assign individuals from the study population into treated and control groups. However, randomization can be challenging to implement in the social sciences, leading researchers to explore quasi-experimental study design alternatives instead [33,34,35]. Propensity score matching (PSM) is a widely used method to address biases arising from observable variables, whereas instrumental variables regression (IV) can effectively mitigate biases from both observable and unobservable factors [36].

PSM relies on observable characteristics of individuals in the sample to create a control group that closely resembles the treated group, differing only in intervention status [37]. Two conditions must be fulfilled for PSM estimation: (i) Observable characteristics must be independent of project outcomes, meaning there should be conditional independence. (ii) There needs to be an overlap in the distributions of observable characteristics between the treated and untreated groups, which is referred to as common support [38].

PSM initiates by conducting a regression of treatment status, where 1 represents the treated group and 0 represents the control group (i.e., non-treated and control), against a set of observable characteristics to compute propensity scores [39].

Typically, the initial step involves probit or logit estimation to compute the propensity scores [40]. These propensity scores are subsequently utilized for matching treated and control observations. Various matching methods are available, such as nearest neighbor (with or without replacement), kernel, caliper, and radius matching [41,42]. Subsequently, balance is assessed by comparing the means of the matching variables between the treated and control groups. The average treatment effect on the treated (ATET) is then computed based on the mean differences between these two matched groups [43].

The robustness of ATET estimates against biases from unobservable factors is investigated using the Rosenbaum Bounds Delta (Г) framework [42,44,45]. This framework relies on the presence of a confounding variable, denoted as W, which influences the likelihood of receiving treatment. Confidence intervals (bounds) are computed at various incremental values for the effect of W on the likelihood of being assigned to the treatment group. Thus, Г represents the odds that unobservable factors affecting treatment status significantly distort the predicted ATET [44]. Furthermore, this framework complements the IV approach described below as it relies on alternative assumptions to account for unobservable factors in the estimation process [44].

Four alternative PSM models are the basis for exploring spillover effects from the JP. These models incorporate the same covariates (x) to generate propensity scores, and the matching algorithm selected is the nearest neighbor criterion with replacement. In the first matching estimation PSM (1), the study compares treatment with the full control group (non-treatment and control). It is expected that spillover effects from the JP to non-treatment will result in a downward bias of impacts in this first PSM specification. PSM (2) matches project communities of HHs with the control group, which serves as a pure control, and we expect no spillover in this case. In PSM (3), treatment is matched with control, and this is the specification selected to determine the impact of the Jumpstarting Project on the treated. PSM (4) evaluates the magnitude of spillover by matching the treated with the non-treated.

Furthermore, as [19] noted, potential bias arises from the endogeneity of project participation, prompting the study to employ instrumental variable regression (IV) to mitigate biases from both observable and unobservable factors [19,46]. Procedural estimation with IV necessitates the use of a suitable instrument (z) that must satisfy two critical conditions: (1) it must be correlated with the regressor, i.e., participation; and (2) it must be independent of the error term [47]. An instrumental variable that has been utilized in this context is the intention-to-treat (ITT), which is adapted from the experimental medical literature [35,48,49]. In this case, ITT equals 1 for eligible members of the sample regardless of participation (treated) and 0 for non-eligible individuals. Additionally, ITT is linked to the households’ location in project communities, such that ITT = 1 if the households belong to project communities.

Instrumental variable regression is based on a two-step estimation approach [50]. In the first step, denoted in Equation (3), OLS is used to estimate participation as a function of ITT (PC) and x to get predicted participation (treated):

Treated = α_iii + _ρPC + β_iiix + u_iii

(3)

In the second step, a model is estimated where the predicted value of participation (treated) from the first step is used as a regressor. In this study, the second-step model (4) will be estimated using both OLS and FR, and the OLS version is denoted as follows:

ADOPT = α_iv + ꞇtreated + β_ivx + u_iv

(4)

Post-estimation results are used to determine the strength of the instrument, and in cases where a single instrument is used for estimation, the general rule of thumb is an F-test result greater than 10 [19,36,48].

3. Methodology

3.1. Study Area

3.1.1. Osun State

In Osun State, South West Nigeria, the Jumpstarting program was introduced. Hence, Osun State was selected as one of the study areas. Osun State is situated between longitudes 40′E and 50′E of the Greenwich Meridian and latitudes 70′N and 80′N of the equator in the South West of Nigeria. It is bordered in the south by Ogun State, in the north by Kwara State, in the west by Oyo State, and in the east by Ondo State. The State has a total population of 3,423,535 people and a land area of 10,245.00 km² [51]. Osun State is divided into thirty (30) Local Government Areas (LGAs). The state experiences two separate seasons: a wet season and a dry season. The dry season starts in November and lasts until March, whereas the rainy season lasts from April to October. The growth of several food and income crops is encouraged by this circumstance. The average annual temperature is around 27 °C, the average annual rainfall is between 2000 mm and 2200 mm, and the relative humidity is 79.90%. The white-, yellow-, and orange-fleshed sweet potato (OFSP) can be grown in optimum conditions thanks to the climate, vegetation, and edaphic elements. At both the commercial and subsistence levels, agriculture is practiced. One of the main food crops grown in the state is the sweet potato, which is produced in smaller quantities than in Oyo State in South West, Nigeria. The state also produces yam, maize, cassava, and other food crops.

3.1.2. Kwara State

In Kwara State, North Central Nigeria, the Jumpstarting program was also introduced. The coordinates for Kwara State are 8°30′N5000′E. It has a 36,825 sq km area with a temperature range of 26 to 32 °C with a high relative humidity. With an average yearly rainfall of 1200 mm, the vegetation of Kwara State is guinea woodland, which includes all kinds of flora and wildlife. Many people in the state of Kwara rely heavily on agriculture as their primary source of income. More than 75% of the people in Kwara State are employed and supported financially by the agricultural sector. Kwara State, which comprises l6 LGAs, has a population of 2,378,957 people according to the National Population Commission [51]. In Nigeria, Kwara State is mostly in charge of producing a variety of food crops, including cassava, sweet potato, yam, watermelon, maize, etc. Orange-fleshed sweet potato (OFSP) can be grown in both small and large amounts in Kwara State due to the climate and soil conditions. The choice of Kwara State as the research location where the Jumpstarting initiative was launched was made possible by the fact that the state is the highest grower of sweet potatoes in North Central, Nigeria.

3.2. Data Source, Sampling Techniques, and Sample Size

The research utilized cross-sectional data collected from smallholder sweet potato farming households in Nigeria, specifically in Osun and Kwara States. Respondents were selected from both intervention communities and non-intervention communities within the two states. Within the intervention communities, the study targeted both producers who participated in the project interventions (referred to as the “treated”) and producers who did not participate in the project interventions (referred to as the “non-treated”). The control group consisted of non-treated respondents residing outside the intervention communities.

Specifically, a community was designated as an intervention or project area if the project team and partners had conducted activities such as awareness creation, market sensitization, and farmer training, and had distributed planting materials within that community. On the other hand, communities where no project activities took place served as the control or non-intervention group. The selection of control communities involved a simple random sampling without replacement from a list of communities situated at least 5 km away from the nearest intervention community. Additionally, the selected control communities were confirmed to have not received any promotion or interventions related to OFSP from the project and its partners.

The primary objective of interviewing the non-treated respondents was to capture any potential spillover effects of the project in the intervention communities. On the other hand, the control group interviews aimed to assess and understand the situation in the non-intervention communities. Sampling proceeded as follows: Multistage sampling techniques were used to select respondents for the study. In the first stage, the purposive sampling technique was used to select four intervention communities where the Jumpstarting Project was implemented and two non-intervention communities. In total, six communities were selected in both Osun and Kwara States. In the second stage, the intervention communities were stratified into participants and non-participants: 20 participants and 30 non-participants were selected using the population proportion size from the list of sweet potato farmers in each of the four intervention communities making a total of 80 participants and 120 non-participants, respectively. In the third stage, 60 sweet potato farmers were randomly selected from each of the two (2) non-intervention communities, giving a total of 120 control groups. Thus, a total of 320 respondents were selected for the study. Primary data were collected using a pretested structured questionnaire through interviews.

Moreover, prior to conducting the interviews, selected respondents in the study sites were requested to provide their consent voluntarily. The interviews commenced only after obtaining consent from the respondents. Trained enumerators conducted personal interviews to collect the data. The information gathered encompassed various socioeconomic characteristics of the respondents, such as age, gender, educational qualification, farming experience, farm size, household size, and more. Additionally, data were collected on sweet potato (and OFSP) production and marketing activities, production costs, household livestock ownership and control, access to savings and credit facilities, availability of extension services, income-earning activities, demographic details, sources of weather information, and agronomic and pest management practices.

Adoption of OFSP varieties, in this study, is defined as the growth of OFSP varieties during the previous production period (2020/2021 season). Two OFSP varieties were considered in this study—namely, Mother’s Delight and King J—in the two states due to their popularity among smallholder farmers. Farmers were first asked whether they knew OFSP varieties and, if affirmative, asked whether they had planted the OFSP varieties during the previous production period. The interviews were conducted using the local language.

3.3. Data Analysis

3.3.1. Sequential Probit (Determinants of Awareness, Registration of Farmers Groups, and Collection of OFSP Vine)

Participation in the Jumpstarting Project involves a sequence of steps and choices. In this setup, this process involves becoming aware of the Jumpstarting Project, registering for the Jumpstarting Project through the farmers’ association, and collecting orange-fleshed sweet potato vines from implementers. Registration is a strict subset of awareness, and the collection of orange-fleshed sweet potato vines is a strict subset of registration. The sequential probit model listed below was used to model the above sequential process:

Pr(I_k,i = 1/X_i, Sⁱ, V_i, Z_i, I_k-l,i = 1) = p_ik

(5)

where p_ik is the standard probit model represented by pik = Φ (Xi, Si, Vi, Zi, β). Herein, Φ represents the cumulative distribution function and β captures vectors of parameters to be estimated. Ik is an indicator function that takes a value of one if farmer i passes transition k (becoming aware of the Jumpstarting Project, registering for the Jumpstarting Project, and collecting orange-fleshed sweet potato vines from implementers) and zero otherwise. Xi captures a vector of household i’s characteristics such as household size, age, education, etc. that affect participation in the Jumpstarting Project. Si captures participation in social network activities such as membership in cooperatives, etc. [52], while Vi captures state-level fixed effects to control state-level heterogeneity in the implementation of the Jumpstarting Project. The variable Zi is the instrument: the number of years the farmer has resided in the village. According to implementers, all registered association of farmers will collect orange-fleshed sweet potato vines. As such, the selection bias occurs at the awareness and registration stage. It was assumed that those farmers who have lived in the village for long time are more likely to be aware of the Jumpstarting Project as they have more connections. Similarly, they are also more likely to be recognized by village leaders and, hence, can be considered as genuine sweet potato farmers at the time of registration, which increases the likelihood of participating in the Jumpstarting Project.

3.3.2. Endogenous Treatment Effect (ETE)

The results of direct comparisons between adopting and non-adopting households may be misleading since these groups may consistently differ in terms of both observable and unobserved variables [53]. Due to a sample size that does not fairly represent the target population, the measure of association between the treatment (in this example, OFSP adoption) and the result (here, the income of OFSP production) may be skewed. The traditional statistical approach to address this bias uses instrumental variable regression, which enables quantification of the impact of technology adoption on outcome variables of interest while removing the impact of reverse causation or simultaneity [53,54,55,56,57,58,59,60,61,62,63,64]. However, interaction variables between the endogenous adoption variable and other covariates (such as education) need to be included in the model when the influence of technology is not uniform among sample households. With more interaction terms, more instrumental variables are needed to simply identify the model. In general, it is challenging to even identify one viable instrumental variable for a model, and frequently, the instrumentation choice is questioned since it does not adhere to the exclusion restriction. When one attempts to quantify the heterogeneous consequences of an endogenous variable, the size of this issue increases significantly. Using an endogenous treatment effects (ETE) approach, in which adoption is considered as a regime changer and a single instrumental variable (selection instrument) is sufficient to capture the heterogeneous effects of the technology, is a more elegant and straightforward way to handle self-selection bias. This method also provides a more effective solution to the problem. The ETE framework makes the implicit assumption that the error terms are both independent and uniformly distributed. This signifies that the outcome and treatment status of each respondent are not related to the outcome and treatment status of any other persons in the population. Consequently, although these models are less suited for modeling correlated data resulting from hierarchical or longitudinal study designs, they are great for estimating impacts from cross-sectional datasets, such as the one we used in our research. The primary advantages of an endogenous switching model are that it makes it possible to model both the allocation of households to various treatments as well as the effects of treatment on other outcomes, all while estimating the degree to which common, unmeasured variables affect the outcome and the explanatory variables. In other words, endogenous switching models allow for modeling both the allocation of households to various treatments as well as the effects of treatment on other outcomes. This method takes into account the possibility of a selection bias and simulates the outcomes that would have occurred if non-adopters had been included in the adoptive group. The ETE model requires two steps of estimation in order to be accurate. The first step is a selection equation that is based on a binary choice function. In this stage, it is hypothesized that a variety of farm-household characteristics would decide whether a certain household will adopt a certain technology (Ai) following [53]:

A_i* = z_iα + ȵ_i where A_i = {¹₀ if A_i* > 0 otherwise

(6)

The benefits that are anticipated to be gained from the implementation of technology are represented by the realization Ai of the dichotomous latent variable Ai*; zi represent the observed farm-household variables that have an effect on the adoption variable, α stands for the parameter vector that needs to be estimated, and ȵi denotes the unobserved heterogeneity. The outcome of interest is modeled based on the observed adoption realization Ai in the second stage of the ETE estimation process. The ETE estimation is broken down into its component parts by [65]. In order to explain the outcome of interest, depending on the selection function, two regime equations are specified as follows:

Regime 1: Y_1i = x_1iψ₁ + Ɛ_1i if A_i = 1

(7a)

Regime 2: Y_2i = x_2iψ₂ + Ɛ_2i if A_i = 0

(7b)

where xi are characteristics of the farm households that have an effect on the outcome variable (Yi). The parameter vectors ψ1 and ψ2 are to be estimated, and the error terms for regimes 1 and 2 are denoted by the Ɛ1 and Ɛ2, respectively. The selection equation needs to include at least one variable (i.e., a selection instrument) that is connected to adoption but not correlated directly with the outcome (exclusion limitations) [53]. This is necessary in order to ensure that the model is identified with an adequate degree of precision. The selection instruments for this study were variables that connected to the characteristics of social networks (the usage of NM OFSP). In the ETE framework, selection instruments are necessary in order to identify the model, just like they are in the two-stage least squares or control function approach. The authors of [53] proposed a straightforward falsification test as a means of determining whether these instruments should be admitted into evidence. A valid selection instrument will have an effect on adoption, but it will not have any effect on the outcome among those who do not adopt.

The ETE model was applied in the research in order to analyze the expected outcomes in the counterfactual hypothetical instances where adopter household members did not adopt the technology and to compare the expected outcomes in terms of income of OFSP adopters and non-adopters. The study has determined that adoption typically has the following average effects on a population of agricultural households: E(Y1|Ai = 1 − Y2|Ai = 0) where E represents mathematical expectation operator. This is indicated by the average treatment effects (ATE) parameter, which, according to the impact evaluation literature, stands for the “supply of the technology” effect or, in our case, the impact of providing farmers with the OFSP technology. The average treatment effect on the treated (ATT), which limits the computation of the average treatment effect to the subpopulation of adopters, is the effect parameter that determines the causal effect of adoption in the presence of non-compliance, that is, E(Y₁ − Y₂)|A_i = 1.

3.3.3. Foster, Greer, and Thorbecke Poverty Index

To evaluate the poverty status of the smallholder farmers, Foster, Greer, and Thorbecke poverty index was employed. The analysis of productivity and poverty has traditionally relied heavily on asset-based measures of poverty, which have been used in several studies in an effort to address the many limitations of the Multidimensional Poverty Index (MPI) applied to poor smallholder HHs, including issues with data availability and reliability [66,67,68,69,70,71,72]. Experts have stated that from the capacities framework, many of the systematic and persistent depravations encountered by smallholders across the various dimensions of human welfare are frequently ignored by commonly used measurements of poverty [73,74,75,76,77,78].

Although the headcount ratios and poverty gap, which is consumption measures of poverty classified under the Foster–Greer-Thorbecke (FGT) are frequently used to assess the impact of agricultural development projects on poverty due to the availability and dependability of data across the majority of sub-Saharan African countries, including Nigeria. In refs. [21,45,79,80,81,82,83], expenditures on non-food items, overall expenditures, and indicators connected to poverty are all calculated on a per capita basis. The study employed the headcount poverty ratio as an extra welfare metric in addition to expenditure data. Food and other expenditures amounts are added to determine the total expenditure. The cost of purchasing food and the assumed value of consumption from one’s own produce together make up a household’s food consumption expenditure. The final welfare-related outcome indicator, also known as the headcount ratio, calculates the percentage of households that fall below the poverty line. The authors of [84,85,86,87,88,89] suggested using per capita total expenditures to assess the poverty status of households. The headcount ratio (P0) is formally determined as follows:

$$P\alpha ={\displaystyle \frac{1}{n}}{\displaystyle \sum _{i=1}^q(}{\displaystyle \frac{z-yi}{z}})\alpha$$

(8)

where n represents the total number of farmers in the group; q represents the number of poor farmers; Z represents the poverty line created; Yi represents the value of per capita expenditure of the ith farmer; and α represents the poverty aversion parameter. The following was the classification of the poverty level: (a) Non-poor: The respondents in this group had per capita expenditures that are higher than the poverty threshold. That is p > 2/3 of the mean per capita expenditure per month. (b) Poor: The respondents in this group had per capita consumption expenditures that are lower than the poverty threshold. That is p < 2/3 of the mean per capita expenditure per month.

3.3.4. Endogenous Switching Probit Model (ESPM)

This study is also interested in estimating the effect of orange-fleshed sweet potato (OFSP) adoption on binary outcomes like the incidence of poverty. Contrarily, it might be challenging to apply nonlinear models for continuous outcome variables when sample selection and endogenous switching are present for binary outcomes [90,91,92]. Therefore, two-stage approaches used in evaluations (like Heckman’s sample selection model) would produce contradictory results and erroneous inferences. Therefore, the endogenous switching probit (ESPM) framework was used in the present study, which is similar to the endogenous switching regression for continuous outcomes [92,93]. The ESPM model was used to demonstrate the impact of adopting OFSP on the poverty status of farm households. This was accomplished in two separate analytical steps. The choice to adopt OFSP was calculated in the first step using a probit model. In the second stage, the study used a probit regression with selectivity correction to investigate the relationship between a set of explanatory variables conditional on the farm households’ decision to adopt and a binary outcome variable (poverty headcount). The decisions by farm households to adopt OFSP were represented by the following latent response models:

$$T_i^{*}=M_i\gamma +{\tau }_i$$

(9)

$$T_i=\left\{\begin{array}{c}1, if T_i^{*}>0\\ 0, otherwise\end{array}\right.$$

(10)

where $T_i^{*}$ is a continuous latent variable; γ represents parameters to be estimated; ${\tau }_i$ is an error term. The binary response was defined as follows:

$$g_i^{*}=w_i\mathsf{\lambda }+{\mu T}_i+{\tau }_i$$

(11)

$$g_i=\left\{\begin{array}{c}1, if g_i^{*}>0\\ 0, otherwise\end{array}\right.$$

(12)

where $g_i$ represents the most important outcome variables; $g_i^{*}$ is a continuous latent variable, λ represents a vector of parameter estimated; μ represents the coefficient of endogenous treatment dummy; τ_i represents the residual term. The endogenous switching issue here is that the response g_i, for the i^th farm household is not always observed. In addition, $g_i$, is assumed to depend on endogenous dummy T_i, including a vector of the explanatory variable m_i. The endogenous dummy T_i also depends on the vector of the explanatory variable m_i, and there is a possibility that vector w_i and M_i, are correlated. Due to unobserved endogeneity, the causative impact would lead to erroneous estimates in direct estimation of Equation (11) and interpretation. By simultaneously estimating the selection and outcome equation with the proper instrumentation of the OFSP adoption decision, the ESPM regression would correct this bias [93] (lokshin and Sajaia, 2011). In a two-stage treatment framework, the ESPM framework models the choice to adopt OFSP and its effect on multiple binary outcomes. A probit model was used in the first step to model and estimate farm households’ decision to adopt OFSP. In the second stage, a probit model with selectivity correction was used to ascertain the association between the binary outcomes, OFSP adoption, and explanatory variables. The study stated the binary outcomes contingent on the adoption of OFSP as an endogenous switching regime model, following [93]:

$$Regime 1\left(adopters\right): g_{1i}^{*}={\lambda }_1{W}_{1i}+{\varphi }_{1i} g_{1i}=1(g_{1i}^{*}>0)$$

(13)

$$Regime 1\left(Non-adopters\right): g_{0i}^{*}={\lambda }_0{W}_{0i}+{\varphi }_{1i} g_{0i}=1\left(g_{0i}^{*}>0\right)$$

(14)

Observed g_i reflects the latent variables’ dichotomous realization and is described as follows:

$$g_i=\left\{\begin{array}{c}{g}_{1i}, if T_i=1\\ g_{0i}, if T_i=0\end{array}\right.$$

(15)

where g_1i and g_0i, represent latent variables determining observed binary outcomes, g₁ and g₀, for OFSP adopters and non-adopters, respectively; W₁ and W₀ represent vectors of the weakly exogenous variable; M_i represents the vector of variables that determine a switch between regimes; λ₁ and λ₀ represent vectors of parameters estimated; ϕ_1i and ϕ_0i represent the error term of outcome equations. In accordance with [93], the study evaluated an endogenous switching full information maximum likelihood (FIML) probit model to calculate the relevant parameters. By utilizing the analytical framework suggested by [93], the study also calculated the effects of OFSP adoption on the farmers’ poverty status. Furthermore, according to [94], the defined endogenous switching probit model permitted the derivation of probabilities in hypothetical situations. We calculated the average treatment effect on the treated (ATET) and the average treatment effect on the untreated (ATUT) using the following formulas:

$${TET}_j=pr\left(g_{1j}=1|T=1\right)-pr\left(g_{0j}=1|\mathrm{T}=1\right)$$

(16)

$${TUT}_j=p\tau \left(g_{1j}=1|T=0\right)-pr\left(g_{0j}=1|\mathrm{T}=0\right)$$

(17)

3.3.5. Robustness Check (Propensity Score Matching)

The model assumption of the ETE and ESPM estimation, i.e., the choice of instrumental variables, may have an impact on the outcomes. The PSM technique was judged essential to apply in order to confirm the reliability of the estimated treatment effect data derived from the ETE and ESPM. The authors of [42] claim that PSM provides an estimate of the impact of an adoption-related “treatment” variable on an outcome variable that is largely devoid of bias resulting from a relationship between the adoption-related treatment status and observable factors. Matching techniques, however, are not robust to “hidden bias” brought on by unobserved factors that have an impact on both the outcome variable and assignment to treatment at the same time. The PSM is based on the conditional independence assumption (CIA) (see [95]), which states that outcomes are independent of the treatment written as TG1, G0 T|M, conditional on observables characteristic of the rural farm households (M). This assumption controls for differences in observable covariates that might influence a rural farm household’s decision to adopt the orange-fleshed sweet potato. The shared support or overlap requirement is a further supposition: 0 < (T = 1|M) < 1.

This requirement guarantees that, in the propensity score distribution, the comparison observations are “nearby” the treatment observations [93]. According to [96], it explicitly ensures that people with the same observable qualities have a favorable possibility of being in both categories. In order for the estimation to be performed on farmers with common support, we carried out this prerequisite. As a result, according to [97], the average treatment effect on the treated (ATET) is the difference between the mean outcomes of matched adopters and non-adopters with shared support. According to [98], treatment observations with limited common support should be abandoned. We can only deduce causality intuitively in the context of common support. A vital step is to perform a balancing test. That is, to ascertain if:

$$\widehat{P}\left(M|\mathrm{T}=1\right)=\widehat{P}\left(M|\mathrm{T}=0\right)$$

(18)

The likelihood that a rural household will adopt OFSP given M is expressed by the propensity score (P(m)), which is represented as follows:

$$P\left(m\right)=\mathrm{Pr}\left(T=1|\mathrm{M}=\mathrm{m}\right)$$

(19)

where T = adoption of orange-fleshed sweet potato; M = observable characteristics of farm households.

Calculating the average treatment effect (ATE) requires more information than simply predicting propensity. According to each adopter’s propensity score, it is critical to look for the proper counterfactual or counterfactuals that match them. Therefore, selecting a matching algorithm is the next step. The nearest neighbor method and kernel matching are the most popular matching techniques. According to the closest propensity scores, adopters and non-adopters are matched using the nearest neighbor matching method [99]. The counterfactual for the adopter units was created using these matched non-adopter units. By deducting a weighted average of the outcomes in the comparison group from each result observation in the treatment group, the kernel-based matching method calculates the effects of the treatment. Each non-adopter unit is given a weight according to how far away it is from the adopter unit. According to [94], the three matching estimators of ATET can be described as follows using their framework:

$$ATET={\displaystyle \frac{1}{q^1}}\sum 1\left\{\left(G_{1i}|T_i=1\right)-\sum {jc}_{\mathrm{1,0}}\left(G_{0i}|T_i=0\right)\right\}$$

(20)

where c stands for a set of scaled weights that calculate the distance between each non-adopter and the target adopters, and q¹ is the number of adopter cases. According to [94], these estimators differ mostly in the number of matches selected for each target instance that needs to be matched as well as how these multiple matches are weighted, _l,0, if more than one is employed. Then, by averaging the changes in income and poverty across rural households that used the OFSP and those that did not (see, for example [100]) the treatment effect on the treated (ATET) is calculated as follows:

$$\begin{array}{c}E\left(G_1-G_0|\mathrm{T}=1\right)=\mathrm{E}\left[E\left(G_1-G_0|\mathrm{T}=1,\mathrm{P}\left(\mathrm{m}\right))\right)\right]\\ =E[E\left(G_1|\mathrm{T}=1,\mathrm{P}\left(\mathrm{m}\right)\right)=\mathrm{E}\left[\left(G_0|\mathrm{T}=0,\mathrm{P}(\mathrm{m}\right)\right]\end{array}$$

(21)

The distributions of all baseline variables are equal between the farm households who adopted improved technology (the treated group) and the non-adopters (the control group) according to the research of [98]. Since “blocks” are groups of participants with similar propensities, this “balancing property” indicates that if we successfully adjust for the propensity score while comparing the groups, we have succeeded in transforming the observational data into a sort of randomized block experiment.

The variables of interest, the description of variables, and the socioeconomic characteristics are presented in Table A1 (Appendix A). About 35% of the respondents adopted OFSP, with 54.38% being male farmers, indicating a larger proportion of male farmers being involved in sweet potato production. The mean age was 49.64 years, indicating that the sweet potato farmers were in their active and productive age, the majority (66.56%) were married with 7.59 persons per household, which might be attributed to the fact that many of the households were married, hence medium to large household size. The majority (85.94%) of the sweet potato farmers were literate having one form of education, including primary, secondary, or tertiary education, which is expected to positively influence the sustainability of OFSP adoption. The mean years of farming experience were 23.57 years, indicating that the sweet potato farmers were in the production of sweet potatoes for a longer period. The average farm sizes cultivated by the farmers were 2.1 ha, which is an indication that the sweet potato farmers were predominantly smallholder farmers. All of the adopters of OFSP were willing to take risks on newly introduced technology, which might have informed their decision to cultivate OFSP while only 37.98% of the non-adopters were willing to take risks on new technology. The majority (83.93%) of the adopters were members of farmers’ associations, hence they might have the opportunity to enjoy group dynamics and farm inputs. The majority (66.07%) of the adopters were also found to have access to extension services when compared with the non-adopters (37.02%). The adopters might have been pre-informed about the technology and Jumpstarting Project even before its implementation. The majority (7l.43%) of the adopters had children under five years old when compared with non-adopters (32.69%). Having children under the age of five might have also influenced adopters’ decision to cultivate OFSP.

4. Results and Discussion

4.1. Determinants of Awareness, Farmer Group Registration, and Collection of OFSP Vines

The results of the sequential probit comprising the Jumpstarting Project awareness, farmer group registration, and collection of OFSP vines are presented in

Table 1. Parameter estimates of the sequential probit model.

	Jumpstarting Awareness	Jumpstarting Registration	Collection of OFSP Vine
Variable	Coefficient	Coefficient	Coefficient
Age	0.0325 *** (0.0106)	0.0327 ** (0.0164)	0.3789 *** (0.1002)
Marital status	−0.0114 (0.0203)	0.3274 *** (0.0801)	0.0774 ** (0.0312)
Education	0.1137 *** (0.0206)	0.0393 ** (0.0163)	0.0250 (0.0238)
Gender	0.3204 *** (0.1028)	0.0849 * (0.0448)	0.4446 *** (0.1287)
Household size	−0.0250 (0.0238)	0.0124 (0.0184)	0.0351 (0.0243)
Farming experience	−0.0001 (0.0006)	−0.0006 (0.0008)	−0.0003 (0.0007)
Membership in association	0.0026 *** (0.0006)	0.0038 *** (0.0013)	−0.0009 (0.0017)
Access to credit	−0.0003 (0.0003)	0.0003 (0.0004)	−0.0003 (0.0007)
Access to off-farm work	−0.0002 (0.0003)	−0.0008 (0.0005)	0.0006 (0.0005)
Risk aversion	0.0031 ** (0.0017)	0.0003 (0.0004)	−0.0001 (0.0006)
Access to extension contacts	0.0031 ** (0.0017)	0.0006 (0.0005)	−0.0006 (0.0008)
Access to climate information	−0.0008 (0.0005)	−0.0009 (0.0017)	−0.0003 (0.0015)
Number of years in the community	0.0026 ** (0.0013)	0.0002 (0.0005)	−0.0008 (0.0005)
Primary occupation	0.0001 * (0.0001)	−0.0002 (0.0011)	0.0005 (0.0004)
Visiting the agricultural field office	0.0002 ** (0.0001)	0.0023 ** (0.0009)	0.0001 (0.0006)
State fixed effect	Yes	Yes	Yes
Wald chi2(27)	365.3	274.5	235.7
Pseudo R2	0.47	0.63	0.51
Goodness of fit measure	1.62	1.47	1.35

Figures in parentheses are standard error; ***, **, * represent 1%, 5%, and 10% probability levels, respectively.

. The instrumental variable (IV) (number of years in the community) has been found to affect the probability of being aware of the Jumpstarting Project but not the conditional probability of registration and collection of OFSP vines. This implies that selection bias for the program has been remedied by the IV at the level of awareness. The goodness of fit results of 1.62, 1.47, and 1.35 for awareness, registration, and collection of vines indicate a good choice of selected explanatory variables.

With respect to the determinants of sweet potato farmers’ participation at the different stages of the Jumpstarting Project, it was found that age, education, gender, membership in associations, risk aversion, access to extension contacts, number of years in the community, primary occupation, and visits to the agricultural field office influenced farmers awareness of jumpstarting projects. For sweet potato farmers’ registration, it was found that age, marital status, education, gender, membership in associations, and visits to the agricultural field office affected farmers’ registration in jumpstarting projects. It was equally found that age, marital status, and gender influenced the collection of OFSP vines.

Age was found to be positively and significantly associated with awareness, registration, and collection of OFSP vines. This implies that the age of the respondents increases the likelihood of being aware of the Jumpstarting Project, with registering for the program conditional on awareness and collecting OFSP vines through the intervention project conditional on registration. Older farmers are more likely to be connected; hence, they are more likely to be aware of incoming programs in their respective communities. Also, at the registration stage, older farmers might be given credence for registering for new intervention programs, which might be out of respect for their experience and for elders in the community. Older farmers might also be given credence at the stage of collection of OFSP vines conditional on their registration for a jumpstarting project.

Marital status significantly and positively influenced registration for the jumpstarting project conditional on awareness and collection of OFSP vines, which, in turn, was conditional on the registration. Married respondents are more likely to register for intervention projects such as the Jumpstarting Project with the hope of improving their livelihood. They are more likely to have a medium to large household size, hence their decision to register and collect OFSP vines.

Education positively and significantly influenced awareness and registration, which was conditional on the awareness. This implies that education increases the likelihood of awareness and registration for the project. Farmers who are educated tend to search for information that will help them increase their productivity, hence, getting aware of intervention projects such as jumpstarting projects. Farmers who are educated are also more likely to register for jumpstarting projects based on the fact that they are aware of the existence of the project.

Gender positively and significantly influenced awareness, registration, which was conditional on awareness, and the collection of vines, which was conditional on the registration. This indicates that male farmers are more likely to be aware of a jumpstarting project, register for the project, and collect OFSP vines than female farmers. This might not be connected to the fact that the male gender has more access to farm resources, including the implementation of intervention projects such as the Jumpstarting Project. Hence, the male gender was more aware and registered for the program, and thus, the collection of OFSP, which was conditional on the registration, tends to favor the male gender.

As expected, membership in the association positively and significantly influenced farmers’ awareness, registration, which was conditional on the awareness, and collection of vines, which was conditional on the registration. This is an indication that membership in the association increases the likelihood of being aware of a jumpstarting project, registering for the project, and collecting OFSP vines. This might be due to the fact that farmers who are members of the association tend to enjoy group dynamics and access to information. Thus, farmers might have received information about the implementation of a jumpstarting project in their community through their various associations.

Risk aversion and access to extension contacts positively and significantly influenced farmers’ awareness of jumpstarting projects. This implies that farmers being risk averse and having access to extension services increases the probability of their being aware of the jumpstarting project in their community. Farmers who are always willing to take risks with new technology are more likely to be aware of a new intervention project being implemented in their community. Also, farmers who had access to extension agents are more likely to have access to up-to-date information, including having access to information about the implementation of jumpstarting projects in their community.

The number of years of stay in the community, primary occupation, and visits to the agricultural field office positively and significantly influenced farmers’ awareness of the implementation of the jumpstarting project. This indicates that the number of years in the community, primary occupation, and visits to the agricultural field office increase the likelihood of being aware of the implementation of a jumpstarting project in their community. Farmers who have stayed in the community for a long time are more likely to have socioeconomic and political power in their community; hence, when new programs are being introduced, they receive firsthand information about the implementation of the project. Also, when new technologies are being introduced, farmers who are predominant users of the technology are first targeted; hence, farmers whose primary occupation is sweet potato production are more likely to be aware of the implementation of a jumpstarting project that helps to promote the adoption of the orange-fleshed sweet potato. In addition, farmers who have visited the agricultural field office are more likely to be aware of the implementation of jumpstarting projects in their community. They might have received information from their field officers, thus being aware of the JP. Visits to the agricultural field office also influenced registration for the jumpstarting project, which was conditional on awareness. Farmers who have visited the agricultural extension office might have been informed of the need to register for the project, including having access to OFSP vines and other resources through the jumpstarting project.

4.2. Sustainability of the Jumpstarting Project (JP) on the Adoption of the Orange-Fleshed Sweet Potato (OFSP)

Presented in

Table 2. Regression results for the adoption of the orange-fleshed sweet potato.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Variable	(i) OlS R	(i) OlS U	(ii) OlS R	(ii) OlS U	(i) FR R+	(i) FR U+	(ii) FR R+	(ii) FR U+
Treated	0.1662 *** (0.0138)	0.0055 ** (0.0024)			0.0049 *** (0.0018)	0.0140 * (0.0008)
Non-treated	0.0061 (0.4718)	0.0027 (0.0033)			0.0041 * (0.0022)	0.0807 *** (0.0144)
PC			0.0063 *** (0.0023)	0.0098 *** (0.0021)			0.0063 ** (0.0029)	0.0026 * (0.0014)
Age		0.0063 ** (0.0029)		−0.0002 (0.0034)		−0.0005 (0.0008)		0.0048 (0.0045)
Marital status		−0.0001 (0.0021)		0.0112 ** (0.0044)		0.0001 (0.0003)		0.0005 (0.0016)
Education		0.0026 * (0.0014)		0.0145 ** (0.0075)		0.0003 (0.0005)		−0.0014 (0.0024)
Gender		−0.0022 (0.0024)		0.0154 *** (0.0038)		0.0007 (0.0007)		−0.0047 (0.0035)
Household size		0.0032 *** (0.0009)		−0.0019 (0.0014)		0.0007 *** (0.0002)		0.2998 *** (0.0922)
Farming experience		−0.0018 (0.0012)		−0.0001 (0.0005)		0.0003 ** (0.0001)		0.0656 (0.1087)
Farm size		0.0078 (0.0110)		0.0006 (0.0007)		−0.0001 (0.0002)		−0.1234 (0.0968)
Membership in association		0.0002 ** (0.0001)		0.0015 (0.0011)		0.0006 (0.0013)		0.159 (0.1282)
Access to credit		0.0056 (0.0037)		0.0010 *** (0.0002)		0.0005 * (0.0003)		0.0334 (0.3497)
Risk aversion		0.0050 (0.0032)		−0.0017 (0.0014)		−0.0007 (0.0005)		−0.2137 (0.1750)
Access to extension contacts		0.0017 (0.0024)		−0.0001 (0.0005)		0.0015 *** (0.0006)		−0.0491 (0.1384)
Access to climate information		0.0003 (0.0016)		0.0004 (0.0007)		0.0032 (0.0022)		0.4067* (0.2466)
Years stayed in the community		−0.0001 (0.0021)		0.0017 * (0.0010)		−0.0016 (0.0066)		−0.0491 (0.1384)
Project community l_Osun		0.0021 ** (0.0008)		−0.0004 (0.0004)		0.0007 *** (0.0002)		0.2635 ** (0.1131)
Project community 2_Osun		0.0127* (0.0070)		0.0009 *** (0.0002)		0.0019 *** (0.0006)		0.3558 *** (0.1362)
Project community l_Kwara		0.0006 ** (0.003)		0.0002 (0.0006)		0.0036 ** (0.0018)		−0.2137 (0.1750)
Project community 2_Kwara		0.0049 ** (0.0023)		0.0012 *** (0.0002)		0.0035 * (0.0021)		−0.0586 (0.221)
Constant		−0.0002 (0.0034)		0.0004 ** (0.0002)		0.0018 ** (0.0007)		−0.0147 (0.0145)
R2	0.637	0.592	0.637	0.592
F	5.737	12.63	6.519	14.32

Figures in parentheses are standard error; ***, **, * represent 1%, 5%, and 10% probability levels, respectively; ⁺ Marginal effects; ^R restricted, ^U unrestricted.

is the result of the sustainability outcomes of the Jumpstarting Project on the adoption of the orange-fleshed sweet potato from the econometric framework model, which comprises models (i) and (ii). With respect to the alternative specifications, eight models were estimated, including unrestricted (2, 4, 6, 8) and restricted (1, 3, 5, 7) models. OLS was used in estimating model (i) (1, 2, 5, 6) while fractional regression was used to estimate model (ii) (3, 4, 7, 8). The results were found to be consistent across specifications indicating a significant and positive sustainability of the Jumpstarting Project on the adoption of the orange-fleshed sweet potato by the treated. Various criteria were used in the selection of the preferred model. Unrestricted versions of the model were preferred to the restricted versions based on the result of the F-statistics. Thus, model (i) is selected because more information was included in the model while estimating the effects for the treated and non-treated rather than pooling them together into a single project community group. Last, the outcome indicator “ADOPT” was captured in fractional form (%) following [19]; fractional regression was preferred to OLS. Thus, column (6), which is an unrestricted version of the model (i), was selected for onward discussion of findings presented in Table 2.

The results show that the adoption of the orange-fleshed sweet potato for the treated is 14% under (6), which is statistically significant at a 10% level of probability. In addition, the parameter estimates of the adoption rates based on intervention communities indicate significant differences. For instance, project community 1_Osun and project community 2_Osun indicate that the adoption of OFSP was sustained and increased by 0.7% and +1.9%, respectively, in the first and second intervention communities in Osun State. Furthermore, project community 1_Kwara and project community 2_Kwara show that the adoption rate of OFSP was sustained and increased by 3.6% and 3.5%, respectively, in the two intervention communities in Kwara State. This implies that the implementation of the Jumpstarting Project ensures the sustainability and spread and increases the adoption of the orange-fleshed sweet potato, especially in the intervention communities. Household size, farming experience, access to credit, and access to extension services positively and significantly influenced the adoption of OFSP at 1%, 5%, 10%, and 1%, respectively. This implies that household size, farming experience, access to credit, and access to extension services increase the adoption of OFSP by 0.7%, 0.3%, 0.5%, and 1.5%, respectively.

The parameter estimates of the four PSM models are presented in

Table 3. Estimates of the propensity score matching.

	PSM(1)	PSM(2)	PSM(3)	PSM(4)
	Treated/Non-Treated and Control	PC/Control	Treated/Control	Treated/Non-Treated
Age	0.0001 (0.0006)	0.0018 ** (0.0007)	0.0001 (0.0002)	−0.0001 (0.0021)
Marital status	0.0005 (0.0003)	−0.0006 (0.0008)	0.0002 (0.0002)	−0.0022 (0.0024)
Education	0.0003 (0.0003)	−0.0001 (0.0006)	0.0004 (0.0006)	0.0026 * (0.0014)
Gender	−0.0002 (0.0003)	0.0003 (0.0004)	0.0001 (0.0006)	0.0036 (0.01733)
Household size	0.0003 * (0.0002)	−0.0008 (0.0005)	0.0005 (0.0004)	−0.0167 (0.04808)
Farming experience	0.0001 (0.0001)	0.0007 * (0.0004)	0.0007 (0.0009)	−0.0341 (0.0222)
Farm size	−0.0002 (0.0002)	0.0005 (0.0004)	0.0007 (0.0009)	−0.0005298 (0.0008184)
Membership in association	0.0015 ** (0.0007)	0.0001 (0.0005)	−0.0005 (0.0006)	0.1022 (0.4034)
Access to credit	0.0001 * (0.0001)	−0.0002 (0.0004)	0.0005 * (0.0003)	0.0125 (0.0273)
Risk aversion	−0.0001 (0.0004)	−0.0009 * (0.0005)	0.0036 ** (0.0018)	0.1137 *** (0.0232)
Access to extension contacts	0.0001 (0.0005)	−0.0004 (0.0006)	−0.0030 * (0.0018)	0.0212 (0.0242)
Access to climate information	−0.0009 *** (0.0003)	−0.0009 (0.0007)	0.0005 (0.0004)	0.0434 (0.030)
Number of years of stay in the community	−0.0003 (0.0015)	0.0002 (0.0004)	−0.0002 (0.0004)	−0.0108 (0.0283)
Project community 1_Osun	0.0002 (0.0005)	−0.0001 (0.0001)	−0.0004 (0.0005)	−0.0060 (0.0228)
Project community 2_Osun	0.0007 (0.0005)	0.0001 (0.0002)	−0.0004 (0.0006)	0.0041 (0.0239)
Project community 1_Kwara	0.0001 (0.0006)	0.0004 (0.0006)	−0.0009 (0.0007)	0.0422 (0.026)
Project community 2_Kwara	0.0023 ** (0.0009)	0.0001 (0.0001)	0.0002 (0.0004)	−0.0169 (0.0248)
Constant	0.0017 *** (0.0004)	0.0012 *** (0.0004)	0.0007 (0.0009)	−0.0005 (0.0006)

Figures in parentheses are standard error; ***, **, * represent 1%, 5%, and 10% probability levels, respectively.

and the corresponding ATET and Rosenbaum Bounds Delta (Г) are presented in

Table 4. Estimates of the average treatment effect on the treated and Rosenbaum bounds.

	PSM(1)	PSM(2)	PSM(3)	PSM(4)
	Treated/Non-Treated and Control	PC/Control	Treated/Control	Treated/Non-Treated
ATET	0.2319 *** (0.0129)	0.0231 * (0.0122)	−0.04297 (0.04989)	0.1415 *** (0.0313)
Rosenbaum Bounds Delta (Г)	2.45+	2.15	3.35+	3.75+

Figures in parentheses are standard error; ***, * represent 1%, and 10% probability levels, respectively. + Upper bound.

. The results from Table 3 show that age (PSM 1), education (PSM 4), household size (PSM 1), farming experience (PSM 2), membership in the association (PSM 1), access to credit (PSM 1, 3), risk aversion (PSM 2, 3, 4), access to extension contact (PSM 3), access to climate information (PSM 1), and project community 2_Kwara are statistically significant to the estimation of propensity scores. The kernel distribution was used for matching where common support conditions were met. PSM (1) is highlighted in this study because they estimate the direct effect of the Jumpstarting Project on the adoption, indicating a 23.1% ATET with regards to the orange-fleshed sweet potato adoption and is significant at 1% probability level. Rosenbaum Bounds Delta (Г) of 2.45 is obviously within an acceptable range as suggested by previous studies [8,19,45]. This shows that the estimates are robust to the influence of unobservables. PSM (2) confirms possible spillover with a 4.2% difference between project communities and control groups, which is statistically significant indicating that the Jumpstarting Project has sustained the spread of OFSP adoption beyond project intervention communities. PSM (3) shows that the difference between the treated and the control is 4.2% although not statistically significant at 1%. PSM (4) shows that the difference between the treated and the non-treated 14.1% which is statistically significant at 1%. This implies that OFSP might have spillover to neighbors after the implementation of the Jumpstarting Project in the intervention communities.

Controlling for unobservables using instrumental variables shows that the Jumpstarting Project effects are greater than the initial estimates from the alternative model (i) and (ii) specifications. The results are presented in

Table 5. Instrumental variables regression: intention-to-treat (ITT).

	IV(1)	IV(2a)	IV(2b)
	(iii) First Stage	(iv) Second Stage: OlS	(iv) Second Stage: FR +
Treated	0.0892 *** (0.0035)
PC		−0.2504 *** (0.0043)	−0.2790 *** (0.0031)
Age	0.05842 * (0.03237)	0.0353 (0.1292)	0.0130 ** (0.0614)
Marital status	0.05642 ** (0.02751)	−0.0769 ** (0.0035)	−0.0987 ** (0.0464)
Education	−0.0936 (0.1039)	0.0739 * (0.0422)	0.0156 *** (0.0052)
Gender	−0.0026 (0.01811)	0.0788 *** (0.0280)	0.01002 ** (0.0453)
Household size	0.02824 (0.01228)	0.0144 (0.0120)	0.0109 (0.0191)
Farming experience	0.01068 (0.01062)	−0.0123 (0.0191)	−0.0365 (0.1280)
Farm size	−0.0752 (0.1149)	0.0983 (0.0103)	−0.1038 (0.1415)
Membership in association	0.01304 ** (0.0058)	0.01394 (0.0109)	0.0114 (0.0263)
Access to credit	0.000119 (0.0052)	−0.0462 (0.1386)	−0.0833 (0.1512)
Risk aversion	0.0875 (0.0753)	0.0605 (0.1476)	0.01346 * (0.0070)
Access to extension contacts	0.0187 (0.0164)	0.0129 (0.0636)	−0.0553 (0.1367)
Access to climate information	0.0524 *** (0.0111)	0.0560 (0.0537)	0.0772 (0.0719)
Number of years of stay in the community	0.0605 (0.1476)	0.0111 (0.0129)	0.0174 (0.0172)
Project community 1_Osun	0.01292 (0.0636)	−0.0211 ** (0.0106)	0.0139 (0.1958)
Project community 2_Osun	0.0560 (0.0537)	0.0803 *** (0.0209)	−0.0453 (0.4189)
Project community 1_Kwara	0.0111 (0.0129)	0.0134 * (0.0070)	0.0194 (0.0234)
Project community 2_Kwara	−0.0211 ** (0.0106)	−0.0553 (0.1367)	−0.0121 (0.2193)
Constant	0.0803 *** (0.0209)	0.0772 (0.0719)
R2	0.532	0.579
F	11.57	13.46

Figures in parentheses are standard error; ***, **, * represent 1%, 5%, and 10% probability levels, respectively; ⁺ marginal effects.

. The instrument used (ITT) is a strong instrument since the result of the F-stat from the first IV(1) post estimation stage has a value of 11.57. The estimated results of the second stage for the Jumpstarting Project outcomes differ slightly between OLS and FR, that is, IV(2a) and IV(2b) models. The mean effects of the JP on the orange-fleshed sweet potato adoption were 25% and 27.9% for IV(2a) and IV(2b) models, respectively, which was significant at 1% level of probability each. This result is consistent with the PSM (1) ATET estimates of 23.1% in Table 4. This consistency in the use of PSM and IV regression indicates robust results from alternative methodologies and, also, indicates the fact that the Jumpstarting Project has sustained and increased the adoption of orange-fleshed sweet potatoes even after years of its implementation in the project communities.

4.3. Estimation of the OFSP Adoption Effect on Yield and Income

Presented in

Table 6. Parameter estimates of the endogenous treatment effects (ETE) of yield (Model l) and income (Model 2) of adoption and non-adoption of OFSP.

	Model 1: Yield Effect				Model 2: Income Effect
	Adopters (Treated) (N = 112)		Non-Adopters (Non-Treated and Control (N = 208)		Adopters (Treated) (N = 112)		Non-Adopters (Non-Treated and Control (N = 208)
Variable	Coefficient	Std. Err.	Coefficient	Std. Err	Coefficient	Std. Err	Coefficient	Std. Err
Labor expenses	0.00328	0.0021	−0.0001	0.0004	−0.0003	0.0003	0.00001	0.0000
Fertilizer qty	0.0010 ***	0.0002	0.0040 ***	0.0006	0.0002 ***	0.0001	0.0593	0.3186
King J	0.0007	0.0010	−0.0960	0.1055	0.0008	0.0005	−0.0032	0.0021
Mothers’ Delight	0.0029 ***	0.0010	0.0003	0.0003	0.0011 ***	0.0003	−0.0039	0.0028
Age	0.0003	0.0002	0.1795 ***	0.0794	−0.00002	0.00007	0.0008	0.0012
Marital status	0.0002	0.0003	0.2379 ***	.0814	0.0013 ***	0.0004	−0.0007	0.0022
Education	0.0043 ***	0.0010	0.1285	0.1098	−0.00009	0.00014	0.0014	0.0009
Gender	0.0058 ***	0.0016	0.0002	0.0002	0.00001	0.00002	0.0011	0.0009
Household size	0.0070 ***	0.0021	0.0004	0.0004	−0.00008	0.00008	0.0043 ***	0.0013
Farming experience	0.0056 ***	0.0013	2.3312 ***	0.2907	−0.0009	0.0462	−0.0040	0.0038
Farm size	0.0001	0.0001	0.6933	0.7131	0.0647	0.0335	−0.0045	0.0036
Membership in association	0.0044 ***	0.0014	0.3601	0.4250	0.0479	0.0317	0.0029	0.0010
Access to credit	0.0043 ***	0.0013	−0.0782	0.1628	0.0134	0.0119	0.0030	0.5211
Risk aversion	−0.0011	0.0242	−0.00004	0.0004	0.00015	0.0003	0.0008	0.0009
Access to extension contacts	−0.0076	0.0433	−0.0010	0.0010	−0.0455	0.1036	0.0021	0.6209
Access to climate information	−0.0039	0.0028	−0.0002	0.0016	−0.0413	0.0428	0.0088 ***	0.0020
Years of stay in the community	0.0008	0.0012	0.1868	0.3021	0.0053	0.0087	0.0003	0.0244
Project community l_Osun	0.0042	0.0029	0.0006	0.0043	0.0082	0.0110	0.0136	0.0102
Project community 2_Osun	0.0007	0.0022	−0.00008	0.00006	0.1518	0.1686	0.0004	0.0019
Project community l_Kwara	0.0014	0.0009	−0.00008	0.00007	0.2291 ***	0.0599	−0.0024	0.0017
Project community 2_Kwara	−0.0024	0.0008	0.0004	0.0024	0.0008	0.0005	0.0306	0.3102
Constant	0.0002	0.0002	0.0001	0.0008	0.0011 ***	0.0003	0.0094 ***	0.0043
IMR	0.0021 **	0.0009	0.0388 *	0.0207	0.0443 *	0.0244	0.0153 *	0.0078
Endogeneity test chi-square (2) =	37.53 ***				46.92 ***
Wald chi2	3102.377
Prob > chi2	0.000
LR test of indep.	0.0261
Loglikelihood	527.281

NB: The second stage of ETE model (outcome equation) is presented. The first stage estimates of selection equation are equivalent to Model 1 in Table 1 ***, **, * represent 1%, 5%, and l0% probability levels, respectively; IMR = inverse mills ration derived from the selection equation: endogeneity test = treatment and outcome unobservables are uncorrelated.

is the result of the second stage of the endogenous treatment effects (ETE) regression model showing the determinants of yield and income effects of adopting the orange-fleshed sweet potato. The probit model was used in the first stage (selection equation) of the ETE model. However, the estimates of the probit model (selection equation) are equivalent to that of Model 1 obtained and presented in Table 2, hence the justification for the presentation of the second stage of the ETE model in Table 6, which comprises the yield effects (Model 1) and income effects (Model 2). Both the dependent variable and some of the independent variables are logged for the two models. Thus, the coefficients are regarded as elasticities.

Accordingly, Model l in Table 6 indicates that the quantity of fertilizer applied positively and significantly affects yields of adopters of the orange-fleshed sweet potato and non-adopters (producers of the white and yellow sweet potato). This implies that the application of fertilizer increases the yield of both OFSP and traditional white and yellow potato by 0.1% and 0.4% for a 1%. Adoption of Mothers’ Delight, one of the varieties promoted through the Jumpstarting Project, was positive and statistically significantly influenced yield of the adopters. Thus, the Mothers’ Delight variety increases the farmers’ yield by 2.9%. It should also be noted that this particular variety was popular among the adopters, which might be attributed to its high yielding attribute. Age and marital status were positively and significantly associated with the yield of non-adopters. This indicates that older married farmers will likely experience increased yield of the white and yellow potato due to the fact that they might have gained experience over the years that will help them improve their farming practices. Education, household size, gender, and farming experience positively and significantly influenced OFSP yield. This implies that education, household size, gender, and farming experience increase yield by 0.4%, 0.7%, 0.5%, and 0.5%, respectively. Farmers who are educated might have learned different training skills on potato production, which will subsequently help them increase their farm yield through knowledge transfer, thus helping them increase their yield. Farmers who have a medium to large household size might also experience more yield due to the fact that they have more available family labor to work on the farm. Access to more available labor might translate into improved farming activities/practices that will help increase farmers’ yield. Male OFSP farmers tend to increase yield because they have access to more farm resources when compared with their female counterparts. Farming experience also positively and significantly influenced the yields of white and yellow sweet potato farmers at a l% level of probability. This implies that farmers who were involved in sweet potato production for a long period of time are more likely to experience increased yields due the fact that they might have accumulated experience on farming practices that can increase farm yield. Furthermore, membership in an association and access to credit positively and statistically significantly affects OFSP yields at l% probability level each. This implies that being a member of the farmers’ organization and having access to credit facilities increases the yield of OFSP by 0.4% each. Farmers that are members of social networks or have access to loan facilities have the likelihood of having access to production resources through their association or purchasing of resources through the loan facilities obtained. This is expected to translate into increased yields for farmers.

The results show that the endogeneity test on correlation of treatment and outcome unobservables accepts the null hypothesis at p < 0.01. This indicates that there are unobserved variables affecting both adoption and yields of OFSP, thus justifying the choice of ETE model. The highly significant IMR obtained from the selection equation implies that the unobserved factors are positively correlated with OFSP yields.

Based on the result of the average treatment effect presented in

Table 7. Treatment effects of the adoption of OFSP on yield and income from the ETE.

	Model 1: Yield Effect		Model 2: Income Effect
	Coefficient	Standard Error	Coefficient	Standard Error
Average treatment effect (ATE)	0.7227 ***	0.1258	0.0442 **	0.0194
Average treatment effect on the treated (ATET)	0.3638 ***	0.3009	0.3061 ***	0.0102

***, ** represent 1%, 5% probability levels, respectively; ATE = average treatment effect (OFSP adopters vs. non-adopters); ATET = average treatment effect (OFSP adopters vs. non-adopters) on the treated, i.e., among OFSP adopters (explanations in the text).

, it can be inferred that the adoption of the orange-fleshed sweet potato accrued yield benefits for the adopters. This is shown by the positive and statistically significant ATE estimate. In addition, among the farmers who adopted OFSP, the parameter estimate of the ATET is positive and highly statistically significant. Estimating the difference between the calculated counterfactual yield among OFSP adopters (that is, the estimated yield they would have attained without adopting OFSP) and the addition of the counterfactual yield plus the estimated ATET translates into a yield gain of 36.3 kg/ha (Table 7).

In Model 2, the same set of explanatory variables was used to assess the determinants of income effects of the adoption of the orange-fleshed sweet potato and the results are presented in Table 6. Only the outcome equations (second stage) are also presented in Model 2 since the selection equation (first stage) is an equivalent of Model 1 in Table 2, previously presented. From Model 2, the quantity of fertilizer applied positively and statistically significantly affects income of the adopters of OFSP. This implies that fertilizer application increases the income of the adopters of OFSP by 0.2%. This might be connected to the fact that relatively little quantities of fertilizer are required for OFSP production if the farmer intends to apply fertilizer at all. This helps the adopters reduce production costs, thus translating into increased income for the OFSP farmers. Cultivation of the Mothers’ Delight variety was positive and statistically significant at a 1% probability level in influencing the income of the adopters of OFSP. This implies that the cultivation of the Mothers’ Delight variety increases the income of the adopters by 0.1%. This might be attributed to the fact that this particular variety (Mothers’ Delight) produces higher yield, which is expected to translate into higher income. Marital status was also positive and statistically significant at a 1% probability level. This indicates that married households are expected to have relatively large household sizes, which makes for available family labor. Availability of family labor ensures that the cost of hired labor is reduced, which helps reduce production costs. This is expected to lead to increased incomes for the adopters of OFSP. Residing in project community 1_Kwara was positive and statistically significant at a 1% probability level. This implies that adopters residing in the intervention communities experienced an increased income. For the non-adopters, household size was positive and statistically significant at a 1% level of probability. This implies that increased household size increases the income of the farmers. As earlier explained, households that are relatively large in size utilize family labor for farm production, thus reducing cost of hired labor in the process. This subsequently translates into increased incomes since the cost of hired labor takes a huge share of the cost of sweet potato production. Access to climate information was positive and statistically significant at l%. This indicates that having access to climate information affects the income of non-adopters. Farmers who have access to climate information are more likely to adopt climate adaptation strategies that will help them increase their farm yield and, subsequently, translate that into increasing their incomes.

Just like in Model 1 in Table 7, the endogeneity test in Model 2 and the highly statistically significant IMRs for the adopters and non-adopters indicate the existence of unobserved factors affecting both OFSP adoption and non-adoption, further justifying the choice of the ETE model as presented in Table 7. The results of the ATE and ATET for Model 2 are presented in Table 7 Model 2 produces a highly significant positive ATE of the OFSP adoption that is per unit income translating to NGN 44.2/kg. The model further produces a highly statistically significant and positive ATET of NGN 306.1/kg. It is, however, necessary to indicate that these results are very robust.

4.4. Effects of Adoption of Orange-Fleshed Sweet Potato on the Poverty Status

4.4.1. Estimated Incidence of Poverty Headcount, Depth and Poverty Severity

Presented in

Table 8. Poverty analysis of households by adoption status.

Poverty Indices	Pooled	Adopters	Non-Adopters
Poverty headcount	0.574	0.3864	0.6428
Poverty depth	0.2837	0.2528	0.3791
Poverty severity	0.2169	0.1563	0.2947

is the result of the computed poverty status of the sampled households by adoption status. In order to compute the poverty status of the respondents, a poverty line was constructed. The poverty line indicates the threshold with which a household can be separated as poor or non-poor. The poverty line computed for the adopters, non-adopters, and pooled sample was NGN 12,346, NGN 12,533, and NGN 12,465, respectively. The poverty line was computed as the two-third mean per capita expenditure per household. Households with per capita consumption expenditure above the poverty line were regarded as non-poor while those lower than the poverty line were classified as poor households. The FGT poverty analysis was also used to calculate all the relevant poverty indices, with the results from the computation of the FGT indices reported in Table 8.

The results indicate that the poverty headcount, poverty depth, and poverty severity were 0.3864, 0.2528, and 0.1563, respectively, for the adopters of the orange-fleshed sweet potato while for the non-adopters, it was 0.6428, 0.3791, and 0.2947, respectively. This is an indication of higher poverty headcount, poverty depth, and poverty severity among non-adopters compared with the adopters. Thus, about 64% of the non-adopters were living below the poverty line. The results of the poverty depth implies that the non-adopting households will need to mobilize additional 37% of their current amount in order to move completely out of poverty while it takes only 25% for the adopters. The result underscores the role of improved technology adoption, such as OFSP, in lifting farmers out of poverty. Although it has not completely eradicated poverty among the adopters, it has narrowed down the incidence of poverty among them to the level required to escape poverty.

4.4.2. Determinants of Poverty Headcount among the Adopters and Non-Adopters of OFSP

The result of the parameter estimates of the endogenous switching probit regression model (ESPRM) showing the factors affecting households’ poverty headcount among the adopters and non-adopters of OFSP are presented in

Table 9. Parameter estimates of the endogenous switching probit regression model.

Variable	Adopters		Non-Adopters
	Coefficient	Std. Err	Coefficient	Std. Err
Age	−0.018	0.013	0.053 **	0.022
Gender	0.133	0.264	0.010	0.014
Education	−0.060	0.068	0.032	0.032
Marital status	0.002	0.015	0.206	0.032
Household size	−0.093	0.119	−0.003	0.025
Farming experience	−1.137 ***	0.383	0.580	0.361
Farm size	−2.705 ***	0.384	−0.333	0.418
Membership in association	0.016	0.011	−0.038	0.054
Access to credit	−2.419 ***	0.385	0.050	0.053
Risk aversion	1.586 ***	0.308	1.280 **	0.634
Access to extension contacts	−0.391 **	0.191	0.061	0.450
Project community l_Osun	0.000	0.003	0.467	1.451
Project community 2_Osun	−0.010	0.0069	−0.233 ***	0.051
Project community l_Kwara	0.003	0.0064	−2.833 ***	0.945
Project community 2_Kwara	−0.297	0.279	−0.129 ***	0.020
Constant	−4.359 ***	1.120	5.7808 **	2.754
/athrhol	1.893 ***	0.715
/athrho0	0.203	0.296
rhol	−0.023 *	0.012
rho0	−0.409 **	0.156
lR chi2(l5)	143.80
Prob > chi2	0.0000

Note: ***, ** and * represent significance level at 1%, 5%, and 10%, respectively.

. The correlation coefficients rho_1 and rho_0 are statistically significant and negatively correlated indicating an association between adoption and poverty headcount. The results show that poverty headcount has increased significantly among non-adopting households by age and risk aversion while it has reduced significantly by residing in project community 2_Osun and project community 1and2_Kwara. As farmers grow older, their level of involvement in farming activities decreases, thereby becoming less productive. This, subsequently, affects their income, thus leading to the increased poverty headcount. In addition, non-adopters are risk averse; thus, they are not willing to invest in new improved technology that might potentially increase their productivity. Producing on the same scale with tradition technology over years might actually increase their poverty headcount. However, non-adopters who reside in intervention communities have the tendency to reduce their poverty headcount. They might have access to information on improved farming practices within their community as a result of spillover from participants in the intervention project, thus increasing their productivity and reducing their poverty headcount. For the adopters of the orange-fleshed sweet potato, the result shows that poverty headcount is reduced significantly by farming experience, farm size, access to credit, and access to extension contacts. Adopters who have been into sweet potato production for many years might have accumulated useful experiences, which can be transferred to their farm and increase their productivity. This tends to increase their income, thus reducing their household poverty headcount. Farmers who cultivate on relatively large farms are expected to have an increased income from their farm outputs, which is expected to help reduce their household poverty headcount. Access to credit affords the farmers the opportunities to purchase necessary farm resources and increase their scale of production. This will help the farmers increase their yields and income and, subsequently, reduce their households’ poverty headcount. Furthermore, having access to extension services affords the farmers the opportunity to have access to up-to-date information on current production practices, including how to adapt to current climate crises that is affecting/lowering crop yield. This information, if successfully applied, can help farmers increase their farm yields and income, which subsequently leads to reduced household poverty headcount.

4.4.3. Treatment Effects for Poverty from Endogenous Switching Probit Regression Model

Presented in

Table 10. Treatment effect of ESPM on poverty.

Outcome Variable	Treatment Effects	Average Treatment Effect (ATE)
Poverty headcount (%)	Adopted farm households (ATET)	0.053 ** (0.022)
	Non-adopted farm households (ATUT)	−0.217 *** (0.081)

Note: *** and ** represent significance levels at 1% and 5%, respectively.

is the result of the estimate of the average treatment effects on the treated (ATET), indicating the effect of the adoption of the orange-fleshed sweet potato on poverty status after accounting for observables and unobservables. The result indicates that the adopters benefitted from the adoption of the orange-fleshed sweet potato. For instance, the probability of being poor among the households that adopted OFSP would be 5.3% more if these households did not adopt OFSP as indicated by the ATET, which is statistically significant at a 5% level of probability. In a similar situation, if the non-adopting households had adopted the orange-fleshed sweet potato, the probability of being poor would have been 21.7% less. The implication of this is that the non-adopters would have been able to reduce their poverty headcount if they had adopted the orange-fleshed sweet potato, which is the average treatment effect on the untreated (ATUT). The ATUT is also statistically significant at a l% level of probability, implying that the non-adopters would be better off, that is, less poor, if they adopted the orange-fleshed sweet potato.

4.4.4. Robustness Check

The endogenous treatment effect (ETE) model and endogenous switching probit model (ESPM) have been found to be sensitive to the exclusion restriction assumption, hence the need for the use of PSM to check the robustness of the estimated effects from ETE and ESPM models. The same variables were used to estimate the propensity scores. Nearest neighbor (NNM) and kernel-based (KBM) matchings were used to estimate the effects of the adoption of the orange-fleshed sweet potato on yields, income, and poverty headcount. The results of the ATT estimated from the PSM matchings are presented in

Table 11. Effects of adoption of OFSP on yield.

Variables	Sample	Treated	Controls	Differences	Std. Err	t-Statistics
Nearest Neighbor Matching (NNM)
Yield (kg)	Unmatched	346.632	299.649	46.983	11.092	4.24
	ATT	346.632	306.531	40.101 ***	12.215	3.28
	ATU	299.649	311.736	12.087	-	-
	ATE			19.452	-	-
	Effect (%)			0.57 ***
Kernel-Based Matching
Yield (kg)	Unmatched	346.632	299.649	46.649	11.092	4.24
	ATT	346.632	309.562	37.070 ***	13.422	2.76
	ATU	299.649	310.184	10.535	-	-
	ATE			17.758
	Effect (%)			0.44 ***

Note: *** represent significance levels at 1%.

,

Table 12. Effects of adoption of OFSP on income.

Variables	Sample	Treated	Controls	Differences	Std. Err	t-Statistics
Nearest Neighbor Matching (NNM)
Income (NGN)	Unmatched	64,382.482	39,675.763	24,706.719	3856.254	6.40
	ATT	64,382.482	43,748.218	20,598.264 ***	4201.236	4.90
	ATU	39,675.763	48,046.372	8370.609	-	-
	ATE			9573.472	-	-
	Effect (%)			0.38 ***
Kernel-Based Matching (KBM)
Income (NGN)	Unmatched	64,382.482	39,675.763	24,706.719	3856.254	6.40
	ATT	64,382.482	43,559.852	20,822.630 ***	4673.727	4.46
	ATU	39,675.763	46,836.946	7161.183	-	-
	ATE			9915.57	-	-
	Effect (%)			0.27 ***

Note: *** represent significance levels at 1%.

, and

Table 13. Effects of adoption of OFSP on the poverty headcount.

Variables	Sample	Treated	Controls	Differences	Std. Err	t-Statistics
Nearest Neighbor Matching (NNM)
Poverty headcount (%)	Unmatched	0.53	0.71	−0.18	0.0536	−3.36
	ATT	0.53	0.67	−0.08 ***	0.0332	−2.41
	ATU	0.71	0.64	−0.07	-	-
	ATE			−0.09	-	-
Kernel-Based Matching (KBM)
Poverty headcount (%)	Unmatched	0.53	0.71	−0.18	0.0536	−3.36
	ATT	0.53	0.66	−0.13 ***	0.0628	−2.07
	ATU	0.71	0.64	−0.07	-	-
	ATE			0.09	-	-

Note: *** represent significance levels at 1%.

for effects on yield, income, and poverty headcount, respectively. With respect to the effects of the adoption of the orange-fleshed sweet potato on yields, there is a positive and significant average treatment effect on the treated of 40.101 kg and 37.070 kg for the NNM and KBM, respectively. The ATET is the average difference between kilogram yields of similar pairs of farm households but belongs to a different adoption status. In addition, the ATE that gives the average effects of adoption on yields for a farming household is 19.452 kg and 17.758 kg for NNM and KBM, respectively. Consequently, the adoption of the orange-fleshed sweet potato has increase yield by 44−57%.

The income from OFSP of adopters is positive and statistically significantly higher than that of the non-adopters by NGN 20,598.264 (Table 12). Thus, the adoption of OFSP is found to increase income by 27–38% of the adopting farming households. This result implies that an increase in household income as a result of the adoption of OFSP is expected to lead to a decrease in households’ poverty headcount. The PSM results presented in Table 13 consequently show that adoption of OFSP has the potential to significantly reduce the poverty headcount by 8–l3%.

5. Conclusions

Using household-level data in Nigeria, this paper examined how the implementation of a farmer-led multiplication–dissemination Jumpstarting Project has sustained the adoption of the orange-fleshed sweet potato after several years of its implementation. The study employed quasi-maximum likelihood fractional regression and PSM combined with instrumental variable regression to ascertain the sustainability of JP on the adoption of OFSP while the endogenous treatment effect (ETE) and endogenous switching probit model (ESPM) was used to examine the effects of the adoption of OFSP on productivity and poverty status of the farmers.

Our main empirical findings are as follows. First, the implementation of the Jumpstarting Project in project communities in the two states indicates that the adoption of OFSP was sustained and increased. The result of the PSM confirms possible spillover between project communities and control groups. Second, estimating the difference between the calculated counterfactual yield among OFSP adopters (that is, the estimated yield they would have attained without adopting OFSP) and the addition of the counterfactual yield plus the estimated ATET translates into a yield gain for the adopters. The result of the ATE and ATET for the income showed a highly significant positive ATE of the OFSP adoption that is per unit income translating to the income increase for the adopters. Third, the result of the average treatment effects on the treated (ATET) from the endogenous switching probit regression model showed that the probability of being poor for the households that adopted OFSP would be higher if these households did not adopt OFSP, and if the non-adopting households had adopted the orange-fleshed sweet potato, the probability of being poor would have been reduced.

5.1. Policy Implications

Targeting rural farmers and increasing their yields and income through the development of new and improved agricultural technologies have been one of the challenges of development and emergency response. However, to ensure the diffusion and adoption of these technologies, efforts in the form of intervention projects to promote the adoption of these technologies have a great role to play. This study used responses from counterfactual non-participants, control, as well as participants in the Jumpstarting Project, which allows for better attribution of the outcomes to the project and aims to contribute to the literature by providing a micro perspective on the effect of the intervention project on the promotion of the adoption of improved technologies since efforts to economically improve the rural livelihood of the farmers include the introduction of improved technologies to them.

Our results suggest that appropriate government agencies, including extension agents, should be motivated to intensify the campaign with the view of sensitizing and motivating farmers toward the adoption of improved technologies such as OFSP. In addition, to achieve the poverty alleviation goal in Nigeria, a subset of millennium developmental goals, it is important for government and research institutes such as IITA to consider the role played by intervention projects in the promotion of the adoption of improved agricultural technologies, which subsequently increases farmers yields and income and reduces the poverty headcount. This study, therefore, recommends replication and scaling up of the Jumpstarting Project in other states in Nigeria, having impacted positively on the poverty status of the adopters of OFSP in the study area.

The findings from this study will be useful for planning public investment in improved agricultural technologies in Nigeria and other developing countries. This study will also complement the existing studies that investigated the success of improved agricultural technologies, offer suggestions for replication and future improvements, and act as a springboard to undertake further detailed and comprehensive studies in Nigeria.

5.2. Limitations of the Study and Suggestions for Future Study

We are unable to capture the dynamic relationships between OFSP adoption, OFSP yield, and farm revenue since our empirical analyses are based on 1-year cross-sectional data. However, we think these are interesting topics for further investigation once the data needed are accessible.