Farmer preferences for adopting drought-tolerant maize varieties: evidence from a choice experiment in Nigeria. (Premilinary Draft-Not to be cited)

Authors: Zainab Oyetunde-Usman and Apurba Shee

Submitted for presentation in the lightning session of the 2021 AAEA & WAEA Joint Annual Meeting.

Abstract

Maize is an important food crop in sub-Saharan Africa and provides staple food needs for the majority of rural maize farm households, yet it is highly vulnerable to prolonged dry spells and drought stress which has impacted poor productivity and welfare. In Nigeria, a similar event unfolds in the maize production system despite the availability of the drought-tolerant maize varieties. Poor adoption of drought-tolerant maize varieties persistently derails the goals of improving food security and the welfare of the populace. This study questions the likelihood of farmers’ perceived preferences for various drought-tolerant maize traits and the implicit value they are willing to place on it, as a drawback to rapid adoption. This research thus adopts the DCE approach to elicit preference information on the attributes and the implicit price farmers are willing to pay using choice cards and household data from 320 maize farm households in Gwarzo and Rano local government areas in Kano State Nigeria. The Mixed logit and Hierarchical Bayesian model are comparatively used for more robust analysis in preference and willingness to pay space. The result shows that the most preferred attributes by farm households in the study areas include high drought-tolerant, yield, high nitrogen use efficiency, and early maturing attributes. The WTP estimates show that maize farmers place a high value on drought-tolerant attributes which is 1.11 times and 1.29 times more than the value farmers place on yield and high nitrogen use efficiency attributes respectively. Also, a very high negative turnoff is expressed for preference for non-resistance to the Striga attribute and the value farmers are willing to place on it. This study recommends policy needs to facilitate the design of maize varieties in a way that preference attributes are reflected, including creating targeted awareness on the needs of different interest groups.

Introduction

The underlying goal of agricultural technology interventions is to enhance rapid uptake among farm households to solve problems of low productivity, food insecurity, and poor welfare (Asfaw et al. 2012; Mathenge, Smale and Olwande 2014; Awotide et al. 2015; Abdoulaye, Wossen and Awotide 2018). However, low and slow adoption plagues agricultural technology adoption in various contexts across sub-Saharan Africa ( SSA) low (Abebe et al. 2013; Ward et al. 2018; Kagoya, Paudel and Daniel 2018). This also applies to the adoption of drought-tolerant maize varieties (DTMVs); a low-cost initiative to tackle climate risk among maize farm households in SSA. The DTMVs under the drought-tolerant maize for Africa (DTMA) project was introduced to 13 countries including Nigeria across SSA to provide resilience to drought and climate variation and also to improve farmers' welfare (Kostandini, La Rovere and Abdoulaye 2013). Among the objectives of the DTMA project was to increase maize yield under drought stress conditions and also to increase the average productivity of small farm-holders by 20-30% (Fisher et al. 2015). Evidence of the impact of DTMVs have been widely assessed across adopting countries in SSA and has been found positive and significant in increasing productivity, household welfare, and food security among adopters ( Bezu et al. 2014; Wossen, Abdoulaye, Alene, Feleke, et al. 2017; Abdoulaye, Wossen and Awotide 2018; Kassie et al. 2018). In particular to Nigeria, the positive impact of adopting DTMVs on productivity and welfare has been established and found to increase maize yields by 13.3%, reduced the level of variance by 53%, and downside risk exposure by 81% among adopters (Wossen et al. 2017). Despite this, significant variations exist in the adoption of DTMVs in drought-prone agricultural zones in Nigeria (Abdoulaye et al. 2018). The situation is unsettling with prevailing drought and desertification issues in agricultural areas zones in Nigeria (Medugu, Majid and Choji 2008; Eze 2018; Hassan, Fullen and Oloke 2019). There are also growing concerns on projected loss in maize yield by 20% or more in 2050 due to dryer and hotter temperature(Lobell et al. 2011). This might complicate the immediate welfare of the rural agricultural populace in which recent poverty indices show that, of the people that live below $1.90 per day, 84.6% lived in rural areas (World Bank Brief, 2020). Also, the food insecurity level may be worsened as Nigeria is currently ranked 98th out of 107 countries with a global hunger index of 29.2 and the food insecurity is classified as serious in the severity chart (GHI, 2020).

Understanding the constraints or incentives that influence farm households’ decision to adopt DTMVs is key to enhancing adoption and in past studies, poor adoption of DTMVs has been attributed to differences in topography, plot, and socioeconomic conditions (Fisher and Carr 2015; Holden and Fisher 2015; Katengeza, Holden and Lunduka 2019). This approach, although significant for policy designs focusing on observable factors of farm households, may not be enough as several underlying behavioural attributes and perception to adoption may have been overlooked. To highlight, the design of DTMVs incorporates several other attributes in varying forms that can pique the interest of maize farmers differently and can inform their decisions to adopt or not. This suggests that it is not enough to judge the adoption of DTMVs based on households’ characteristics without considering choices of DTMVs attributes that maximise households’ utility. How to promote DTMVs that better reflects farm households’ constraints and preferences is a question relevant in the face of increasing climate risks and given projections of increasing population and the need to meet food needs and improve farm households’ welfare.

A discrete choice experiment is a popular approach to eliciting preference information and a large body of literature reports the use of DCE in various contexts in applied behavioral studies in fields such as management, hospitality, health, transport (Jagger and Jumbe 2016; Alemu and Olsen 2018; Penn and Hu 2020; Hu et al. 2020; Potoglou et al. 2020; Chen 2021; Nthambi, Markova-Nenova and Wätzold 2021) inclusive of a broad presence in the field of agriculture and natural resources management (Ward et al. 2014; Ward and Singh 2015; Ward et al. 2016; Owusu Coffie et al. 2016; Joshi, Khan and Kishore 2019; Krah et al. 2019a; Oyinbo et al. 2019; Teferi et al. 2020; Shee, Turvey and Marr 2020). In this context, the DCE has been applied to assessing trait preferences in improved seeds; for example, Ward et al. 2014 employed DCE to examine farmers’ preferences for drought-tolerant traits in rice and explore heterogeneity in these preferences in India. Within the context of managing weather risks and access to credit for farmers; Teferi et al. 2020 use DCE to investigate the preferences and willingness to pay for frost and yellow resistance grain yield, grain colour, plant height, and length of maturity traits for wheat in Ethiopia. The use of DCE extends to examining preferences in climate-smart agricultural techniques (Schaafsma, Ferrini and Turner 2019; Krah et al. 2019a). Within the framework of managing agricultural risks, Shee et al. 2020) applied DCE to examine the demand and supply side preferences for insurance-linked credit among farm households in Kenya.

Based on this background, this study uses the DCE experiment approach to elucidate information on maize farm households’ preferences for DTMVs. Given that the price effect may play a role in the demand for improved maize varieties, we explore the implicit price they are willing to pay for defined attributes. Attempt to elicit information on trait preferences in DTMVs is still quite scarce. A recent one is in Kassie et al. 2017, which examined taste parameter preferences and the implicit price maize farm households are willing to pay for in Zimbabwe. However, estimation is limited to the conventional approach. This study differs by comparatively employing both the conventional classical approach and the Bayesian estimation approach in preferences and willingness-to-pay space.

This paper further contributes to better understanding the characteristics of the trait in DTMVs that are valuable to farmers to guide plant breeders and policymakers in the design and promotions of DTMVs in Nigeria. The objectives are as follows: (1) to elicit information on farmers' preferences for DTMVs (2) examine the heterogeneity in DTMVs preferences (3) estimate the willingness to pay for DTMVs attributes.

The next section provides some background information generally on maize, DTMVs, and their attributes in Nigeria. The third section describes the data and empirical framework, the results and discussion were presented in section four and the last section concludes the paper.

2.0 Context of the Study

In Nigeria, maize is the second most widely grown crop after Cassava based on the land areas covered and production indices (FAOSTAT, 2019). It is cultivated on about 4.9 million hectares of land (FAOSTAT, 2019) and available in most of the country especially in the Savanna zone due to the presence of high radiation which is favourable for its growth (Bello et al. 2014). In West Africa, Nigeria contributes 43% of maize grown (Tegbaru et al. 2020), and based on recent production statistics, Nigeria is the second-largest maize producer in Africa with over 100 million tonnes after South Africa (FAOSTAT, 2019). Maize as a staple food crop is consumed in many forms as a main dish, infant foods and snacks, and a core ingredient in animal feed (Ekpa et al. 2018; Adewopo 2019). Despite its importance, a major concern is a lag in productivity which has not kept pace with the rate of growth in the area and with population growth. By way of example, comparing yield in 1961 (805t/ha) to 2018(2092.4t/ha) there is a gradual increase of approximately 1.6%, however, growth in area and population increased by 2.5% and 3.3% respectively in the same period (FAOSTAT, 2020). Also, among the top producers in Africa, Nigeria is marginally a net maize importer (Adewopo 2019). The situation is disconcerting as there are projections of the population doubling by 2015 (IF, 2019) and this may worsen existing food insecurity and welfare issues.

Like other SSA countries, drought is a major culprit in the lag in maize productivity (Fisher et al. 2015). Similar, to most SSA agricultural systems, agricultural production is rainfed and thus, crop failure is inevitable due to prevalent extreme conditions such as erratic rainfall patterns, extreme temperatures. In maize production, drought stress attacks the most critical stages of growth which include, early in the growing season, flowering stage, and the mid to late grain filling (Heisey and Edmeades, 1999). This usually results in a reduction in the number of ears per plant and grain yield (Azeez et al. 2005; Olaoye et al. 2009). Beyond poor productivity, drought stress indirectly aggravates poverty and food insecurity level, for example, extreme cases of drought stress have led to famine (Mortimore 1989). Also, drought in the periods of 1971-1972 exacerbated existing poverty and starvation and significantly reduced agricultural contribution to the GDP from 18.4% to 7.3% (Abubakar and Yamusa, 2013). The introduction of DTMVs provided a means to mitigating climate risk at important growth stages and can increase the productivity of farm households by 20-30% (Kostandini et al. 2015).

The DTMVs compared to traditional varieties achieve higher productivity features in the face of extreme drought conditions because it exhibits trait features that manifest in the early stage seedling vigor and leaf rolling, in this case, leaf rolling takes longer under early-season drought stress and the flowering stage has shorter or narrower anthesis silking interval, also crop show stay-green attributes under moisture stress (Kassie et al. 2017). Besides, some cultivars of the DTMVs have traits that are resistant to diseases such as the maize streak virus and enhances better tolerance to low soil nitrogen (Fisher et al. 2015). Plant breeders have also made efforts to incorporate some DTMVs with strains that are resistant to parasitic weed problems (Azeez et al. 2005). Parasitic weed such as Striga hermonthica infestations is highly endemic in the northern region of Nigeria where cereal production is dominant and it constitutes one of the most severe constraints to production (Dugje, Kamara, and Omoigui 2006; Kamara et al. 2020). Striga infestations can account for as high as 100% of the farmlands and can force farmers to abandon their farmlands (Ekeleme et al. 2014). Some other DTMVs features include extra-early maturing features (less than 90 days), increasing values in the number of ears per plant, and number of kernels per ear (Kassie et al. 2017). With the aforementioned, the adoption of DTMVs provides several features beyond mitigating drought, however with very low demand and variation in adoption, it suffice to say that DTMVs have received little attention in terms of farm-households’ attributes preferences which may be constraining adoption. It is also important to note that the ultimate measure of disseminating improved varieties is to ensure that demand and adoption is rapid among users. As such, adequate understanding of DTMVs preferred attributes will help plant breeders to integrate this into product profiles which will help to increase adoption of DTMVs, meet the food needs of the populace and improve farm households’ welfare.

3.0 Empirical Methodology

The study uses experimental choice modelling methods to assess maize farmers' preferences for DTMVs attributes. Discrete Choice Experiment (DCE) is rooted in consumer behavior theory in which the optimal goal is utility maximization (Lancaster, 1966). Following Louviere, Flynn, and Carson 2010, the use of DCE in adoption theories is consistent with the economic random utility theory in which individuals are assumed to have a ‘utility’ for each choice of alternative technologies which cannot be observed known as ‘latent utility’. This latent utility comprises of the systematic and random components. Similar to the aforementioned conventional approaches, the DCE systematic components explain the differences in choices and observable heterogenous factors driving adoption, while the random components explain the unobserved factors affecting choices. Accordingly, heterogeneity can also exist in unobservable random components associated with individuals and not necessarily the choices.

In the case of this study, we applied the DCE in assessing preferences for DTMVs attributes which constitute attributes of yield, maturity, Striga-resisting features, maize cob features, and nitrogen use efficiency capabilities. The DCE approach also allows the interaction of some households’ factors to enable heterogeneity estimations of decisions on attributes. Finally, the DCE allows the estimation of marginal preferences for each DTMVs attributes.

3.1 Design of Discrete Choice Experiment

a. Identifying Attributes

The design of a choice experiment typically follows a five-stage approach which includes identification of attributes, identification of levels, experimental design, data collection, and analysis of data (Kjær 2005). Identifying attributes involve various consultation and collaborations with identified stakeholders. For this study, identifying attributes for choice experiment involved collaborating first with the research team at the IITA responsible for the deployment of DTMA intervention specifically in Kano State, Nigeria where the study took place. Within the team, we consulted with the plant breeders in the IITA under the DTMA Nigeria team to understand existing varieties in the study location. The collaborative effort also includes discussing with team leads at the forefront of DTMVs intervention, this includes extension workers, key farmers working with a network of farmers and seed dealers in the regions under study. An additional effort was exploring a database of existing maize varieties that are drought tolerant and have been deployed for farmers on the Nigeria seed portal database. Attributes and level adapted were based on all consultations and this is presented in Table 1. The attributes selected for our choice experiment include yield, maturity/duration, resistance to Striga, nitrogen use efficiency, cob size, grain size, price, and tolerance to drought.

Yield is an important attribute as it defines farm households' earnings and overall welfare and food security, and in most empirical study is a highly preferred trait (Dao et al. 2015; Kassie et al. 2017). It as well defines the main objective of ensuring food availability in retrospect to the adoption of agricultural technology from as early as the green revolution era (Evenson and Gollin 2003; Zilberman et al. 2005). In this study, we considered potential yield attainable under favourable conditions as designed by plant breeders for varieties of drought-tolerant maize, ranging from a lower rank to a higher rank this include four levels which are 3.0t/ha, 6.0t/ha, 8.5t/ha and 9.0t/ha. Studies have shown that potential attributes of yield as high as 10 t/ha in drought-prone zones, provided conditions are favourable (African Centre for Biodiversity ACB, 2017).

Disparities in attributes of DTMVs includes preferences for either early maturing, medium or late maturing. In the scientific context, duration or maturity periods can be advantageous or otherwise depending on the farmers' knowledge of use or preferences. For example, the late maize varieties when sown earlier can utilize the longer period of grain filling and be found to gain more yield than early maturing varieties (Bello et al. 2012). At the same time, late maize varieties have the tendency of having high plant height which makes them susceptible to lodging, in contrast, early varieties were found to be shorter, high yielding, better nitrogen use efficiency, and agronomic performance (Liu and Wiatrak 2011). Our experimental choice design includes three-level for duration attribute which includes early for varieties that mature within 90 days; medium for varieties that mature between 90 to 120 days and late for varieties that mature after 120 days. Nitrogen is one of the yield-limiting nutrients in maize production in Nigeria (Kamara, Ewansiha, and Menkir 2014). It is highly mobile and subject to excessive loss in the soil (Pasley et al. 2020). Among existing varieties deployed are maize cultivars that perform well under low nitrogen conditions (Bänziger and Lafitte 1997; Kamara et al. 2005; Badu-Apraku et al. 2019). Also, cultivars that are tolerant to drought are efficient in the uptake of Nitrogen suggesting that such cultivars require less investment in nitrogen fertilizer application (Kamara, Ewansiha and Tofa 2019). Based on this, in our experimental study, we incorporated three levels of nitrogen use efficiency; low, medium, and high.

Cob and grain size are common attributes in maize cultivars (Abate et al. 2013; Tadesse, Buah, et al. 2013; Medhin and Ayalew 2014) and their sizes have a lot to do with the potential yield and marketability of outputs (Kassie et al. 2017). In our experimental design, cob-size levels include small, medium, and large, while grain size levels are small and large. We varied drought-tolerant levels into low, medium, and high to assess households’ perception of their degree of preference drought-tolerance attributes in maize. The price attribute represents the common cost of a kilogram of seed farm households ever purchased DTMVs. This will be used to estimate the willingness to pay in preference and WTP- space.

b. D-optimal choice set design

In designing choice sets, it is important to control for the non-dominance of one attribute over the other. The D-optimal design approach takes this into account by explicitly considering the importance of attribute levels and at the same time ensuring that the alternatives in the choice set provide more information about the trade-off between the different attributes (Carlsson and Martinsson 2003). We specified the D-optimality criterion using a modified Federov search algorithm based on calculating the determinants of the variance-covariance matrix of the parameters from a non-linear logit model as applied in similar contexts (see. Shee, Turvey and Marr 2020). From the eight attributes, we generated varying choices using the JMP software. In the design process, we considered varying four key attributes while four remain fixed. This is to give the respondents a clearer view of choices and to understand the changes appropriately. The fixed attribute in the sample choice card include yield, maturity/duration, price of maize grain per kg, and tolerant to drought, while others varies. The result generated a set of 50 unique choice sets which were assigned to 5 different blocks, such that each respondent was required to respond to 10 choice sets. The choice set were constructed with three alternatives including an opt-out option (see figure 1). In the case of this study, the opt- out option is necessary as it represents real life situation and reflects existing preferences for non-drought tolerant or traditional varieties. Overlooking the effect of opt-out effect in DCE simply show that participants preferences are not adequately considered and could lead to inaccurate measurement of attribute preferences and error in policy recommendation (Boxall, Adamowicz and Moon 2009; Campbell and Erdem 2019). Also, the DCE is highly susceptible to hypothetical bias (Lusk and Chroeder 2009; Moser, Raffaelli and Notaro 2014). However, in an unforced situation such as the use of the opt-out option does not completely exclude errors in estimation, the common bias in the experimental process is the omission and avoiding choice which may include opting for the opt-out option which is simpler to understand (Boxall et al. 2009). To control for this, participants are repeatedly reminded of the free-will to not choose between the designed drought-tolerant varieties. Such an approach culminates into a repeated reminder of the opt-out approach which has been found to reduce or mitigate hypothetical bias (Ladenburg and Olsen 2014; Alemu and Olsen 2018). In our experimental approach, the opt-out option is repeated in each choice card and participants are made to understand that opt-out options represents their non-interest in any of the drought- tolerant choices and indicate their preferences for lowest levels in each attributes which are relatively close to attributes of traditional varieties and serves as the status quo for this study.

Table 1. Discrete Choice Experiment Attributes and Corresponding Levels

Attributes Description Levels

Yield Grain yield measured in ton/hectare 3.0t/ha, 6.0t/ha. 8.5t/ha, 9.0t/ha Maturity/duration Measured by the time from planting to maturity Early maturing, of maize crop. Less than 90 days is early medium maturing and maturing; between 90 to 120 days is medium late maturing. maturing and more than 120 days are late maturing. Resistance to Striga The ability of maize variety to resist parasitism by Yes, No Striga hermonthica; a popular parasitic weed in cereal production Nitrogen use efficiency The ability of maize variety to efficiently take up Low, medium, high nitrogen in the soil Cob size Observation is jointly based on the maize length Small, medium, and and diameter. large Grain size Observation based on the relative kernel size Small, large

Price of maize seed per Maize seed price in Nigerian naira/kg 200, 250, 300, 400 Kg (Nigerian Naira) Tolerant to drought The ability of maize variety to have high seedling Low, medium, high vigor, shorter/narrower anthesis silking interval, and a stay-green attributes under moisture stress

Figure 1. Sample Choice card

Block1- choice set 9 of 10 MAIZE SEED MAIZE SEED A MAIZE SEED B MAIZE SEED C Opt-out CHARACTERISTICS Yield I DO NOT LIKE ANY OF THESE 3.0t/ha 3.0t/ha 3.0t/ha CHOICES Maturity/Duration

Medium maturing Late maturing early maturing Resistance to Striga

Yes Yes Yes Nitrogen use efficiency

High Medium Low Cob size

Medium Medium Medium Grain size

Small Small Small Price of maize grain per kg

300 200 400 Tolerant to drought

Low High Medium Tick selected choice

3.2 Econometric Framework

In the case of this study, maize farm households choose their most preferred attributes of DTMVs among different alternatives presented to them in a choice card. Based on the assumption of utility maximization (McFadden 1974), we model farm households choices based on using a random utility framework which states that the utility an individual obtains from choosing is and can be illustrated as follow: 𝑖𝑖 𝑡𝑡 𝑈𝑈𝑖𝑖𝑖𝑖 = + (1) ′ Where represents the latent𝑈𝑈 𝑖𝑖𝑖𝑖unobservable𝛽𝛽 𝑉𝑉𝑖𝑖𝑖𝑖 𝜀𝜀 maximum𝑖𝑖𝑖𝑖 utility that individual with choice alternative ; is the systematic observable component of utility (deterministic component) 𝑈𝑈𝑖𝑖𝑖𝑖 𝑖𝑖 that individual associates with alternative , this aspect affects the respondents’ decisions. 𝑡𝑡 𝑉𝑉𝑖𝑖𝑖𝑖 represents the vector of coefficient estimates and is the random or error component 𝑖𝑖 𝑡𝑡 𝛽𝛽 associated with individual and choice . The error component is expected to follow the 𝜀𝜀𝑖𝑖𝑖𝑖 identically and independently distributed (IID) Gumbel distribution (Meyerhoff, Boeri and 𝑖𝑖 𝑡𝑡 Hartje 2014)and it is probabilistic and describes how individual choices among alternatives impacts changes in either choice option or individual covariates or both. The probabilistic discrete choice model of individual associated with choice from a set of competing alternatives can be illustrated given a choice set , utility is maximised if the probability of 𝑖𝑖 𝑡𝑡 adopting choice is greater than the probability of adopting all other j alternative choices in 𝐶𝐶𝑡𝑡 set . 𝑖𝑖 ( |𝐶𝐶𝑡𝑡 ) = [( + )] > ( + )], for all j alternatives in choice sets . (2)

It𝑃𝑃 i𝑖𝑖s 𝐶𝐶important𝑡𝑡 𝑃𝑃 𝑉𝑉𝑖𝑖𝑖𝑖 to note𝜀𝜀𝑖𝑖𝑖𝑖 that 𝑀𝑀𝑀𝑀𝑀𝑀variations𝑉𝑉𝑗𝑗𝑗𝑗 exist𝜀𝜀𝑗𝑗𝑗𝑗 in assumptions to the distributions𝐶𝐶 of𝑡𝑡 the random utility components . From equation 1, we first model the probability of individual choosing alternatives over other alternatives in a choice set represented by a multinomial 𝜀𝜀𝑖𝑖𝑖𝑖 𝑖𝑖 logit (MNL) model as follows: 𝑡𝑡 𝑗𝑗 ( ) = (3) 𝑒𝑒𝑒𝑒𝑒𝑒 𝑉𝑉𝑖𝑖𝑖𝑖 𝑖𝑖𝑖𝑖 𝑗𝑗 𝑃𝑃 ∑𝑗𝑗=1 𝑉𝑉𝑖𝑖𝑖𝑖 Where = and is the choice set. represents the mean marginal utility for each of the DTMVs attributes′ . However, there is a shortcoming in the assumption of the Independence 𝑉𝑉𝑖𝑖𝑖𝑖 𝛽𝛽 𝑉𝑉𝑖𝑖𝑖𝑖 𝐽𝐽 𝛽𝛽 of irrelevant alternatives (IIA) property of the MNL model. This assumes that maize farmers' preferences for DTMVs attributes are the same across all choices, this does not hold if the preferences of individuals and attributes of DTMVs are different. In the case of this study, maize farm households’ characteristics are heterogenous, it is expected that their preferences for DTMVs attributes are also heterogenous. The MXL or random parameters logit allows for modelling such heterogeneities and it has been applied in similar scenarios (Hole and Kolstad 2012; Joshi et al. 2019; Krah et al. 2019b; Shee et al. 2020). The mixed logit model (MXL) also known as the random parameter logit overcomes this limitation of the MNL by allowing random parameter estimates to vary across maize farmers, the choice of attributes, and the alternatives (Ward and Singh 2015). The MXL model also allows for the estimation of heterogeneities within the observed and unobserved attributes of the data set while varying the parameter estimates. it allows for random taste variations, unrestricted substitution patterns, and the correlation of error terms (Meyerhoff et al. 2014). In this study, we express the MXL model with and without farm households’ heterogeneities. Applying the MXL model to our study excluding farm household heterogeneity, we express the probability that individual selects a set of DTMVs attributes in a choice set is expressed as follows:

( ) 𝑖𝑖 ( | , ) = 𝑡𝑡 𝑗𝑗 (4) ( ) 𝑛𝑛 𝑁𝑁 exp 𝑉𝑉𝑡𝑡𝑡𝑡𝑡𝑡 𝑖𝑖 𝑖𝑖 𝑖𝑖 𝑛𝑛=1 𝑗𝑗 𝑃𝑃 𝑘𝑘 𝛽𝛽 𝑉𝑉 ∐ ∑𝑗𝑗=1 exp 𝑉𝑉𝑗𝑗𝑗𝑗𝑗𝑗 Where represent the series of choices over N choice scenarios presented to individual . To ensure that𝑛𝑛 parameter estimates are predicted accurately, random variation is accommodated 𝑘𝑘𝑖𝑖 𝑖𝑖 to ensure that preferences vary across individuals in the estimation process. Also, to estimate heterogeneity resulting from individual-specific factors, we interact utility estimates of the DTMVs attributes with some socioeconomic characteristics of maize farm households in the MXL model by first expressing a linear utility specification expressed as follows:

= + + (5) ′ ′ Where is the utility𝑉𝑉𝑡𝑡𝑡𝑡𝑡𝑡 associated𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝛽𝛽𝑖𝑖 with𝑧𝑧𝑖𝑖 𝛿𝛿 individual𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 choosing during choice task , is a vector of the attributes for the nth alternative, is a vector of individual taste parameter s 𝑉𝑉𝑡𝑡𝑡𝑡𝑡𝑡 𝑖𝑖 𝑡𝑡 𝑛𝑛 𝑥𝑥𝑡𝑡𝑡𝑡𝑡𝑡 mapping these attributes into utility, is a vector of terms defining individual-specific 𝛽𝛽𝑖𝑖 characteristics and maps these characteristics into utility associated with the choice of a 𝑧𝑧𝑖𝑖 particular alternative. The individual-specific functions are interacted with the attributes of 𝛿𝛿 the DTMVs and the researcher specifies a distribution for , ( | ) where are the parameters of the distribution which has a mean vector and covariance matrix , ~ ( , ). 𝛽𝛽𝑖𝑖 𝑓𝑓 𝛽𝛽 𝜗𝜗 𝜗𝜗 The probability that individual choosing alternative during choice task is expressed as: 𝑏𝑏 𝑆𝑆 𝛽𝛽𝑖𝑖 𝑁𝑁 𝑏𝑏 𝑆𝑆 𝑖𝑖 exp( ) 𝑡𝑡 𝑛𝑛 ( ) = ( | ) (6) exp′ 𝑥𝑥𝑡𝑡𝑡𝑡𝑡𝑡𝛽𝛽𝑖𝑖 𝑡𝑡𝑡𝑡𝑡𝑡 𝑗𝑗 𝑃𝑃 𝜗𝜗 � �𝑖𝑖 ′ 𝑓𝑓 𝛽𝛽 𝜗𝜗 𝑑𝑑𝑑𝑑 Where is the number of alternatives∑𝑗𝑗=1 presented�𝑥𝑥𝑡𝑡𝑡𝑡𝑡𝑡𝛽𝛽 𝑖𝑖in� each choice task, ( | ) is the probability density function of the taste parameters conditioned by a vector parameter characterizing the 𝑗𝑗 𝑓𝑓 𝛽𝛽 𝜗𝜗 distribution. is the vector of non-random alternative specific coefficients that captures the effects on the choice probability. Once the distribution of the taste parameters has been 𝛿𝛿𝑡𝑡 specified, equation (5) can be estimated using the simulated maximum likelihood method.

For a more robust analysis, this study employs the Hierarchical Bayesian mixed logit model. The advantage of the Hierarchical Bayesian procedures over the conventional classical mixed logit method is that the common approach of using maximization of the likelihood function in classical methods is not required in the Bayesian procedure, this help to overcome the problems of convergence which can be due to poor starting values in the model or the inclusion of bounded distributions (Train 2002). Also, the Bayesian procedure, under more relaxed conditions attain desirable estimation properties such as consistency and efficiency (Train 2002). The Bayesian approach is not completely elusive of issues, because it uses an iterative process to converge with a sufficient number of iterations to draw from the posteriors, estimations can as well be complicated (Regier et al. 2009). Due to the Bayesian approach of using the Markov Chain Monte Carlo (MCMC) techniques, it is usually not easy to ascertain if the MCMC algorithm is drawing from the distribution. Notwithstanding, this study applies both approaches with a view to comparatively inform the relative advantage of the mixed logit model and the hierarchical Bayesian approach in mixed logit.

Following Shee et al. 2020, we specified the prior beliefs about and are specified as ~ (0, ), is large, and ~ ( , 1) for 1, where stands for inverted Gamma 𝑏𝑏 𝑆𝑆 𝑏𝑏 𝑁𝑁 𝑣𝑣 𝑣𝑣 𝑆𝑆 𝐼𝐼𝐼𝐼 𝑣𝑣 𝑣𝑣 → 𝐼𝐼𝐼𝐼 distribution. The parameters and are called population level parameters. We use Gibbs sampling to estimate three sets of parameters , . The posterior for , is: 𝑏𝑏 𝑆𝑆 𝑖𝑖 𝑖𝑖 𝑏𝑏 𝑆𝑆(𝑎𝑎𝑎𝑎𝑎𝑎| 𝛽𝛽)∀𝑖𝑖 𝑏𝑏 𝑆𝑆 𝑎𝑎𝑎𝑎𝑎𝑎 𝛽𝛽 ∀𝑖𝑖 ( , , | ) ′ , ( , ) (7) exp�𝑥𝑥𝑡𝑡𝑡𝑡𝑡𝑡𝛽𝛽𝑖𝑖� 𝑖𝑖 𝑖𝑖 𝑗𝑗 ′ 𝑖𝑖 𝐾𝐾 𝑏𝑏 𝑆𝑆 𝛽𝛽 𝑌𝑌 ∝ ∐ ∑𝑗𝑗=1 exp�𝑥𝑥𝑡𝑡𝑡𝑡𝑡𝑡𝛽𝛽𝑖𝑖� ∅ 𝛽𝛽 𝑏𝑏 𝑆𝑆 𝑘𝑘 𝑏𝑏 𝑆𝑆 The Gibbs sampling involves taking a sequence of draws in which each draws for a parameter is estimated conditional on the parameters in the model in a hierarchical procedure. During estimation, the algorithm starts with initial values of , and . The nth iteration of the Gibbs sampling can be estimated using the following steps0 (1)0 take 0a draw of , from ( , ) 𝑏𝑏 𝑆𝑆 𝛽𝛽𝑖𝑖 where is the mean conditional on and ; (2) take a draw of conditional𝑛𝑛 on and 𝑏𝑏 𝑓𝑓 𝛽𝛽 𝑆𝑆 and (3) take a draw of conditional0 on 0 and . These steps𝑛𝑛 are repeated sequential𝑛𝑛 ly0 𝛽𝛽 𝑆𝑆 𝛽𝛽𝑛𝑛 𝑆𝑆 𝑏𝑏 𝛽𝛽𝑖𝑖 over many iterations until𝑛𝑛 the values have converge𝑛𝑛 𝑛𝑛d to draws in the posterior. A number of 𝛽𝛽𝑖𝑖 𝑏𝑏 𝑆𝑆 draws from the posterior are then used to calculate the required statistics.

In both the Mixed logit and Bayesian model, we compared two model estimates, in the first, the price of DTMVs is fixed, while the coefficient of attributes and other alternative specific constants were allowed to vary as independently normally distributed. In this specification, the choice variation is for different attributes is captured and across individuals in the sample. Also, the standard errors reflect the panel nature of the data (Lancsar, Fiebig, and Hole 2017) as each respondent completes 10 choice cards. The unconditional choice probabilities are estimated by simulated Maximum Likelihood.

Estimated parameters from the mixed logit model and Bayesian approach can be used to obtain willingness to pay (WTP) measures. The WTP is calculated as the change in price or premium in order to keep the same level of utility after an attribute changes. The WTP for the nth attribute is calculated as:

= (8) 2𝛽𝛽𝑛𝑛 𝑛𝑛 𝑝𝑝 Where is the estimated𝑊𝑊𝑊𝑊𝑊𝑊 parameter− 𝛽𝛽 of the attribute and is the estimated coefficient of seed price in the context of this study. Following previous studies Shee et al. 2020, the WTP is 𝛽𝛽𝑛𝑛 𝑛𝑛𝑛𝑛ℎ 𝛽𝛽𝑝𝑝 harmlessly multiplied by 2 due to the use of effects coding. Estimated coefficients of WTP in preference space represents individual farm households’ preferences or marginal utilities for various attributes of DTMVs. In the preference space, while the coefficients of other attributes are allowed to vary normally, the price coefficient is specified to be fixed across all observations. The implication of this is that attributes distribution in the WTP model will be the same as the distribution of random coefficient, at the same time, the mean and variance is scaled by the fixed price coefficient to provide a meaningful interpretation (Revelt and Train, 1998; Ward et al., 2014). The limitation of this approach is that it not realistic since the premium effect is not likely to be fixed across all attributes

In estimating WTP in price-space, also known as willingness to pay space, the model is re- parametised in a way that estimated coefficients directly represent the tradeoffs individual are willing to make and the tradeoffs are estimated in terms of the premium of the price individuals are willing to pay (Hole and Kolstad 2012; Ortega et al. 2016). The advantage of this approach over the preference state is that the estimation approach facilitates direct control of the marginal rate of substitution, unlike the preference state which relies on the ratio of two marginal utility with potentially undefined properties which may lead to errors in results (Ortega et al. 2016). The estimation in the WTP space has been found to give more realistic estimates of the WTP (Scarpa, Thiene, and Marangon 2008; Hole and Kolstad 2012; Teferi et al. 2020). To model this, we re-specify the utility individual derives from choosing during choice task is specified as a function of individual taste parameter and individual 𝑖𝑖 𝑡𝑡 specific characteristics 𝑖𝑖𝑖𝑖𝑖𝑖 𝑛𝑛 ′ 𝑥𝑥 𝑧𝑧=𝑖𝑖 + + (8) ′ Where and are𝑉𝑉𝑡𝑡𝑡𝑡𝑡𝑡 individual𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖𝛽𝛽𝑖𝑖 -specific𝑧𝑧𝑖𝑖𝑖𝑖𝑖𝑖𝛿𝛿𝑖𝑖 coefficients𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖 for attributes and individual-specific characteristics. is assumed to be an extreme value distributed with variance given by 𝛽𝛽𝑖𝑖 𝛿𝛿𝑖𝑖 ( ), where 𝑖𝑖𝑖𝑖𝑖𝑖 is an individual-specific scale parameter. As shown by Train and Weeks 2 𝜀𝜀 Π (2005),2 dividing equation 8 by does not affect behavior and result in a new error term which 𝜇𝜇𝑖𝑖 6 𝜇𝜇𝑖𝑖 is IID extreme value distributed𝑖𝑖 with variance equal to : 𝜇𝜇 2 Π = + + 6 (9) ′ Where = / and𝑉𝑉𝑖𝑖𝑖𝑖𝑖𝑖 = 𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖/ 𝜆𝜆 𝑖𝑖 𝑧𝑧𝑖𝑖𝑖𝑖𝑖𝑖𝑐𝑐𝑖𝑖 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖

Where 𝜆𝜆𝑖𝑖 𝛽𝛽is𝑖𝑖 the𝜇𝜇𝑖𝑖 utility𝑐𝑐𝑖𝑖 associated𝛿𝛿𝑖𝑖 𝜇𝜇𝑖𝑖 with individual , is a vector of the attributes for the nth alternative, is a vector of individual taste parameter s mapping these attributes into 𝑉𝑉𝑡𝑡𝑡𝑡𝑡𝑡 𝑖𝑖 𝑥𝑥𝑡𝑡𝑡𝑡𝑡𝑡 utility, is a vector of terms defining individual-specific characteristics and maps these 𝛽𝛽𝑖𝑖 characteristics into utility associated with the choice of a particular alternative. 𝑧𝑧𝑖𝑖 𝛿𝛿 the WTP for an attribute is noted as the ratio of the attribute coefficient to the price coefficient i.e. = = 𝛽𝛽𝑖𝑖 = . The utility function in WTP space (Scarpa et al. 2008; Train and Weeks, 𝜆𝜆𝑖𝑖 𝜇𝜇𝑖𝑖 𝛽𝛽𝑖𝑖 𝛿𝛿 𝑖𝑖 𝑐𝑐𝑖𝑖 𝑖𝑖 𝛿𝛿𝑖𝑖 2005)𝑤𝑤 can thus 𝜇𝜇be𝑖𝑖 written as: = + ( ) + + (10) ′ 𝑉𝑉𝑡𝑡𝑡𝑡𝑡𝑡 −𝑐𝑐𝑖𝑖𝑧𝑧𝑖𝑖𝑖𝑖𝑖𝑖 𝑐𝑐𝑖𝑖 𝑤𝑤𝑖𝑖 𝑥𝑥𝑖𝑖𝑖𝑖𝑖𝑖 𝜀𝜀𝑖𝑖𝑖𝑖𝑖𝑖

3.2 Data and Summary Statistics a. Study area and data collection

This study employs a multistage sampling procedure. The first stage includes randomly selecting a participating State out of the 18 States that participated in the DTMA project through the IITA. A substage to this was the consideration of States in the Savanna Zones that are prone to drought and have had episodes of drought occurrences. Among these States, Kano State was randomly selected. Kano State is located in the North-West region of Nigeria in the Sudan Savanna agroecological zone (figure 2) and is one of the largest agricultural producing states. Due to the topographical location of Kano state in the Sudan Savanna Zone of Nigeria, the State is prone to drought effects. (Achugbu and Anugwo 2016). Kano State is major maize producing States in Nigeria and benefitted from the DTMA project through the IITA. For this study, Gwarzo and Rano LGAs were one of the benefiting LGAs and were selected as part of the fulfilment of the objective of this study. However, not all communities in both LGAs were target implementation sites for the DTMA project. As such, the next stage of sample selection randomly selected 8 communities in each LGAs. Within which 4 were DTMA project intervention areas and non-intervention areas. In total, sample selection was drawn from 8 intervention and 8 non-intervention areas (Figure 3), totalling 16 communities. From each community, 20 maize-producing households were randomly selected using existing maize database listings. The sample size consists of 320 maize farm households in total. This sample is heterogenous and relatively represents maize producing households in drought prevalent communities in Kano State, Nigeria.

In the data collection stage, asides presentation of choice cards, a short survey was presented to households covering their respective socio-economic characteristics. Table 2. below presents the households’ attributes and the statistical differences based on the adoption of DTMVs. The data show that the average household size is approximately 14.5 members on average and the average age of the household head is approximately 43 years. Moreover, while respondents’ average years of education is 5.4 years, the maximum years of education on the average in the household is 9.56 years, this can reflect possible literacy effect on farm households attributes decisions. The result further shows a significant difference in education variables between adopters and non-adopters of DTMVs. Small-holder farmers are located on an average of 33.44 minutes away from the nearest seed market. Respondents in the study area on average are highly experienced in maize production (15.52 years), however only 9.12 years of experience in the adoption of improved maize varieties. In terms of institutional variables, 88% received one form of a loan or the other from local private agricultural firms, 34% belonged to credit institution groups and 61% are members of agricultural groups. Access to extension information shows that only 42% of farm households on average visited extension agents in the past agricultural seasons, while only 52% were visited by extension agents. The result further revealed that on average, the total land area allocated to maize production is 5.32 acres, out of which maize farm households allocated 1.8 acres to the production of DTMVs. The data further show a significant difference between adopters and non-adopters for some variables. For instance, adopters have higher total livestock units than non-adopters and 76% of adopters have ease availability of DTMVs in their community compared to 18% of non-adopters. Similarly, adopters are significantly (p<0.01) more literate with 6.16 years of education compared to non-adopters with 4.58 years of education on average. In addition, adopters of DTMVs have more years of farming experience, access to extension information, and access to agricultural groups platforms compared to non- adopters.

Figure 2: Geographical location of the study area

Figure 3: Illustrations of LGAs and communities from which samples were selected.

Kano State

Gwarzo LGA Rano LGA

A. Intervention Communities A. Intervention Communities - Katsinawa - Garebi

- Fada - Gidan Zangi - Ungwar Rajab - Doka - Katsira - Yelwa

B. Non-Intervention Communities B. Non-Intervention Communities

- Salihawa - War - Kanyar Gurgu - Zambur - Dakwara - Karofin Talan

- Balodawa - Zambur Dutse

Table 2: Summary statistics of farm households’ socio-economic characteristics

Maize farm households’ characteristics Adopters Non-adopters of Total of DTMVs DTMVs Gender (1=male, 0 otherwise) 0.95 0.97 0.96 Age (years) 43.61 42.77 43.21 Education (years) 6.16** 4.58** 5.4 Maximum education (years) 10.13* 8.95* 9.56 Household size 13.69** 15.29** 14.46 Farm experience (years) 25.43* 23.01* 24.26 Experience with improved maize 9.77 8.42 9.12 varieties (years) Maize farming experience (years) 15.79 15.22 15.52 Maize land (acre) 5.16 5.48 5.32 Land own(acres) 8.76 9.77 9.24 Number DTMVs 2.25*** 0.33*** 1.32 DTMVs availability (1=yes; 0 0.76*** 0.18*** 0.48 otherwise) Distance to seed market (minutes) 29.76*** 37.52*** 33.44 TLU 3.19** 2.40** 2.81 Soil water (1=yes; 0 otherwise) 0.94 0.91 0.93 Received loan (1=yes; 0 otherwise) 0.84** 0.92** 0.88 Credit group (1=yes; 0 otherwise) 0.40** 0.29** 0.34 Member of agric group (1=yes; 0 0.71*** 0.51*** 0.61 otherwise) Receive temperature information 0.84** 0.92** 0.88 (1=yes; 0 otherwise) Farmer visit extension agent (1=yes; 0 0.57*** 0.25*** 0.42 otherwise) Extension visit farmer (1=yes; 0 0.71*** 0.31*** 0.52 otherwise) *. **, *** represents significant level at 10%, 5% and 1% respectively

4.0 Results and Discussions a. Mixed logit estimation in preference space with and without interactions

This section presents the empirical results from the maximum simulated mixed logit model in preference space without interactions (M1) and with the incorporation of interactions (M2). The significance (p<0.01) of the log-likelihood in the two models indicate an acceptable model fit. For the mixed logit estimation in M1 and M2, significant coefficients for attributes exhibit a similar sign. In both models, the coefficient estimates reveal that farmers have the lowest preference for non-resistance to Striga attribute. This, however, relates to the existing prevalence of Striga infestations and it is one of the pre-existing issues maize farm households face during production seasons (Badu-Apraku et al. 2018). In terms of nitrogen use efficiency, maize farmers have higher preferences for the attribute of high nitrogen use efficiency attributes and a significantly lower preference for the attribute of low nitrogen use efficiency. Drought tolerance is supposedly a key attribute in drought-prone agricultural zones and expectedly maize farmers in the study areas significantly (p < 0.01) prefer DTMVs with high tolerance to drought, they as well have a high negative preference for DTMVs with low tolerance to drought attributes. The mean coefficient in both models is the highest and shows that it is the most preferred attribute. In a similar case study, Kassie et al. 2017 found drought- tolerant attribute as one of the most preferred traits among maize farm households’ in Zimbabwe. Within the contexts of this study, preference for very high yield attributes (9t/ha) is the second-highest after drought attributes. The result further reveals a significant and increasing preference for yield attributes as the yield level increases. Comparing drought and yield attributes, this study confirms that, although productivity may be of concern to farmers as it is the main direct indicator of utility maximization, farmers are as well concerned about the suitability of variety to the pre-existing condition of the environment such as drought and Striga infestations as earlier mentioned. We further found a positive but insignificant preference for large cob-size, large grain size, early and late duration attributes. In both models, the mean coefficient of fixed price attribute however shows a significantly decreasing preference as the price attribute increases. This further suggests that price is considered a potential turn-off among the attributes.

Maize farmers’ heterogenous effects on attribute preferences are further shown in M2. We however present only significant attributes in Table 3. below. The result shows significant preference at p<0.01 for yield attributes and resistance to Striga attributes differ between communities in Gwarzo and Rano local government areas. While farm households in Gwarzo local government areas have a high preference for very high yielding attributes (9t/ha), farm households in Rano local government area have less preference for maize varieties that are not resistant to Striga. Besides, current adopters of DTMVs compared to non-adopters have less preference for very high yielding attributes of maize variety, which signifies that non- adopters are attracted by the high yielding attributes and this can be a catch if well promoted to increase adoption decisions. Equally, compared to non-adopters, early maturing attributes is more preferred by adopters of DTMVs.

Table 3. Estimates of Mixed logit model in preference space

Preference space Preference space with without interaction interactions (M2) (M1) Coef. Std. Err. Coef. Std. Err. Mean Coefficient Price 0.002*** 0.001 0.002*** 0.001 Maturity/duration(early) 0.142 0.088 -0.292 0.443 Maturity/duration(late) 0.026 0.081 -0.043 0.114 Resistant to Striga (no) -1.605*** 0.102 -2.182*** 0.495 Nitrogen use efficiency (High) 0.993*** 0.092 1.198*** 0.480 Nitrogen use efficiency (Low) -0.868*** 0.134 -0.956*** 0.192 Cob size (large) 0.069 0.087 0.044 0.126 Cob size medium -0.072 0.089 -0.193 0.129 Grain size (large) 0.089 0.072 0.090 0.106 Tolerant to drought (high) 1.269*** 0.090 1.858*** 0.473 Tolerant to drought (low) -1.027*** 0.130 -1.185*** 0.181 Yield (medium, 6.0t/ha) 0.316*** 0.112 0.243 0.161 Yield (8.5t/ha, high) 0.888*** 0.107 0.934* 0.549 Yield (9t/ha, very high) 1.086 *** 0.116 1.916*** 0.562 Gwarzo lga * Yield (9t/ha, very 0.367* 0.192 high) Gwarzo * Resistant to Striga (no) 0.811*** 0.180 Adopt DTMVs * Yield (9t/ha, -0.524** 0.226 very high) Adopt DTMVs * Duration 0.341** 0.173 (early)

Standard Deviation Price Maturity/duration(early) 0.550*** 0.134 0.500*** 0.144 Maturity/duration(late) -0.322** 0.165 0.308 0.228 Resistant to Striga (no) 0.871*** 0.121 0.767*** 0.125 High nitrogen use efficiency 0.799*** 0.119 0.779*** 0.120 Low nitrogen use efficiency 0.627*** 0.226 0.683*** 0.215 Cob-size large 0.415*** 0.165 0.411** 0.169 Co-bsize medium 0.110 0.250 0.153 0.217 Grain size (large) 0.094 0.196 0.072 0.223 Tolerant to drought (high) 0.546*** 0.128 0.536*** 0.129 Tolerant to drought (low) 0.799*** 0.155 0.786*** 0.173 Yield (medium, 6.0t/ha) -0.269 0.243 -0.195 0.365 Yield (8.5t/ha, high) -0.375** 0.191 -0.292 0.229 Yield (9t/ha, very high) 0.696*** 0.161 0.621*** 0.175 Mean price SD price Number of observations 12800 12800 12800 12800 Number of choices 3200 3200 3200 3200 Number of Halton draw 500 500 500 500 Simulated log-likelihood -2792.700 -2762.2744 Wald chi-square (14) LR chi(13) 97.42*** 82.96*** Prob > chi2 0.000 0.000 *. **, *** represents significant level at 10%, 5% and 1% respectively b. Willingness to pay for maize attributes

The WTP estimates presented in table 4 below were derived from the maximum simulated likelihood estimates of the mixed logit model with and without interaction terms at 95% confidence intervals. Both models show similar signs except in attributes of early maturity. In both models, the result shows that sampled farm households’ are willing to respectively pay 1028.12 & 1406.22 more if the variety is not Striga resistant. Surprisingly, a similar pattern exists for drought-tolerant attributes as results show that farmers are willing to pay more for variety that is low tolerant and less for variety that is highly tolerant to drought. Equally, maize varieties attribute such as yield, high nitrogen use efficiency, and large cob-size and large grain size exhibit similar lower willingness to pay estimates.

Overall, in both models, the WTP result does not adequately project farmers’ rationality to optimally maximise utility. Also, as discussed earlier, the coefficient estimates in this context are modelled over a fixed price which does not truly reflect the heterogeneity in farmers’ perceived price for each attribute. The estimation of mixed logit in WTP space presented in Table 5 adequately fill in for this shortcoming. In the table, M3 and M4 respectively represent models of WTP space with and with interactions. The results are different from Table 4 and on the contrary show that in M3 and M4 respectively, among all the attributes considered, farmers value most seed varieties that are highly tolerant to drought and very high yielding (9.0t/ha). In the WTP space model without interaction, the attribute of high drought tolerance is 1.11 times more than the farmer places for every increase in very yield attribute (9.0t/ha). Similarly, the drought tolerance attribute value is 1.29 times more than the value farmers place for every time they decide to consider a high nitrogen use efficiency attribute. While the drought tolerance attribute is highly relevant in the M3, the very high yielding attribute is of the highest value in the M4 among other attributes. The drought tolerance attribute in this case is the second most preferred farmers are willing to pay more value on with farmers willing to offer 1.04times less compared to the high yield attribute. In both models, willingness to pay estimates increases as the yield level increases, suggesting that high value is placed on high yielding varieties, for example, the WTP for very high yield (9.0t/ha) attributes is about 3.2times more than medium yield attribute (6.0t/ha). Disutility for an attribute can be seen in negative coefficients and represents farmers' lower preference and value for such attribute. For example, significant estimates of nitrogen use efficiency attributes show that farmers attach a positive and higher value to the variety that has high nitrogen use efficiency but will however pay less for variety that is low in nitrogen use efficiency. Equally, preferences for varieties that are not resistant to Striga are highly negative as farmers are willing to pay 836.656 (in M3) and 1171.748 (in M4) less for non-resistant to Striga varieties. The result shows that the implicit price on non-resistant to Striga attribute is highly negative in both models and the value is 1.56 times and 1.81times respectively lower than low drought tolerant attributes in M3 and M4. The implicit price place on the premium is negatively significant and shows that farmers are willing to pay 6.261 (in M3) and 6.219 (in M4) respectively.

The interaction estimates in M4 show that, compared to farmers in Rano Lga, farmers in Gwarzo Lga are willing to pay more for variety that is not resistant to Striga suggesting that Striga tolerant attribute is less attractive to farm households in Gwarzo Lga. Also, while adopters of DTMVs are willing to pay less for very high yielding variety, they are willing to pay more for variety with early maturing attributes. Also, compared to non-adopters, adopters are surprisingly willing to pay more for low drought-tolerant varieties. Overall, of all the interactions used in this model, adopters are significantly mostly attracted to the early maturing attributes of seed varieties. This may be related to the ability to plant more cycles of maize in an agricultural year if the variety is early maturing. The overall mean price in the WTP space in both models (M3 and M4) significantly (p <0.01) decreased by 0.002 with a standard deviation of 0.001.

Table 4. Willingness to Pay estimate from the Mixed logit with interaction

WTP estimates from M1 WTP estimates from M2 WTP LL UL WTP LL UL Duration (early) -91.01 -212.22 30.20 188.18 -390.53 766.89 Duration (late) -16.61 -118.19 84.97 27.80 -118.88 174.48 Resistant to Striga (no) 1028.18 319.98 1736.38 1406.22 265.76 2546.68 Nitrogen use efficiency (high) -636.38 -1075.02 -197.75 -772.33 -1573.12 28.47 Nitrogen use efficiency (Low) 556.18 148.17 964.19 616.23 135.71 1096.76 Cob-size (large) -44.21 -158.00 69.58 -28.31 -189.05 132.42 Cob-size medium 45.98 -62.54 154.50 124.41 -44.81 293.64 Grain size (large) -57.11 -148.60 34.38 -57.94 -191.28 75.41 Tolerant to drought (high) -812.92 -1350.50 -275.33 -1197.44 -2188.04 -206.85 Tolerant to drought (low) 658.05 174.80 1141.30 763.77 186.08 1341.46 Yield (medium, 6.0t/ha) -202.43 -424.43 19.57 -156.72 -404.69 91.26 Yield (8.5t/ha, high) -569.03 -983.94 -154.12 -601.83 -1410.47 206.81 Yield (9t/ha, very high) -695.61 -1212.13 -179.09 -1234.59 -2346.54 -122.64 Notes:

Table 5. Estimates of Mixed logit model in WTP space

WTP space without WTP space with interactions interactions (M3) (M4) Coef. Std. Err. Coef. Std. Err. Mean Coefficient Price -6.261*** 0.293 -6.219*** 0.282 Maturity/duration(early) 63.005 47.254 -143.616 229.375 Maturity/duration(late) -8.266 43.021 -63.995 63.938 Resistant to Striga (no) -836.656*** 253.775 -1171.748*** 422.036 Nitrogen use efficiency (High) 513.682*** 156.330 603.818** 301.684 Nitrogen use efficiency (Low) -453.334*** 152.555 -520.149*** 197.297 Cob size (large) 41.182 48.113 40.051 66.090 Cob size medium -38.906 47.113 -97.879 68.790 Grain size (large) 58.021 38.183 64.479 54.020 Tolerant to drought (high) 663.706*** 192.208 974.704** 380.964 Tolerant to drought (low) -535.904*** 180.823 -645.540*** 210.246 Yield (medium, 6.0t/ha) 188.966*** 92.205 159.682 103.396 Yield (8.5t/ha, high) 510.270*** 162.772 526.827* 308.026 Yield (9t/ha, very high) 598.613*** 198.472 1011.023** 429.659 Gwarzo * Resistant to Striga (no) 443.423*** 157.236 Adopt * Yield (9t/ha, very high) -281.503* 146.672 Adopt *Tolerant to drought 225.288* 133.094 (low) Adopt *Maturity/duration(early) 173.565* 100.072

Standard Deviation Price 0.436*** 0.061 0.447*** 0.061 Early maturing -281.812*** 103.774 -262.625*** 99.618 Late-maturing -180.361* 101.008 -221.986** 98.217 Resistant to Striga (no) -380.226*** 134.952 309.065*** 108.383 High nitrogen use efficiency -320.601*** 120.427 310.347*** 107.887 Low nitrogen use efficiency -208.899 230.055 -246.616 364.544 Cobsize large 218.579** 99.982 226.652** 91.626 Cobsize medium 76.607 117.359 69.368 143.835 Grain size (large) -41.506 115.202 24.097 122.827 Tolerant to drought (high) -190.812* 104.746 -163.789 101.186 Tolerant to drought (low) 261.334 207.444 291.690 184.805 Yield (medium, 6.0t/ha) -144.995 130.020 -130.691 102.673 Yield (8.5t/ha, high) -57.476 152.666 -49.787 149.914 Yield (9t/ha, very high) -349.867*** 129.681 -313.453*** 122.034 Mean price -0.002*** 0.001 -0.002*** 0.001 SD price 0.001*** 0.000 0.001*** 0.003 Number of observations 12800 12800 12800 12800 Number of choices 3200 3200 3200 3200 Number of Halton draw 500 500 500 500 Simulated log-likelihood -2783.296 -2748.548 Wald chi-square (14) 13367.37*** 12123.90*** Prob > chi2 0.000 0.000 *. **, *** represents significant level at 10%, 5% and 1% respectively c. The Hierarchical Bayesian Estimates in preference and WTP space

Farm households’ attributes ranking in the bayesian model in preference state (M5) is different from the mixed logit model in preference state (M1) and compared to M1, all attributes are significant at p < 0.01. The result shows that maize farmers have the highest positive preference for variety with large grain size and the lowest preference for low tolerance to drought attribute. The high nitrogen use efficiency attribute is the second most preferred among farmers followed by the high tolerance to drought attribute. Besides, the preference for medium cob size is higher compared to large cob-size and preferences as positively significant for all yield levels, however highest for high yield attribute. Also, while preferences for early maturing variety is positive and significant, it decreases by 205.723 for late-maturing attributes. Other less preferred attributes include non-Striga resistance, low tolerance to drought, and low nitrogen use efficiency attribute.

Coefficient estimates of attributes in the Hierarchical Bayesian in WTP space (M6) are almost similar to the mixed logit in WTP space (M3) in terms of the sign direction with an exception in the significance of duration/maturity attribute. The result shows that the WTP estimates in is highest for the high drought tolerant attribute, in which farmers are willing to pay 1.079 more. This value is 1.35% more than what they are willing to pay for high nitrogen use efficiency. Similarly, maize farmers value high drought-tolerant traits 1.13 times than the value they place on high yield attributes (9.0t/ha). Also, the farmer placed more value on the high drought tolerance trait 7.39 times than the value they attach to early maturing attributes. In this context, farmers place a higher value on high drought tolerant attributes over other attributes, this result is in line with Kassie et al. 2017 similar higher implicit price for drought tolerance traits compared to other attributes. Also, willingness to pay increases as the yield level increases, and as seen in Table 4, farmers are willing to pay 3.41 times for very high yield attribute compared to medium yield attribute. This result is expectedly rational and shows the relevance of yield attributes among maize farm households. In the same vein, farm households’ place 1.73 times more value on large cob-size compared to early maturing attributes. Also, non-resistant to Striga attribute famers’ are willing to pay 1.496 times less for no resistance to Striga attributes, which is 1.74 times and 2.5times less value compared to low drought tolerant attribute and low nitrogen use efficiency respectively. However, farmers attached the lowest value to the price attribute suggesting their willingness to offer 6.803 less on average. Comparing this to the nearest lowest attribute suggests that maize farmers are willing to offer 4.55 times lesser premium compared to what they will offer to pay for non- resistant to Striga attribute.

Table 6. Estimates of Hierarchical Bayesian model in preference and WTP space

Preference (M4) WTP space (M5) Mean Coefficient Coef. Std. Err. Coef. Std. Err. Price 0.361*** 0.000 -6.803*** 0.343 Early-maturing 177.866*** 41.825 0.146* 0.081 Late-maturing -204.723*** 27.239 0.081 0.112 Resistant to Striga (no) -388.282*** 33.639 -1.496*** 0.086 Nitrogen use efficiency(High) 389.044*** 33.566 0.795*** 0.108 Nitrogen use efficiency(Low) -468.985*** 23.139 -0.597*** 0.112 Cob size large 238.251*** 43.924 0.253** 0.111 Cob size medium 261.480*** 36.981 -0.120 0.083 Grain size (large) 461.485*** 62.944 -0.010 0.095 Tolerant to drought (high) 337.890*** 31.236 1.079*** 0.077 Tolerant to drought (low) -494.481*** 48.479 -0.850*** 0.098 Yield (medium, 6.0t/ha) 164.245*** 33.049 0.281*** 0.096 Yield (8.5t/ha, high) 302.065*** 31.565 0.757*** 0.137 Yield (9t/ha, very high) 291.039*** 33.785 0.959*** 0.086 Number of observations 12800 12800 12800 12800 Number of Choices 3200 3200 3200 3200 Number of groups 320 320 320 320 Total draws 4000 4000 4000 4000 Burn-in-draws 1000 1000 1000 1000 *. **, *** represents significant level at 10%, 5% and 1% respectively

5.0 Summary and Conclusion

Increasing adoption of agricultural technology in sub-Saharan Africa remains a challenge in several contexts and continuous research on achieving rapid adoption is of great concern and it is cogent in the adoption of climate mitigation schemes. Drought remains one of the leading drawbacks to increasing maize productivity in Nigeria and has negatively impacted the productivity and welfare of the rural agricultural populace. Similar to most agricultural technology in SSA, low adoption plagues adoption decisions of DTMVs. Thus, this study fills this gap by employing a DCE approach to elicit preference information on existing DTMVs and the implicit price they are willing to pay for the attributes. We compared modelling in preference state and WTP space in two-state; with and without household interactions using the simulated log-likelihood mixed logit model and the Hierarchical Bayesian estimates. From most of the model estimates (M1 to M5), we found the drought-tolerant attribute to be the highly preferred and the one farmers’ are willing to place more value on. We also found significant preference and willingness to pay more for very high yielding attributes in the M6. Our result recommends the need to take into account maize farmers' trait preferences and adequate dissemination to encourage adoption.

All models also show significant turnoff for the price attribute and farmers' willingness to offer less price. Policy intervention may consider the need to subsidise drought-tolerant maize varieties to enable ease accessibilities for farmers' use. Incorporating other highly significantly attribute as expressed in the high disutility for non-resistance to Striga attributes and low nitrogen use efficiency is likely to affect adoption decisions. Furthermore, there is a need to consider heterogeneity in preferences as our result further revealed differences in preference in local government areas and among adopters and non-adopters. For example, there may be a need to promote very high yielding maize varieties in Gwarzo as it is more attractive to farmers resident in this local government area, and more of Striga resistance attributes are preferred in Rano local government areas. Also, the result further shows non-adopters preferences for high tolerant drought attributes and adopters' significant preferences for early maturing varieties. Promotion and dissemination through marketing campaigns should be tailored towards incorporating different preferences in various interest groups.

References

Abate T, Menkir A, MacRobert JF, Tesfahun G, Abdoulaye T, Setimela P, Badu-Apraku B, Makumbi D, Magorokosho C, T.A. (eds. . 2013. “DTMA Highlights for 2012 / 13 Drought Tolerant Maize for Africa.” DTMA Highlights for 2012/13. CIMMYT-Kenya (April):116.

Abdoulaye, T., T. Wossen, and B. Awotide. 2018. “Impacts of improved maize varieties in Nigeria: ex-post assessment of productivity and welfare outcomes.” Food Security (2018) 10:369–379.

Abebe, G.K., J. Bijman, S. Pascucci, and O. Omta. 2013. “Adoption of improved potato varieties in Ethiopia: The role of agricultural knowledge and innovation system and smallholder farmers’ quality assessment.” Agricultural Systems 122 (2013) 22–32.

Abubakar, I., and M. Yamusa. 2013. “Recurrence of Drought in Nigeria: Causes, Effects and Mitigation.” International Journal of Agriculture and Food Science Technology 4(3):169–180.

Achugbu, I.C., and S. Anugwo. 2016. “Drought Trend Analysis in Kano Using Standardized Precipitation Index (SPI).” FUOYE Journal of Engineering and Technology 1(1).

Adewopo, J. 2019. “Smallholder maize-based system: a piece of the puzzle for sustaining food security in Nigeria.” Multifunctional land uses in Africa: sustainable food security solutions.:115–133.

African Centre for Biodiversity (ACB). 2017. “The Water Efficient Maize for Africa Project: Profiteering not Philanthropy.” :59. Available at: https://acbio.org.za/en/water-efficient- maize-africa-project-profiteering-not-philanthropy.

Alemu, M.H., and S.B. Olsen. 2018. “Can a Repeated Opt-Out Reminder mitigate hypothetical bias in discrete choice experiments? An application to consumer valuation of novel food products.” European Review of Agricultural Economics 45(5):749–782.

Asfaw, S., B. Shiferaw, F. Simtowe, and L. Lipper. 2012. “Impact of modern agricultural technologies on smallholder welfare: Evidence from Tanzania and Ethiopia.” Food Policy 37(3):283–295.

Awotide, B.A., A.D. Alene, T. Abdoulaye, and V.M. Manyong. 2015. “Impact of agricultural technology adoption on asset ownership: the case of improved cassava varieties in Nigeria.” Food Security. (2015) 7:1239–1258.

Azeez, J.O., D. Chikoye, A.Y. Kamara, A. Menkir, and M.T. Adetunji. 2005. “Effect of drought and weed management on maize genotypes and the tensiometric soil water content of an eutric nitisol in south western Nigeria.” Plant and Soil 276(1–2):61–68.

Badu-Apraku, B., M.A.B. Fakorede, B. Annor, and A.O. Talabi. 2018. “ Improvement in grain yield and low nitrogen tolerance in maize cultivars of three eras” Experimental Agriculture 54(6):805–823.

Badu-Apraku, B., A.O. Talabi, M.A.B. Fakorede, Y. Fasanmade, M. Gedil, C. Magorokosho, and R. Asiedu. 2019. “Yield gains and associated changes in an early yellow bi-parental maize population following genomic selection for Striga resistance and drought tolerance.” BMC Plant Biology 19(1):1–18.

Bänziger, M., and H.R. Lafitte. 1997. “Efficiency of secondary traits for improving maize for low-nitrogen target environments.” Crop Science 37(4):1110–1117.

Bello, O.B., O.J. Olawuyi, S.Y. Abdulmaliq, S.A. Ige, J. Nahamood, M.A. Azeez, and M.S. Afolabi. 2014. “Yield performance and adaptation of early and intermediate drought- tolerant maize genotypes in Guinea Savanna of Nigeria.” Sarhad Journal of agriculture 30(1):53–66.

Bezu, S., G.T. Kassie, B. Shiferaw, and J. Ricker-Gilbert. 2014. “Impact of improved maize adoption on welfare of farm households in Malawi: A panel data analysis.” World Development Vol. 59, pp. 120–131, 2014.

Boxall, P., W.L. Adamowicz, and A. Moon. 2009. “Complexity in choice experiments: Choice of the status quo alternative and implications for welfare measurement.” Australian Journal of Agricultural and Resource Economics 53(4):503–519.

Brief, E. 2020. “Nigeria.” (October).

Buah, S., J. Kombiok, R. Kanton, N. Denwar, A. Haruna, A. Wiredu, and M. Abdulai. 2013. “Participatory evaluation of drought tolerant maize varieties in the Guinea Savanna of Ghana using mother and baby trial design.” Journal of Science and Technology (Ghana) 33(2):12.

Campbell, D., and S. Erdem. 2019. “Including Opt-Out Options in Discrete Choice Experiments: Issues to Consider.” Patient 12(1):1–14.

Carlsson, F., and P. Martinsson. 2003. “Design techniques for stated preference methods in health economics.” Health Economics 12(4):281–294.

Chen, Q. 2021. “District or distributed space heating in rural residential sector? Empirical evidence from a discrete choice experiment in South China.”

Dao, A., J. Sanou, V. Gracen, and E.Y. Danquah. 2015. “Indentifying farmers’ preferences and constraints to maize production in two agro-ecological zones in Burkina Faso.” Agriculture and Food Security 4(1):1–7.

Dugje, I.Y., A.Y. Kamara, and L.O. Omoigui. 2006. “Infestation of crop fields by Striga species in the savanna zones of northeast Nigeria.” Agriculture, Ecosystems and Environment 116(3–4):251–254.

Ekeleme, F., J.M. Jibrin, A.Y. Kamara, M. Oluoch, A.M. Samndi, and A.A. Fagge. 2014. “Assessment of the relationship between soil properties, Striga hermonthica infestation and the on-farm yields of maize in the dry Savannas of Nigeria.” Crop Protection 66:90– 97.

Ekpa, O., N. Palacios-Rojas, G. Kruseman, V. Fogliano, and A.R. Linnemann. 2018. “Sub- Saharan African maize-based foods: Technological perspectives to increase the food and nutrition security impacts of maize breeding programmes.” Global Food Security 17(January):48–56.

Evenson, R.E., and D. Gollin. 2003. “Assessing the impact of the Green Revolution, 1960 to 2000.” Science 300(5620):758–762.

Eze, J.N. 2018. “Drought occurrences and its implications on the households in Yobe state, Nigeria.” Geoenvironmental Disasters 5(18).

Fisher, M., T. Abate, R.W. Lunduka, W. Asnake, Y. Alemayehu, and R.B. Madulu. 2015. “Drought tolerant maize for farmer adaptation to drought in sub-Saharan Africa: Determinants of adoption in eastern and southern Africa.” Climatic Change 133(2):283– 299.

Fisher, M., and E.R. Carr. 2015. “The influence of gendered roles and responsibilities on the adoption of technologies that mitigate drought risk: The case of drought-tolerant maize seed in eastern Uganda.” Global Environmental Change 35:82–92.

FAOSTAT 2019. Food and Agricultural Oraganisation Statistics 2019 http://www.fao.org/faostat/en/#data

Genti, K., A. Tahirou, E. Olaf, O. Kai, G. Zhe, S. Peter, and A. Menkir. 2004. “Potential Impacts of Drought Tolerant Maize: New Evidence from Farm-trials in Eastern and Southern Africa.” 89th Annual Conference of the Agricultural Economics Society, University of Warwick, England (April):0–28.

GHI 2020. Global Hunger Index. https://www.globalhungerindex.org/ranking.html

Hassan, A.G., M.A. Fullen, and D. Oloke. 2019. “Problems of drought and its management in Yobe State, Nigeria.” Weather and Climate Extremes 23(January):100192.

Holden, S.T., and M. Fisher. 2015. “Subsidies promote use of drought tolerant maize varieties despite variable yield performance under smallholder environments in Malawi.” Food Security (2015) 7:1225–1238.

Hole, A.R., and J.R. Kolstad. 2012. “Mixed logit estimation of willingness to pay distributions: A comparison of models in preference and WTP space using data from a health-related choice experiment.” Empirical Economics 42(2):445–469.

Hu, Y., L.A. House, B.R. McFadden, and Z. Gao. 2020. “The Influence of Choice Context on Consumers’ Preference for GM Orange Juice.” Journal of Agricultural Economics (2016).

Jagger, P., and C. Jumbe. 2016. “Stoves or sugar? Willingness to adopt improved cookstoves in Malawi.” Energy Policy 92:409–419.

Joshi, P.K., M.T. Khan, and A. Kishore. 2019. “Heterogeneity in male and female farmers’ preference for a profit-enhancing and labor-saving technology: The case of Direct- Seeded Rice (DSR) in India.” Canadian Journal of Agricultural Economics 67(3):303–320.

Kagoya, S., K.P. Paudel, and N.L. Daniel. 2018. “Awareness and Adoption of Soil and Water Conservation Technologies in a Developing Country: A Case of Nabajuzi Watershed in Central Uganda.” Environmental Management (2018) 61:188–196.

Kamara, A.Y., S.U. Ewansiha, and A. Menkir. 2014. “Assessment of nitrogen uptake and utilization in drought tolerant and Striga resistant tropical maize varieties.” Archives of Agronomy and Soil Science 60(2):195–207. Kamara, A.Y., S.U. Ewansiha, and A.I. Tofa. 2019. “Yield, N Uptake and N Utilization of Early Maturing, Drought and Striga-Tolerant Maize Varieties under Low N Conditions.” Communications in Soil Science and Plant Analysis 50(4):373–387.

Kamara, A.Y., A. Menkir, S.O. Ajala, and I. Kureh. 2005. “Performance of diverse maize genotypes under nitrogen deficiency in the northern Guinea Savanna of Nigeria.” Experimental Agriculture 41(2):199–212.

Kamara, A.Y., A. Menkir, D. Chikoye, R. Solomon, A.I. Tofa, and L.O. Omoigui. 2020. “Seed dressing maize with imazapyr to control Striga hermonthica in farmers’ fields in the savannas of Nigeria.” Agriculture (Switzerland) 10(3).

Kassie, G.T., A. Abdulai, W.H. Greene, B. Shiferaw, T. Abate, A. Tarekegne, and C. Sutcliffe. 2017. “Modeling Preference and Willingness to Pay for Drought Tolerance (DT) in Maize in Rural Zimbabwe.” World Development 94:465–477.

Kassie, M., P. Marenya, Y. Tessema, M. Jaleta, D. Zeng, O. Erenstein, and D. Rahut. 2018. “Measuring Farm and Market Level Economic Impacts of Improved Maize Production Technologies in Ethiopia: Evidence from Panel Data.” Journal of Agricultural Economics 69(1):76–95.

Katengeza, S.P., S.T. Holden, and R.W. Lunduka. 2019. “Adoption of Drought Tolerant Maize Varieties under Rainfall Stress in Malawi.” Journal of Agricultural Economics 70(1):198–214.

Kjær, T. 2005. “A Review of the Discrete Choice Experiment - With Emphasis on its Application in Healthcare.” Health Economic Papers (1):1–139.

Kostandini, G., R. La Rovere, and T. Abdoulaye. 2013. “Potential impacts of increasing average yields and reducing maize yield variability in Africa.” Food Policy 43:213–226.

Krah, K., H. Michelson, E. Perge, and R. Jindal. 2019a. “Constraints to adopting soil fertility management practices in Malawi: A choice experiment approach.” World Development 124 (2019) 104651.

Krah, K., H. Michelson, E. Perge, and R. Jindal. 2019b. “Constraints to adopting soil fertility management practices in Malawi: A choice experiment approach.” World Development 124:104651.

Ladenburg, J., and S.B. Olsen. 2014. “Augmenting short Cheap Talk scripts with a repeated Opt-Out Reminder in Choice Experiment surveys.” Resource and Energy Economics 37:39–63.

Lancsar, E., D.G. Fiebig, and A.R. Hole. 2017. “Discrete Choice Experiments: A Guide to Model Specification, Estimation and Software.” PharmacoEconomics 35(7):697–716.

Liu, K., and P. Wiatrak. 2011. “Corn (Zea mays L.) plant characteristics and grain yield response to N fertilization programs in No-Tillage system.” American Journal of Agricultural and Biological Science 6(1):172–179.

Lobell, D.B., M. Bänziger, C. Magorokosho, and B. Vivek. 2011. “Nonlinear heat effects on African maize as evidenced by historical yield trials.” Nature Climate Change 1(1):42–45. Louviere, J.J., T.N. Flynn, and R.T. Carson. 2010. “Discrete choice experiments are not conjoint analysis.” Journal of Choice Modelling 3(3):57–72.

Mathenge, M.K., M. Smale, and J. Olwande. 2014. “The impacts of hybrid maize seed on the welfare of farming households in Kenya.” Food Policy 44:262–271.

Medugu, I.N., M.R. Majid, and I.D. Choji. 2008. “A comprehensive approach to drought and desertification in Nigeria: A brief evaluation of government policies.” Management of Environmental Quality: An International Journal 19(6):690–704.

Meyerhoff, J., M. Boeri, and V. Hartje. 2014. “The value of water quality improvements in the region Berlin-Brandenburg as a function of distance and state residency.” Water Resources and Economics 5(2014):49–66.

Moser, R., R. Raffaelli, and S. Notaro. 2014. “Testing hypothetical bias with a real choice experiment using respondents’ own money.” European Review of Agricultural Economics 41(1):25–46.

Nthambi, M., N. Markova-Nenova, and F. Wätzold. 2021. “Quantifying Loss of Benefits from Poor Governance of Climate Change Adaptation Projects: A Discrete Choice Experiment with Farmers in Kenya.” Ecological Economics 179(September 2020):106831.

O. B., B., A. S. Y., I. S. A., M. J., O. F., A. M. A., and A. M. S. 2012. “Evaluation of Early and Late/Intermediate Maize Varieties for Grain Yield Potential and Adaptation to a Southern Guinea Savanna Agro-ecology of Nigeria.” International Journal of Plant Research 2(2):14–21.

Olaoye, G., A. Menkir, S. Ajala, and S. Jacob. 2009. “Evaluation of local maize (Zea mays L.) varieties from Burkina Faso as source of tolerance to drought.” Journal of Applied Biosciences:887–898.

Ortega, D.L., S.J. Hong, H.H. Wang, and L. Wu. 2016. “Emerging markets for imported beef in China: Results from a consumer choice experiment in Beijing.” Meat Science 121:317– 323.

Owusu Coffie, R., M.P. Burton, F.L. Gibson, and A. Hailu. 2016. “Choice of Rice Production Practices in Ghana: A Comparison of Willingness to Pay and Preference Space Estimates.” Journal of Agricultural Economics 67(3):799–819.

Oyinbo, O., J. Chamberlin, B. Vanlauwe, L. Vranken, Y.A. Kamara, P. Craufurd, and M. Maertens. 2019. “Farmers’ preferences for high-input agriculture supported by site- specific extension services: Evidence from a choice experiment in Nigeria.” Agricultural Systems 173 (2019) 12–26.

Pasley, H.R., J.J. Camberato, J.E. Cairns, M. Zaman-Allah, B. Das, and T.J. Vyn. 2020. “Nitrogen rate impacts on tropical maize nitrogen use efficiency and soil nitrogen depletion in eastern and southern Africa.” Nutrient Cycling in Agroecosystems 116(3):397–408.

Penn, J., and W. Hu. 2020. “Reports of Bed Bugs on Hotel Selection: A Choice Experiment.” International Journal of Hospitality Management 89(July):102568.

Potoglou, D., L. Whitmarsh, C. Whittle, I. Tsouros, P. Haggar, and T. Persson. 2020. “To what extent do people value sustainable-resourced materials? A choice experiment with cars and mobile phones across six countries.” Journal of Cleaner Production 246:118957.

Regier, D.A., M. Ryan, E. Phimister, and C.A. Marra. 2009. “Bayesian and classical estimation of mixed logit: An application to genetic testing.” Journal of Health Economics 28(3):598–610.

Scarpa, Thiene, and Marangon. 2008. “Using flexible taste distributions to value collective reputation for environmentally-friendly production methods (Canadian Journal of Agricultural Economics).” Canadian Journal of Agricultural Economics 56(3):145–162.

Schaafsma, M., S. Ferrini, and R.K. Turner. 2019. “Assessing smallholder preferences for incentivised climate-smart agriculture using a discrete choice experiment.” Land Use Policy 88(August):104153.

Shee, A., C.G. Turvey, and A. Marr. 2020. “Heterogeneous Demand and Supply for an Insurance-linked Credit Product in Kenya: A Stated Choice Experiment Approach.” Journal of Agricultural Economics.

Tadesse, D., Z.G. Medhin, and A. Ayalew. 2014. “Participatory on Farm Evaluation of Improved Maize Varieties in Chilga District of North Western Ethiopia.” International Journal of Agriculture and Forestry 2014 4(5):402–407.

Teferi, E.T., G.T. Kassie, M.E. Pe, and C. Fadda. 2020. “Are farmers willing to pay for climate related traits of wheat? Evidence from rural parts of Ethiopia.” Agricultural Systems 185(September):102947. Available at: https://doi.org/10.1016/j.agsy.2020.102947.

Tegbaru, A., A. Menkir, M. Nasser Baco, L. Idrisou, D. Sissoko, A.O. Eyitayo, T. Abate, and A. Tahirou. 2020. “Addressing gendered varietal and trait preferences in West African maize.” World Development Perspectives 20(July):100268.

Train, K.E. 2012. “Bayesian Procedures.” Discrete Choice Methods with Simulation:282–314.

Usk, J.A.L.L., and T.E.D.C.S. Chroeder. 2009. “M Ethodological Aspects of C Hoice E Xperiments in Relation To.” Ethics 86(May):467–482.

Ward, P.S., A.R. Bell, K. Droppelmann, and T.G. Benton. 2018. “Early adoption of conservation agriculture practices: Understanding partial compliance in programs with multiple adoption decisions.” Land Use Policy 70(October 2017):27–37.

Ward, P.S., A.R. Bell, G.M. Parkhurst, K. Droppelmann, and L. Mapemba. 2016. “Heterogeneous preferences and the effects of incentives in promoting conservation agriculture in Malawi.” Agriculture, Ecosystems and Environment 222 (2016) 67–79.

Ward, P.S., D.L. Ortega, D.J. Spielman, and V. Singh. 2014. “Heterogeneous demand for drought-tolerant rice: Evidence from Bihar, India.” World Development 64:125–139.

Ward, P.S., and V. Singh. 2015. “Using Field Experiments to Elicit Risk and Ambiguity Preferences: Behavioural Factors and the Adoption of New Agricultural Technologies in Rural India.” Journal of Development Studies 51(6):707–724.

Wossen, T., T. Abdoulaye, A. Alene, S. Feleke, A. Menkir, and V. Manyong. 2017. “Measuring the impacts of adaptation strategies to drought stress: The case of drought tolerant maize varieties.” Journal of Environmental Management 203 (2017) 106-113.

Zilberman, D., D. Gollin, M. Morris, and D. Byerlee. 2005. “Issues Facing Agricultural Technology Adoption in Developing Countries Technology Adoption in Intensive Post- green Revolution Systems 87(5):1310–1316.