Introduction
Maize (Zea mays L.) is regarded as the most important staple cereal crop in Sub-Saharan Africa (SSA) with a huge potential for addressing the challenge of food insecurity (Abdoulaye et al., Reference Abdoulaye, Wossen and Awotide2018; Badu-Apraku et al., Reference Badu-Apraku, Yallou and Oyekunle2013). The cultivation of maize has gradually spread to the Sudan savannas of West Africa where the growing period is 90–100 days because of the availability of short-season early maturing varieties (Kamara et al., Reference Kamara, Ekeleme, Menkir, Chikoye and Omoigui2009). Despite the increased area under maize production, yields have, however, remained quite low. With the SSA population likely to double by 2050, there is need for food production to keep pace with the rising population. However, maize production is facing serious challenges from biophysical and socioeconomic limitations (Tesfaye et al., Reference Tesfaye, Gbegbelegbe, Cairns, Shiferaw, Prasanna, Sonder, Boote, Makumbi and Robertson2015). In Nigeria, there is an obvious maize yield gap as the average yield of maize was 1.6 Mg ha−1 in 2016 compared to the world average of 5.6 Mg ha−1 (FAO, 2018). The major factors limiting maize yield in the Sudan savannas of Nigeria include the inherently poor soils, frequent droughts and parasitism by Striga hermonthica, lack of proper adherence to improved agronomic practices (especially planting dates and densities) and low use of inputs such as fertilizers and improved seeds (Kamara et al., Reference Kamara, Ewansiha and Menkir2014). A key challenge gaining prominence is weather shocks (early and late season drought) and could be a major issue in the coming years due to climate change (Tesfaye et al., Reference Tesfaye, Gbegbelegbe, Cairns, Shiferaw, Prasanna, Sonder, Boote, Makumbi and Robertson2015). This calls for local adaptation measures through improved varieties well suited to the growing conditions amongst others.
In recent years, new early and extra early maturing maize varieties (short-season maize varieties) have been developed for the Sudan and Sahel agroecologies by International Institute of Tropical Agriculture (IITA) and its partners because of the short growing periods prevalent in these areas and the incidence of drought, especially in view of rapid climatic variability. The developed early maize cultivars were disseminated to farmers using Innovation Platform (IP) under the premise of Integrated Agricultural Research for Development (IAR4D) adopted by Sub-Saharan Africa Challenge Programme (SSA-CP). The SSA-CP proposed an alternative approach to address underperformance of SSA agriculture due to the traditional Agricultural and Research Development (ARD) approach, which is characterised by organisation of research and development as a linear process (Pamuk et al., Reference Pamuk, Bulte and Adekunle2014). The alternative approach known as IAR4D was developed by the Forum for Agricultural Research for Development in Africa (FARA). Its aim is to appropriately embed agricultural research within a larger system of innovation whereby knowledge from numerous sources (comprising all various actors and stakeholders) is integrated and effectively put into use (Sanyang et al., Reference Sanyang, Taonda, Kuiseu, Coulibaly and Konaté2016). The operational structure for this is an IP comprised of partnership of researchers, extension workers, farmer representatives, traditional leaders, private firms, NGOs, and government policy makers, who interact to support sustainable agricultural development.
In the face of climate variability that threatens maize production in SSA coupled with smallholders’ limited access to formal insurance, there is a need for local adaptation measures and technologies to reduce the risk of crop failure and sustain the importance of maize for food security (Tesfaye et al., Reference Tesfaye, Gbegbelegbe, Cairns, Shiferaw, Prasanna, Sonder, Boote, Makumbi and Robertson2015; Wossen et al., Reference Wossen, Abdoulaye, Alene, Feleke, Menkir and Manyong2017). With the promotion of short-season maize varieties via IPs, a proper understanding of the factors influencing the adoption of these varieties and more importantly the impacts of these varieties is necessary for research, development and policy. Literature on the adoption and impacts of agricultural technologies using different methodological approaches is enormous in most parts of SSA but with mixed results (Abdoulaye et al., Reference Abdoulaye, Wossen and Awotide2018; Alene and Manyong, Reference Alene and Manyong2006; Asfaw et al., Reference Asfaw, Shiferaw, Simtowe and Haile2011; Awotide et al., Reference Awotide, Abdoulaye, Alene and Manyong2014; Croppenstedt et al., Reference Croppenstedt, Demeke and Meschi2003; Katengeza et al., Reference Katengeza, Mangisoni, Kassie, Sutcliffe, Langyintuo, La Rovere and Mwangi2012; Khonje et al., Reference Khonje, Manda, Alene and Kassie2015; Manda et al., Reference Manda, Alene, Gardebroek, Kassie and Tembo2016; Ouma et al., Reference Ouma, Bett and Mbataru2014; Shiferaw et al., Reference Shiferaw, Kebede and You2008; Solomon et al., Reference Solomon, Tessema and Bekele2014; Weyessa, Reference Weyessa2014; Wossen et al., Reference Wossen, Abdoulaye, Alene, Feleke, Menkir and Manyong2017). Some point to significant welfare effects while others show no significant effect, which echoes the question of why adoption of supposedly profitable intensification technologies is quite low, and whether such technologies truly produce welfare benefits for smallholders. Unlike previous empirical studies on impacts of agricultural technologies, our interest is on causal effects of the improved maize varieties promoted via IPs and its implications for sustainable maize production in the face of weather shocks. We motivate this research direction because of the limited quantitative evidence on the causal effects of short-season maize varieties promoted via IPs in areas more prone to shortened length of crop growing season such as Sudan savannas.
There is a vast literature on the potentials of IPs. For example, Schut et al. (Reference Schut, Klerkx, Sartas, Lamers, Mc Campbell, Ogbonna, Kaushik, Atta-Krah and Leeuwis2016) contributed to IP literature by providing information on implementation and institutionalization of IPs in AR4D programmes. Also, Schut et al. (Reference Schut, Cadilhon, Misiko and Dror2018a) provided insights on the potentials of IPs by identifying success factors of eight case studies through a meta-analysis of mature IPs. Sanyang et al. (Reference Sanyang, Taonda, Kuiseu, Coulibaly and Konaté2016) highlighted the potentials of IPs and provided the requirements for improving the competence and skills of IPs actors in agricultural value chains, food systems and natural resource management. Schut et al. (Reference Schut, Kamanda, Gramzow, Dubois, Stoian, Andersson, Dror, Sartas, Mur, Kassam, Brouwer, Devaux, Velasco, Flor, Gummert, Buizer, Mcdougall, Davis, Tui and Lundy2018b) provided decision support for research, development and donors on the purposes and conditions under which IPs can contribute to agricultural development outcomes. Despite the acclaimed potentials of IPs, there is still scarce quantitative evidence in the literature to show the status of adoption and causal effects of short-season maize varieties promoted via IPs. Therefore, this study was undertaken to provide empirical evidence on the adoption and impacts of short-season maize varieties 6 years after their promotion via IPs in the Sudan savannas of northern Nigeria.
Materials and Methods
The study was carried out in Musawa and Shanono Local Government Areas (LGAs) of the Sudan savanna taskforce project of SSA-CP, which operates across three Pilot Learning Site (PLS) projects of which Kano–Katsina–Maradi PLS (KKM-PLS) is one of them. The Sudan Savanna taskforce is one of the three taskforces within the KKM-PLS, which aims to ensure sustainable agricultural intensification and integrated natural resource management in Sudan Savanna of West Africa through technical, administrative, marketing and management improvements of the cereal–legume–livestock systems. The Sudan Savanna taskforce established one IP in each of the LGAs. The sampling procedure involved random selection of 10 villages from Musawa and Shanono LGAs to give a total of 20 communities. Thirty maize-based farm households were randomly selected from each community to give 300 households per LGA and a total of 600 sampled households from the two LGAs. Primary data were used for the study, which were collected in 2014, that is, 6 years after the introduction of maize production technologies through the IPs. The data were collected using a structured questionnaire administered to the 600 sampled households by trained enumerators. Data on farmer, household, farm-specific characteristics, institutional characteristics and technology-specific attributes of improved maize were collected. Also, data on adoption of improved maize varieties and outcome variables were elicited from the farmers. The analytical tools employed in data analysis were descriptive statistics used to determine the adoption rate and intensity of adoption of maize varieties. A double-hurdle model was used to determine the factors that influenced the probability of adoption and intensity of adoption of the improved maize varieties. The impact of adoption of the short-season maize was evaluated using propensity score matching (PSM).
Double-hurdle model
The double-hurdle model is a parametric generalization of the Tobit model, in which two separate stochastic processes determine the decision to adopt and the level of adoption of the technology. In the Tobit model, decisions whether or not to adopt and how much to adopt are assumed to be made jointly, and hence, the factors affecting the two level decisions are taken to be the same. However, the decision to adopt may well precede the decision about the intensity of use, and hence, the explanatory variables in the two stages may differ (Asfaw et al., Reference Asfaw, Shiferaw, Simtowe and Haile2011). One of the major drawbacks of the Tobit model is the fact decisions of whether to adopt a technology or not and how much to adopt are assumed to be made jointly, and hence, the factors affecting the two decisions are assumed to be the same.
The double-hurdle model is applied in such a way that both hurdles (the decision to adopt and intensity of adoption) have equations associated with them, incorporating the effects of farmers’ characteristics (Croppenstedt et al., Reference Croppenstedt, Demeke and Meschi2003; Shiferaw et al., Reference Shiferaw, Kebede and You2008). In estimating the double-hurdle model, a Probit regression (using all observations) is followed by a truncated regression on the non-zero observations (Cragg, Reference Cragg1971). The double-hurdle model assumes that households make two sequential decisions with regard to adoption and extent of adoption. Each hurdle is conditioned by the household’s socioeconomic characteristics. In the double-hurdle model, a different latent variable is used to model each decision process. The first hurdle is the selection model, which determines the boundary points of the dependent variable (Di
). The expected utility of adopting a technology (
$D_i^{\rm{*}}$
) is latent. The first hurdle is an adoption equation estimated with a Probit model given as:
$$ \eqalign{ {D_{i}^{\rm{*}} = aZ_{i} + u_{i}} \cr {D_{i} = 1 \ {\rm{if}} \ D_{i}^{\rm{*}} \gt 1} \cr { D_{i} = 0 \ {\rm{if}} \ D_{i}^{\rm{*}} \le 1}},$$
(1)
where Di = observed dependent variable that takes the value of 1 if a farmer i adopts at least one improved maize variety and 0 otherwise, Zi = vector of explanatory variables (farmer, farm-specific characteristics, institutional characteristics and technology-specific characteristics), a = vector of parameter estimates and ui = independently distributed error term. We based the selection of explanatory variables and their expected effects on theory and empirical literature on agricultural technology adoption, which points to some key socioeconomic determinants of adoption (Alene and Manyong, Reference Alene and Manyong2006; Asfaw et al., Reference Asfaw, Shiferaw, Simtowe and Haile2011; Croppenstedt et al., Reference Croppenstedt, Demeke and Meschi2003; Katengeza et al., Reference Katengeza, Mangisoni, Kassie, Sutcliffe, Langyintuo, La Rovere and Mwangi2012; Khonje et al., Reference Khonje, Manda, Alene and Kassie2015; Manda et al., Reference Manda, Alene, Gardebroek, Kassie and Tembo2016; Ouma et al., Reference Ouma, Bett and Mbataru2014; Shiferaw et al., Reference Shiferaw, Kebede and You2008; Solomon et al., Reference Solomon, Tessema and Bekele2014; Weyessa, Reference Weyessa2014). We also included perception of technological traits as motivated by the empirical work of Adesina and Zinnah (Reference Adesina and Zinnah1993). The variables are defined in Table 1.
Summary of definitions and measurement of variables
The second hurdle is the outcome model, which uses a truncated model to determine the intensity of adoption measured in terms of the proportion of farm area allocated to improved maize varieties. Therefore, the second hurdle uses observations only from those maize farmers who indicated a positive value on the use of improved variety. It is worth noting that the adopters of improved maize varieties do not grow the improved varieties at the same level of intensity. Hence, the intensity of adoption (intensity hurdle) of improved maize varieties is estimated using a truncated model, which is a Tobit-like model expressed as:
$$\eqalign{Y_i^* = \beta {X_i} + {v_i} \cr {Y_i} = \left\{ {\matrix{{Y_i^*} \quad {{\rm{ if }} \quad Y_i^* \gt 0 \quad {\rm{ and }} \quad {D_i}^* \gt 0} \cr 0 \quad\quad {{\rm{otherwise}}}} } \right.} $$
(2)
Where Yi = proportion of land allocated to improved maize by farmer i, Xi = vector of explanatory variables, β = vector of parameter estimates and vi = error term.
The observed value of the proportion of land allocated to improved maize is, therefore, given by:
$${Y_i} = {D_i^*}Y_i^{{*}}.$$
(3)
The error terms of the two decision models (adoption model and intensity of adoption model) are distributed as follows:
$$\left\{ \!\!\!\matrix{u_i \sim \ N ( 0,1) \cr \,\ v_i \sim \ N( 0,\sigma ^2)} \right..$$
(4)
The error terms ui and vi are usually assumed to be independently and normally distributed. It is assumed that for each respondent, the decision whether to adopt the technology and the decision about the adoption level are made independently. In the double-hurdle model, a variable appearing in both equations may have opposite effects in the two equations.
Propensity score matching
The PSM technique is a non-parametric approach that involves constructing a statistical comparison group by modeling the probability of adopting a technology on the basis of observed characteristics that are unaffected by the technology. Unlike the parametric methods of impact evaluation, PSM requires no assumption about the functional form in specifying the relationship between outcomes and predictors of outcome (Ali and Abdulai, Reference Ali and Abdulai2010). Basically, the PSM framework matches the observations of adopters and non-adopters according to their propensity scores. The propensity score is the predicted probabilities of adopting a superior technology conditional on the covariates (Rosenbaum and Rubin, Reference Rosenbaum and Rubin1983). The underlying principle of PSM is that the propensity scores from an estimated Probit model are used to find matches for technology adopters.
The estimation of average treatment effect on the treated (ATT) is specified as follows:
$${\rm{ATT}} = E{\{H_1}{\rm{|}}D = 1\} - E{\{H_0}{\rm{|}}D = 1\}.$$
(5)
The problem with the estimation of equation (5) is that E{H 0|D = 1} is not observable. However, it is possible to estimate equation (6) by replacing E{H 0|D = 1} with E{H 0|D = 0}as follows:
$${\rm{ATT}} = E\{{H_1}{\rm{|}}D = 1\} - E{\rm{\{ }}{H_0}{\rm{|}}D = 0\}.$$
(6)
Estimation of equation (6) is only possible when E{H 0|D = 1} = E{H 0|D = 0}, and this occurs when treatment is randomly assigned. In the absence of random assignment of treatment, PSM creates a counterfactual by modeling the probability of adoption conditional on observed characteristics, which are unaffected by the adoption. Therefore, controlling for observable characteristics (Z) assumes that technology adoption is random conditional on the observables.
$${\rm{ATT}} = E[{\rm{\{ }}{H_1}{\rm{|}}D = 1,P\left( Z \right)\} - E{\rm{\{ }}{H_0}{\rm{|}}D = 0,{\rm{}}P\left( Z \right)\}],$$
(7)
Where H 1 = value of the outcome for adopters of the new technology, H 0 = value of the outcome for non-adopters of the new technology, D = Adoption (1 for adopters of short-season maize, 0 otherwise) and Z = vector of explanatory variables.
This study employed nearest neighbor matching (NNM) in which individuals from the adopters and non-adopters that are closest in terms of propensity scores are matched and kernel-based matching (KBM) methods in which the weighted average of the outcome variable for all individuals in the group of non-adopters is used to construct the counterfactual outcome, giving more importance to those observations that provide a better match. Two important assumptions of PSM that needs to be satisfied to ensure reliable estimates are conditional independence and common support condition. Conditional independence assumption (CIA) also known as unconfoundedness implies that given a set of observable covariates, potential outcomes are independent of treatment assignment. The common support condition (overlap condition) requires that there should be sufficient overlap in propensity scores across the adopters and non-adopters. Only observations that fall in the common support region are used for the estimation. The ATT estimates may be biased due to violation of the CIA assumption. We tested for robustness of our results to possible bias using a simulation-based sensitivity test (Ichino et al., Reference Ichino, Mealli and Nannicini2008).
Results and Discussion
Adoption of short-season maize varieties in the study area
Adoption of short-season maize varieties was high in both LGAs (Table 2). In comparison with the status of adoption in a baseline report covering the research area by Ayanwale et al. (Reference Ayanwale, Abdoulaye, Ayedun and Akinola2011), adoption of the short-season maize varieties has increased over time (from 2008 to 2014). For example, adoption of short-season maize varieties in Musawa increased from 24 to 88.7%, while in Shanono, it increased from 60 to 87.3%. With respect to the adoption rate of the individual short-season maize varieties, the improved maize variety with the highest adoption rate was 99EVDTWSTR in both Musawa and Shanono LGAs. This indicates that farmers preferred this variety over the other improved maize varieties. The high adoption rate of 99EVDTWSTR may be due to its high yielding ability in addition to tolerance to drought and the parasitic weed Striga. Although, the other varieties adopted are early maturing and tolerant to Striga, 99EVDTWSTR normally produced grain yield that is higher than that of the other varieties.
Adoption rate of short-season maize varieties by year and varieties
Determinants of farmers’ decision to adopt short-season maize varieties
The likelihood ratio (LR) of 214.01 and 354.84 of the estimated Probit models for Musawa and Shanono LGAs, respectively, was significant at 1% probability level, and this implies the joint significance of the explanatory variables included in the models (Table 3). There is a variation in the factors that influenced the decision of the farmers to adopt short-season maize varieties in the two LGAs, as shown in Table 3. The estimated coefficient of age was positive and significantly related to the probability of adoption of short-season maize varieties at 1 and 10% in Musawa and Shanono, respectively. This suggests that an increase in age of the farmers will enhance their likelihood of adoption. This is because older farmers are assumed to have gained knowledge and experience over time and are better able to evaluate technology information than younger farmers, which can influence their adoption decision. The estimated coefficient of education was positive and significant in Shanono, implying that an increase in the educational level of the farmers has the tendency of improving their likelihood of adoption of short-season maize varieties. This finding compares favourably with that of Khonje et al. (Reference Khonje, Manda, Alene and Kassie2015), who obtained similar result on adoption of improved maize varieties in Eastern Zambia. This is because education improves the managerial skills and human capital of farmers. It also enhances their ability to assess and understand information on improved maize varieties, and this will invariably influence their probability of adoption of the improved maize varieties.
Double-hurdle estimates of determinants of adoption of short-season maize
Note: Values in parentheses are the t-values.
***, **, * implies significant at 1, 5 and 10% probability levels, respectively.
The estimated coefficient of farm size was positive and significant in Musawa (Table 3). This implies that farmers with larger farms are more likely to adopt an improved technology (especially modern varieties) compared with those with small farmers as they can afford to devote part of their fields to try out the technology. This is because land is a vital factor of production in agriculture, and farmers having large farm size are better endowed and can take risk in trying new things than poor farmers with small landholdings, who avoid taking risk in replacing their existing varieties with new ones. In accordance, Mignouna et al. (Reference Mignouna, Manyong, Mutabazi and Senkondo2011) found that farm size was a significant determinant of adoption of improved maize varieties in southern Zambia. In Shanono, the estimated coefficient of credit was positive and significantly related to the probability of adoption of short-season maize varieties (Table 3). This is in line with the findings of Solomon et al. (Reference Solomon, Tessema and Bekele2014) on adoption of improved wheat varieties. This implies that an increase in credit accessible to farmers will enhance their ability to purchase inputs such as improved seeds and, particularly, fertilizer for maize. The estimated coefficient for access to extension services was positive and significant in influencing the farmers’ decision to adopt improved maize varieties in both LGAs. This result is consistent with the findings of Ouma et al. (Reference Ouma, Bett and Mbataru2014) in Kenya. In both LGAs, the estimated coefficient of association membership was positively related to the farmers’ likelihood of adoption of improved maize varieties (Table 3). This implies that probability of adoption tends to increase with increasing the active involvement of farmers in associations. This is plausible as farmer associations afford farmers the opportunity of having access to improved seed varieties and other farming inputs at favourable cost, access to marketing services and knowledge sharing among members of the associations.
The estimated coefficient of drought resistance was positive and significant in influencing the likelihood of adoption of short-season maize varieties in Musawa (Table 3). Drought is a significant constraint in the Sudan Savannas of northern Nigeria where early or late season drought is frequent (Kamara et al., Reference Kamara, Ekeleme, Menkir, Chikoye and Omoigui2009). Drought-tolerant varieties are, therefore, important in such agroecological zone, and this explains the significant influence of drought tolerance on adoption. In both LGAs, the estimated coefficient of pest and disease resistance was positively related to the farmers’ adoption decision (Table 3). The estimated coefficient of early maturity had the expected positive sign and was significant in influencing adoption probability in Shanono (Table 3). This result agrees with Alene and Manyong (Reference Alene and Manyong2006) who found that earliness of maturity enhances the adoption of improved cowpea varieties. In the Sudan Savannas where early and late season drought is frequent, earliness of crop varieties is an important trait for farmers. Several other studies have shown that early maturing crop varieties are more likely to be adopted by farmers (Weyessa, Reference Weyessa2014). The estimated coefficient for high yield trait was positive, and this significantly influenced the farmers’ decision to adopt improved maize varieties in both LGAs (Table 3). This is an indication of the importance of high yield as a desirable quality in adoption behaviour of farmers. Yield is the primary trait for smallholder farmers in addition to other quality traits. High yielding varieties are more likely to be adopted than lower yielding varieties.
Determinants of intensity of adoption of short-season maize varieties
The estimated truncated regression model shows that the LR of 205.39 and 166.76 of the fitted models for data generated from Musawa and Shanono LGAs, respectively, was significant at 1% (Table 3). This indicates the joint significance of the explanatory variables in influencing the intensity of adoption of the short-season maize varieties. The results revealed that there is some variation in the results of the estimated Probit model and the truncated regression model, and this implies that the factors that influenced the farmers’ decision to adopt improved maize varieties were not exactly the same factors that influenced their intensity of adoption in both Musawa and Shanono LGAs. This further justifies the use of the double-hurdle model in examining the adoption of short-season maize varieties in the study area. In Shanono, age was significant and negatively related to adoption intensity of short-season maize varieties. This implies that older farmers have less tendency of increasing their adoption intensity compared to younger farmers, and a plausible explanation for this is that older farmers are usually risk averse while younger farmers are typically less risk-averse and are more willing to try new technologies. The estimated coefficient of education was significant and positively related to intensity of adoption of short-season maize varieties in Musawa (Table 3). This result agrees with Hassen et al. (Reference Hassen, Emana, Kassa and Haji2012), who established that education was positive and significant in influencing the intensity of adoption of chemical fertilizer technology in North eastern highlands of Ethiopia.
In Musawa, the estimated coefficient of household size was negatively related to intensity of adoption of short-season maize varieties and was significant at 10%, indicating that intensity of adoption tends to decrease with an increase in household size (Table 3). This is because a large household size implies more mouths to be fed. This could place a high demand on the ability of the household to purchase improved seeds and complementary inputs such as fertilizer especially for smallholder farmers. This is in conformity with the findings of Ouma et al. (Reference Ouma, Bett and Mbataru2014), who found that household size had negative effect on intensity of adoption of improved maize varieties in Kenya. In Shanono, the estimated coefficient of farm size was positively related to the farmers’ decision on the intensity of adoption of improved maize varieties (Table 3). This is expected, because increasing intensity of adoption would mean increasing the land area allocated to improved maize varieties. In other words, access to larger land holding is an incentive for enhancing intensity of adoption of short-season maize varieties. In Musawa, the estimated coefficient of credit was significant and negatively related to the adoption intensity of short-season maize varieties (Table 3). This result is against a priori expectation as it suggests that decision on intensity of adoption by the adopters tends to decrease as their access to credit increases. This could be attributed to the incidence of diversion of credit meant for farm activities to off-farm activities, thereby making credit counterproductive. The estimated coefficient of extension was positive and significant in influencing the farmers’ decision on the intensity of adoption of improved maize varieties in Shanono (Table 3). This is because access to extension services enhances the acquisition of the relevant information that promotes technology adoption. In fact, information plays a vital role in agricultural technology adoption.
For early maturity, its estimated coefficient had the expected positive sign and was significant in influencing the maize farmers’ decision on the intensity of adoption of short-season maize varieties in both Musawa and Shanono (Table 3). This implies that improved maize varieties with early maturing quality will gain more acceptance by the farmers. This result agrees with Weyessa (Reference Weyessa2014), who found that farmers’ perception of early maturity was positively associated with the use intensity of improved tef varieties in Ethiopia. The estimated coefficient of high yield was positive and significantly influenced the farmers’ intensity of adoption of the maize varieties in Musawa (Table 3). The significance of farmers’ perception of varietal characteristics of the short-season maize varieties in influencing their intensity of adoption is a pointer to the fact that farmers are desirous of certain qualities of the improved maize varieties, which will encourage them to increase their extent of adoption of the varieties or otherwise.
Beneficial effects of adoption of short-season maize varieties
The two main assumptions underlying PSM were evaluated before proceeding to establish the causal effect of adopting the improved maize varieties. The quality of the matching process was assessed to ensure that the common support assumption is met (Supplementary Material Table S1). The specification of the propensity score estimation process was successful regarding balancing distribution of covariates between adopters and non-adopters. This is evident by the low pseudo-R2, low mean standardised percentage bias and insignificant p-values of LR test after matching in comparison with the values before matching as well as a high total percentage bias reduction. To test the robustness of ATT for failure of the CIA assumption, a simulation-based sensitivity analysis as put forward by Ichino et al. (Reference Ichino, Mealli and Nannicini2008) was implemented (Supplementary Material Table S2). Based on a neutral confounder as additional matching factor, the results revealed that the estimates with the inclusion of the binary confounder differ by less than 10% from the baseline matching estimates for all the outcome variables. This result indicates that the PSM technique yields robust estimates of the ATT, and therefore, the baseline ATT is robust to possible deviations from the CIA.
The farm-level productivity performance of the adopters of short-season maize in Shanono LGA was higher than that of the non-adopters with a mean difference of 873.04 kg ha−1 at 1% probability level (Table 4). In Musawa LGA, the productivity increasing effects are much higher (1149.76 kg ha−1) and also significant at 1% (Table 4). This implies that adoption of short-season maize varieties had productivity increasing effects in both LGAs. This result compares favourably with empirical findings that revealed significant productivity increasing effects of improved maize varieties (Abdoulaye et al., Reference Abdoulaye, Wossen and Awotide2018; Khonje et al., Reference Khonje, Manda, Alene and Kassie2015; Wossen et al., Reference Wossen, Abdoulaye, Alene, Feleke, Menkir and Manyong2017). Drought, Striga infestation and poor soil fertility are major limitations to maize production in the Nigerian Sudan Savannas. These constraints often occur in the farmers’ fields causing significant reduction in yield of maize (Kamara et al., Reference Kamara, Ekeleme, Menkir, Chikoye and Omoigui2009). The maize varieties introduced by the Sudan Savanna taskforce in the LGAs were early maturing, allowing them to escape drought and reduce the risk of crop failure. They were also tolerant to drought and Striga giving significant yield advantage over the farmers’ local varieties. This explains the high yield difference between adopters of the improved varieties and non-adopters.
PSM estimates of the impact of adoption of short-season maize
Note: ***p < 0.01, values in parentheses are the t-values, official exchange rate at the time of the study: 1 US Dollar to 198 Naira.
With respect to income of farmers in Shanono LGA, adopters had significantly higher maize and household incomes than the non-adopters (Table 4). The matching results from NNM revealed that the ATT in maize income and household income between the two groups was 
154183.84 (US$198.7) and 321937.97 (US$1625.9), respectively, at 1% probability level (Table 4). In Musawa LGA, the ATT in maize income and household income between the two groups was estimated at 22046.72 (US$111.3) and 28928.13 (US$146.1), respectively, at 1% probability level (Table 4). This indicates that adoption generated positive maize income and household income increasing effects. The direct income effect of technology adoption is on maize income, which produces indirect effects on household income. The income from adoption of short-season maize produces multiplier effect as it can be reinvested into other farm and off-farm activities to generate more income towards achieving sustainable food security and poverty reduction. This is the pathway through which income generated from maize production produces indirect effects on other household income sources, thereby contributing in enhancing household income. In rural Zambia, adopters of a combination of sustainable agricultural practices (SAPs) of maize production had on average between 43 and 75% more income than non-adopters (Manda et al., Reference Manda, Alene, Gardebroek, Kassie and Tembo2016).
There were variations in the impacts of short-season maize varieties in Musawa and Shanano LGAs as productivity effect was higher in Musawa while maize income effect was higher in Shanono (Table 4). We expected Musawa with higher productivity effect to possibly have higher maize income effect especially as farm production cost is similar across the two LGAs. However, the result proved otherwise. This was attributed to the notable output market imperfections in Musawa, which limited sales and led to low market price of produce and, invariably, low returns. In the case of Shanono, the farmers had favourable output prices as it is located near a major grain market known as Dawano market, which affords farmers the opportunity of selling their produce at a favourable price and earn higher returns. In other words, output price relations favoured farmers in Shanono over farmers in Musawa.
In general, the findings show that short-season maize varieties promoted via IPs are not only suitable as risk mitigation technologies in the face of variable climate and missing insurance market for smallholders; they generate beneficial effects for farmers. However, we did not make causal inference on the impact of IP, since our interest is on the impact of short-season maize promoted through the IPs and its implications for sustainable maize production in the face of weather shocks. On this issue, only Pamuk et al. (Reference Pamuk, Bulte and Adekunle2014, Reference Pamuk, Bulte, Adekunle and Diagne2015) have produced quantitative evidence on the impacts of IPs. Our research differs from these previous works in two ways. First, our interest is not necessarily on the impact of IPs but on the impact of short-season maize varieties promoted via IPs. Hence, we refrain from making causal inference on IPs except for the varieties promoted via them. Second, the studies produced short-term effects of IPs as the evaluation was done 2 years after introduction of the IPs. However, our results are long-term effects as we evaluate impact of short-season maize varieties 6 years after their promotion via IPs.
Conclusion
In this paper, we provided quantitative evidence on the state of adoption and beneficial effects of early maturing maize varieties promoted through IPs in short-season environments of Sudan savannas of Nigeria. We used cross-sectional household data elicited from 600 sampled households, double-hurdle model and PSM. The study revealed that there was a remarkable increase in the adoption of short-season maize varieties in 2014 compared to what was obtained in 2008. For instance, adoption rate of short-season maize varieties in one of the LGAs increased from 24 to 88.5%. The results of the estimated double-hurdle model showed that age, extension, association and perception among other variables were the drivers of farmers’ decision to adopt short-season maize varieties across the two LGAs. The intensity of adoption of the varieties across the two LGAs was significantly influenced by household size and perception of earliness of varieties among others. The result of the PSM revealed that adoption of short-season maize varieties increased productivity, maize income and household income with some variations across the LGAs. Even though adoption of the maize varieties increases overtime and produces net benefits for farmers, we cannot completely attribute it to the IPs. In general, our results imply that short-season maize varieties are best suited for sustainable maize production in the short growing conditions than late maturing varieties, which are more vulnerable to early or late season drought. Then, the adoption of short-season maize can ensure sustainable maize production in the face of climate variability and generate beneficial effects for farmers, and agricultural extension systems can leverage the use of IP in promoting short-season varieties and enhance farmers’ adoption decisions.
Supplementary materials
To view supplementary material for this article, please visit https://doi.org/10.1017/S0014479718000467.
Author ORCIDs
Oyakhilomen Oyinbo 0000-0002-9687-3097
Acknowledgements
The authors sincerely acknowledge the Forum for Agricultural Research in Africa (FARA) for funding of the project, and International Institute of Tropical Agriculture (IITA), Nigeria for leading the implementation of the project in Nigeria and also, providing the short-season maize varieties which were disseminated to farmers. Also, the involvement of Agricultural Development Programmes of Kano and Katsina States in the dissemination of the varieties deserves to be highly acknowledged.
Financial support
None.
