Adoption of innovation and farm productivity in the Sudano-Saharan zone in Cameroon

Abstract

Faced with the shocks that are disrupting the agricultural system in Sub-Saharan African countries, farmers are faced with the dual challenge of adapting to the effects of these shocks and improving their productivity. Thus, agricultural innovation is one of the adaptation strategies for improving productivity. The objective of this paper is to assess the impact of the adoption of agricultural innovations on farm productivity in the Sudano-Saharan zone of Cameroon. We use survey data from 721 farms in the North and Far North (Sudano-Sahelian zone) regions of Cameroon carried out within the framework of the project “Contrat de Désendettement et de Développement1 agrosystèmes du Nord”. These data are applied by the endogenous switching regression model, which accounts for the selectivity bias of the sample while taking into account the differential impact between farms that have adopted agricultural innovations and those that have not. The results show that the adoption of agricultural innovations significantly improves farm productivity. The effect of these farm innovation adoptions is about 646.44 kg/ha compared to those that did not adopt any innovations.

Keywords:

• adoption of agricultural innovation
• Cameroon
• farms
• productivity
• Sudan-Saharan zone

JEL Classification Code:

• O55
• D24
• Q12
• N57

1. Introduction

The summit of the Heads of State of the African Union held in Malabo in 2014 on the theme “Growth and transformation of agriculture” suggested the establishment of effective means to increase the performance of the agricultural sector, reduce poverty and ensure food security (Amani et al., 2022; Karume et al., 2022). Indeed, Africa is essentially agricultural and struggles to feed its population due to the failure of the means implemented to improve agricultural productivity and living conditions. This situation is all the more due to the use of rudimentary tools, unsuitable cultivation techniques, the virtual absence of extension agents, limited access to financing and agricultural shocks. In addition to population growth and the effects of climate change, African economies are poorly diversified and heavily dependent on rain-fed agriculture (Mushagalusa Balasha et al., 2021; Rasul, 2021). These characteristics link the development prospects of these economies to the climate and may reduce food productivity (Bossio, 2021; Cline, 2007; Parry et al., 2005). Moreover, IPCC (2021) showed in terms of outlook that access to food will be strongly affected following a decrease of about 50% in subsistence agriculture yields by 2020.

Faced with this poor situation, the identification of innovations is essential to sustain food crop production and ensure food security (Bryan et al., 2009; GIEC & Pachauri, 2014). Innovation is understood as the new combination of production factors that can be expressed through the implementation of new products, new production procedures and the construction of new outlets or access to new resources (Schumpeter, 1934). Ouedraogo et al. (1996) define agricultural innovation as any technique likely to improve agricultural production, whether it comes from the rural environment or is introduced from outside. This agricultural innovation is assimilated to the adoption of new seed varieties, new cultivation techniques, monoculture, irrigation, modification of the dates of semi, a new restructuring of the farm and access to new resources (Basse et al., 2022; FAO, 2016; Onyeneke, 2020). For some authors, these agricultural innovations make it possible to remedy agricultural shocks through their productive efficiency (Asseng & PannelL, 2013; Devkota et al., 2017; Niles et al., 2016; Zhang et al., 2015).

The adoption of innovations to adapt to agricultural and climatic shocks is also defined as the modification of agronomic practices, agricultural processes and capital investments in response to observed or anticipated climatic threats (Alosani et al., 2023; Anesukanjanakul et al., 2019; Easterling et al., 2007; Okello & Luttah, 2022). Households adopt innovative strategies to cope with climate hazards by altering the agricultural calendar, growing different crop species and varieties, implementing soil and water conservation practices, diversifying activities, and establishing an irrigation system (Amani et al., 2022; Bagula et al., 2022; Chuma et al., 2022). In order to cope with so-called climatic hazards, some households use innovations and knowledge adapted to the local context or those of local people based on observation, practices and interpretation of natural phenomena (Molua, 2022).

In Cameroon, agriculture is the main activity that occupies more than 60% of the population and contributes 21% of the gross domestic product (FAO, 2017). The agricultural sector generates an annual average of 22% of GDP, contributes to over 55% of trade, provides over 50% of raw materials for industries, and is the main source of income and employment for the working population (AfDB, 2022). According to the National Institute of Statistics, approximately 90% of Cameroonian household land is less than 2 hectares (INS, 2017). Despite the social importance of the agricultural sector, average cereal productivity in the semi-arid zone remains low (1006.9 kg/ha), compared to the national average of 1269.5 kg/ha (Ahmadou et al., 2023). In the Sudano-Sahelian zone, for example, there has been a decline in the productivity of the main cereals grown (sorghum, millet, maize and rice). This drop in productivity is all the more due to a low use of improved seeds (less than 10% for millet and sorghum, less than 20% for rice and less than 50% for maize); an almost total lack of training for farm households in agricultural trades (more than 80% of households); less than 10% of farm households own agricultural equipment (wheelbarrows, all-purpose carriers, sprayers, watering cans, tractors, etc.) and less than 10% of households have access to financing for their agricultural activity (INS, 2017). Despite these weaknesses, several simulation models predict an increase in temperature and a decrease in rainfall for the coming decades with rather serious impacts on agricultural production (Timgem et al., 2008). There is also growing food insecurity in rural areas (22%) as opposed to urban areas (10.5%) with strong disparities between regions.

The Sudano-Sahelian zone is the most affected by food insecurity, with a prevalence rate of 33.7% (FAO, 2017), an average poverty rate of 74% (INS, 2017) and a drop in agricultural production of over 40% (INS, 2017). This area, which stretches from the Benoué basin to the shores of Lake Chad, is characterized by unstable seasons and dry conditions that make agricultural activity precarious (Wakponou & Nyéladé, 2014), population growth from 1,870,000 in 1976 to 6,350,000 in 2015 according to the Bureau central des recensements et des études de population (BUCREP, 2010) and border insecurity. These characteristics accentuate the pressure on accessibility to agricultural resources and food availability, and pose a challenge in terms of new production methods and strategies for agriculture in the area. The latter are linked to low agricultural productivity, which is also characterized by low input levels, low government subsidies and the use of rudimentary tools (Molua & Utomakili, 1998).

Taking the increase in agricultural productivity as an essential component of the success of all rural development strategies, several studies have studied the impact of strategies for adapting to agricultural shock on agricultural production (Douswe, 2022; Koguia et al., 2021; Molua, 2022; Njoya et al., 2022). Although this substantial literature shows the need for agricultural innovations to address agricultural shocks, it does not improve understanding of the relationship between different coping strategies and their effect on agricultural yields. Furthermore, research in Africa and Cameroon examines the causes and effects of agricultural and climatic shocks on agricultural productivity (Ntali et al., 2023; Tagang et al., 2021). These studies did not consider innovative adoption strategies and their effect on agricultural productivity (Issahaku & Abdulai, 2019; Khanal et al., 2018). While agricultural innovations can contribute to improving yields and agricultural productivity. In order to fill this gap, this work answers the question: what is the impact of the adoption of innovations on the productivity of farms in the Sudano-Saharan zone of Cameroon? Thus, we examine the adoption of agricultural innovations and their impact on the productivity of farms in the Sudano-Saharan zone of Cameroon.

We define agricultural innovations in the face of agricultural and climate shocks more broadly to include crop choice and yield improvement (FAO, 2013). The use of improved seeds to face agricultural and climate shock, is defined as the adoption of an innovative agricultural practice including modern, drought-resistant and early maturing varieties that enables farmers to cope with erratic rainfall or short rainy seasons (Issahaku & Abdulai, 2019). This agricultural innovation also captures the evolution of crops in response to climate variability (DiFalco & Veronesi, 2013). It is common to use new seed varieties adapted to climatic conditions in northern Cameroon. The construction of bunds is an innovative anti-erosion agricultural practice that stabilizes upstream water runoff and protects downstream crop lands from erosion. This agricultural innovation concentrates runoff and organic manure in small pits, promotes seedbed moisture conservation (Zougmore et al., 2014). In the Sudano-Sahelian zone of Cameroon, this bund technique primarily used to minimize soil loss due to erosive rains or to reduce soil water evaporation due to high temperatures could improve crop performance. Reducing crop area is an innovative practice to respond to agricultural hazards such as prolonged dry and rainy seasons. Prolonged flooding due to extended rainy seasons. Crop diversification is an innovative practice adopted to adapt to climate variability. Indeed, in case of unforeseen climatic events that may affect agricultural production, such as the emergence of pests and diseases, sudden onset of frost or drought, erosion and loss of soil fertility, farmers cultivating only one type of crop will be more exposed to high risks. This innovation is a pillar of agroecology that increases the portfolio of crops so that farmers do not depend on a single crop to generate their income. The introduction of a wider range of varieties leads to a diversity of agricultural production that can reduce total crop failure and provide alternative means of income generation for farmers. In northern Cameroon, a diversified plot increases the farmer’s chances of coping with the uncertainty and changes created by climate change (Tagang et al., 2021).

The observation of overall activities of agricultural innovations as well as the individual type of innovations, this work contributes to the literature in more than one way. It builds on recent advances of the impact evaluation literature, in particular the use of endogenous switching regression that allows for selection bias (Abdulai & Huffman, 2014; Issahaku & Abdulai, 2019; Takam‐Fongang et al., 2019). This approach allows to identify specific drivers of agricultural innovation adoption and the impact of innovation adoption on farm productivity as well. Specifically, we first examine the factors that affect farmers’ decision to adopt agricultural innovations and then the impact of the adoption of each type of agricultural innovation on farm productivity. Following this method, which accounts for the sample selection bias while taking into account the differential impact between farms that adopted the innovations and those that did not, we use propensity score matching (PSM) to assess the robustness of the results.

We use survey data from selected farms in the northern and far northern regions of the Sudano-Sahelian zone of Cameroon. Given that our sample is composed of cultivated plots in hectares and quantities produced, we capture productivity as the ratio of quantity produced to the area (in hectares) of the farm.

This paper is organized as follows: the second section presents the literature review, the third section provides the methodological analysis, the fourth section presents the results as well as the discussions and the final section concludes.

2. Literature review

As agricultural innovation is essential to meet future food challenges, several studies have been invested in it to show its importance in the fight against agricultural shocks (climate change, rising input prices…). These agricultural innovations will be able to support developing countries in achieving sustainable development objectives (Alosani et al., 2023; Chishti & Sinha, 2022; Salifu & Salifu, 2015). In the search for sustainable solutions that can be easily applicable in these countries, the authors analyzed the process of agricultural innovation and the factors that lead to the adoption of innovations. To do so, the authors begin with an analysis of the determinants of innovation adoption and then analyse the effect of this innovation adoption on farm productivity. By analyzing the case of the Wa region in Ghana, Salifu and Salifu (2015) show that the adoption of agricultural innovations is related to the marital status, age, education level, and experience of the farmer in growing maize. They also find that the characteristics of improved seeds have an effect on the probability of adopting improved maize varieties. Similarly, Kalinda et al. (2014) find that for other Africans, gender, membership of a farmer organization, farm size, favourable perception of the yield potential of improved maize varieties and the price of their production have a positive effect on the adoption of improved maize varieties. Mabah et al. (2013), Ntsama and Kamgnia Dia (2008) and Takam‐Fongang et al. (2019) on the case of Cameroon, identified the level of education, agricultural training, commercial orientation, membership of farmers’ organizations, farm size, contact with extension agents as factors having a significant impact on the adoption of agricultural innovation activities. These analyses are in line with those of Baffoe-Asare et al. (2013) and Asfaw et al. (2011) in Sierra Leone and Ethiopia, showing the close relationship between socio-economic factors and the adoption of agricultural innovations (Kaczmarek et al., 2021; Mardiharini et al., 2021; Newnham & McMurray, 2021; Tsambou & Fomba, 2023).

In contrast to these authors, who only analyze the determinants of agricultural innovation adoption, some authors conduct impact assessments of these innovation adoptions on agricultural performance. Thus, Kalinda et al. (2014) show that the adoption of agricultural innovations such as improved maize varieties increases yields by 674.37 kg per hectare. Although these results corroborate with those of Manda et al. (2016) and Nabasirye et al. (2012), they are contrary to those of Takam‐Fongang et al. (2019) showing no impact of the adoption of improved varieties on maize farm yields. On the other hand, Abdulai and Huffman (2014) show that household size, farm size, access to credit, membership of a producer organization, access to extension agents and perception of shocks have a positive impact on rice productivity. By isolating these effects using an endogenous switching model, the authors conclude that the adoption of soil and water conservation technology innovations by a farmer increases productivity by more than 448 kg/ha compared to non-adopters. Issoufou et al. (2017) on data from 612 millet farmers, show that the adoption of agricultural innovations in terms of improved seeds increased millet productivity by 406.93 kg/ha. They note a significant difference of 211.74 kg/ha between adopters of agricultural innovations and non-adopters. These authors thus concur with the analyses of Shiferaw et al. (2014) in Ethiopia which showed the positive effect of agricultural innovation adoption on food production.

Although this interesting literature presents very conclusive results, there is a lack of literature on the Sudano-Saharan zone which constitutes the granary of Cameroon. Moreover, this existing literature focuses on a single particular innovation (improved varieties), leaving aside process innovations such as crop diversification, modification of semi dates, construction of bunds, etc., which can be apprehended through crop diversification. This study goes further by taking into account all types of agricultural innovations implemented in agricultural activity. The idea is to evaluate globally and individually the impact of the adoption of innovations on farm productivity.

3. Methodological analysis

In response to the problem of this research, the methodology revolves around data source, model specification and analysis of descriptive statistics.

3.1. Data analysis

Although some work (Khonje et al., 2015; Ntsama & Kamgnia Dia, 2008; Takam‐Fongang et al., 2019) has been carried out on Cameroon, it has focused on the forest zone whereas the Sudano-Saharan zone suffers from a real agricultural problem (FAO, 2017). Thus, this work uses cross-sectional data from a survey of 721 farms in the North and Far North regions of Cameroon’s Sudano-Sahelian zone. These data were collected as part of the agrosystèmes-nord project “Contrat de Désendettement et Développement (C2D)¹ agrosystèmes du nord (Actions 6 et 7)” piloted by the Institut de Recherche Agricole pour le Développement (IRAD). The survey was carried out in the North and Far North regions of Cameroon, which are part of the semi-arid zone. Based on the agro-ecological zoning of Dugué et al. (1994), the survey identified seven localities in the two regions. Only four localities (Mowo, Sirlawe, Zera and Sanguere) were included due to the security situation caused by the Boko Haram problem. A random sample of 800 farms was then drawn. At the end of the survey, 9.87% of the farm questionnaires were rejected. Due to poorly completed questionnaires and doubtful answers, 79 observations were excluded from the analysis, resulting in a sample of 721 farms.

This survey through an open questionnaire includes related questions on the adoption of agricultural innovations, the socio-economic and demographic characteristics of farmers, the production structure of farms, the socio institutional environment and the perception of farmers to agricultural shocks. The objective of this survey was to design, evaluate and disseminate efficient and sustainable agrosystems in rural areas of the Far North of Cameroon. This objective makes it possible to contribute to the sustainable improvement of the productivity of polyculture farms in North Cameroon while proposing techniques that are adapted and efficient to agricultural development structures. Based on the agro-ecological zoning of Dugué et al., (1994), the survey took into account physical criteria (climate and soils), demographic criteria (population density and movement), potentialities and constraints, agricultural conditions, crops in the area and economic criteria (market presence and isolation). Although this technique identified seven zones, the survey covered only four zones according to the security conditions related to the “Boko Haram” problem. In the latter, a reference terroir (Mowo, Siriawe, Zera and Sanguere) was identified with a sample of 200 farms to be obtained by random selection. At the end of this survey, the rejection rate of the questionnaires (badly filled in, doubtful answers, non-agricultural farms) was 7.5%, 10.5%, 17% and 4.5% respectively for Mowo, Siriawe, Zera and Sanguere. This rejection constituted a negligible rate (5.2%) of the sample. This made it possible to have data for four terroirs, including 25.66% at Mowo, 24.83% at Siriawe, 24% at Zera and 25.5% at Sanguere.

The Sudano-Sahelian zone of Cameroon, which stretches from the Benoué basin to the shores of Lake Chad, was chosen for the study because of the acuteness of the effects of climate change, which translates into highly unstable seasons and, as a corollary, a pronounced drying up of the environment, making agricultural activity and food supplies precarious (Wakponou and Nyéladé, 2014). One of the reasons is also the acceleration in population growth (from 1.87 million in 1976 to 6.35 million in 2015) fuelled by migratory flows from the Central African Republic, Chad and Nigeria (INS, 2015). In addition, the context of cross-border insecurity in this area in relation to the expansionist ambitions of the Boko Haram group, accentuate the pressure on accessibility to local agricultural resources and food availability. This presents new challenges for Sudano-Sahelian agriculture, which is increasingly subject to growing demand from agri-food companies (brewing, dairy and livestock industries, oil mills, cookie factories, etc.). The Ministry of Agriculture shows that the average poverty rate in this zone was 74%, compared with 37% at national level, with a deteriorating food situation and falling production forecasts (sometimes in excess of 40%), mainly in the territories most exacerbated by cross-border territorial insecurity (Logone-Et-Chari, Mayo-Sava, Mayo-Tsanaga) and those subject to the effects of migratory flows of internally displaced persons. These indicators reveal the growing precariousness of Cameroon’s Sudano-Sahelian zone, which is why the study area was chosen. It is a semi-arid zone located in the north-west region of Cameroon, divided into 10 departments covering 100,350 km2 (21% of the country’s total surface area), with around 5,530,643 inhabitants (29% of the country’s total population) (BUCREP, 2010). As shown in Figure 1, this area has a rainy season from June to August, and a dry season from September to March.

Figure 1. Agro-ecologicalzoning of the northern and Far North regions in Cameroon.

Source: Adapted from, Dugué et al. (1994).

3.2. Measuring innovation adoption variables and productivity

In explaining that the economy is governed by a phenomenon of creative destruction, Schumpeter (1942) sees innovation as the foundation of capitalism to which all exploitation should adapt. This innovation is the implementation of a new or substantially improved product, process, marketing method or organizational method on the farm (OCDE, 2005) From this definition, agricultural innovation is the use of new methods of agricultural production such as: the use of improved seeds, irrigation, soil conservation, monoculture, modification of semi dates. The adoption of each of these innovations refers to the decision to implement new technical proposals in the existing farming system and to progressively improve their use. Constituting the “essential ingredient” of agricultural productivity, the adoption of these innovations depends on the socio-economic characteristics of the farms (Mabah et al., 2013), the characteristics of the farmers, the information they have, the conditions of access to the necessary resources (Rogers, 2004), the institutional and economic environment (Young, 2011).

In this work, farmers were asked whether they had adopted agricultural innovations in the last 10 years. The adoption of one of these agricultural innovations involves a choice between adopting or not adopting a new agricultural practice or a new production strategy to adapt to agricultural shocks (slowdown in yield growth, negative environmental impact, crop destruction by wild animals, increase in agricultural input prices, climate change, fall in agricultural prices). To this end, the majority of farmers reported having adopted new agricultural practices such as the use of improved seeds, the construction of bunds, the reduction of cultivated area and crop diversification. Thus, the adoption of agricultural innovation is a binary variable that is worth one unit if the farm has adopted a new production strategy and 0 if not.

With regard to productivity, the questionnaire on the production structure provides information on the quantities produced and the areas of agricultural holdings. This productivity is the ratio of the quantity produced in kilograms (kg) to the area in hectares (ha) of each farm. This ties in with the theoretical point that the productivity of the land is the ratio of the annual production quantity to the total area of the production unit. The latter is different from labour productivity, which measures the efficiency of the labour incorporated in the productive process, and capital productivity (Tsambou & Et Fomba, 2021; Tsambou & Fomba, 2023). The other variables are presented in Table 2.

3.3. Specification of the econometric model related to the adoption of agricultural innovations

Assessing the effect of agricultural innovation adoption on farm productivity can be reduced to an impact assessment, so several models such as double difference (Yorobe et al., 2011) and propensity score matching (Mendola, 2007) have been used. But the double difference requires knowledge of the situation of farms before and after the adoption of innovation, whereas propensity score matching does not take into account unobserved effects. In this case, the farms adopted the innovations at different points in time, which most likely leads us to the endogenous switching regression model that takes into account both observed and unobserved factors (Abdulai & Huffman, 2014; DiFalco et al., 2011; Takam‐Fongang et al., 2019). This model (endogenous switching regression model) accounts for the selectivity bias of the sample while taking into account the differential impact between farms that adopted the innovations and those that did not. Based on recent work (Abdulai & Huffman, 2014; Coromaldi et al., 2015; DiFalco & Veronesi, 2013; DiFalco et al., 2011; Kassie et al., 2014; Takam‐Fongang et al., 2019), this work uses the endogenous switching regression model and propensity score matching toassess the robustness of the results.

Initially, since the Sudan-Saharan zone of Cameroon is a region highly vulnerable to agricultural shocks, a farm “dichotomously” chooses (yes or no) to adopt an agricultural innovation if it believes that it will enable it to maximize its future agricultural yield. This dichotomous choice (yes and no) corresponds to a yield function ($Y$) that can be written as:

(1) $Y_i=Z{ }_i^{\prime }\alpha +{\epsilon }_i$(1)

Since this adoption of innovation ($A_i$) is a visible manifestation of the unobservable latentvariable ($A_i^{\ast }$) on the farm, the problem is modelled in terms of a farm performance function (Y). For a farm $i$with characteristics Z_i, the adoption of agricultural innovations is conditional on a certain expected return, while Y(1, Z_i) the refusal to adopt agricultural innovations is conditional on an expected return Y(0, Z_i). We have:

(2) $A_i=\left\{\begin{array}{c}1ifY\left(1,Z_i\right)>Y\left(0,Z_i\right)\\ 0ifY\left(1,Z_i\right)\le Y\left(0,Z_i\right)\end{array}\right.$(2)

Under the basis of the profit maximization theory, the farmer chooses the situation that allows him to maximize his future returns. This brings us back to the case of the latent variable that captures the benefits due to the choice of whether or not to adopt agricultural innovations. Thus, we have:

(3) $A_i=\left\{\begin{array}{c}1if{A}_i^{\ast }>0\\ 0if{A}_i^{\ast }\le 0\end{array}\right.with{A}_i^{\ast }=Y\left(1,Z_i\right)-Y\left(0,Z_i\right)=Z_i\alpha +{\epsilon }_i$(3)

Where $Z_i=\left(1,Z_{i1},Z_{i2},Z_{i3},\dots ,Z_{ik}\right)$ is the vector of the explanatory variables, $\alpha$ the vector of the parameters to be estimated, ${\epsilon }_i$ is the realization of the random events distributed according to a normal law. The adoption of agricultural innovations (A_i = 0 ou A^*_i>0) can be influenced by current climatic factors, the perception of production shocks, access to financing, access to information on available agricultural innovations, access to agricultural extension, the characteristics of the farmer and the characteristics of the production structure.

Second, we model the effect of the adoption of agricultural innovations on farm performance. After exploring several functional forms, the most robust is a quadratic specification whose simplest approach to examining the impact of farm innovation adoption would be to include in the operator performance equation a dummy variable (A) equal 1 if the operator had adopted farm innovations. Then apply ordinary least squares (OLS). However, this approach could result in estimates that are biased by the assumption that the adoption of farm innovations is exogenously determined when it is potentially endogenous. The decision to adopt an agricultural innovation or not is voluntary and may be based on individual farmer self-selection. Farms that have adopted agricultural innovations may have systematically different characteristics from farms that have not adopted.

In addition, farmers may have decided to adopt agricultural innovations based on expected yields. Unobservable characteristics of farms and their operators can influence both the decision to adopt innovations and farm performance, resulting in inconsistent estimates of the effect of creation on farm productivity. Thus, taking into account the endogeneity of the farm innovation adoption decision, we estimate a model of simultaneous equations of innovation adoption and farm yields with endogenous maximum likelihood switching to full information. Unlike studies that use the fitted values automatically generated by the non-linearity of the selection model to control for endogeneity (Aboal & Tacsir, 2017; Fu et al., 2018; Tsambou & Et Fomba, 2021), we use an exclusion restriction to identify this model (Abdulai & Huffman, 2014; Tsambou & Fomba, 2023). This restriction is necessary when there are certain variables that directly affect the selection variable (adoption of innovations), but not the outcome variable approximated by agricultural productivity (Coromaldi et al., 2015).

Thus, we use variables related to information on agricultural shocks (information from radio, climate information and the perception of production shocks) as instruments for selecting the agricultural performance function. After economic intuition, the eligibility of these instruments is effective after a test of sample homogeneity and a falsification test (DiFalco et al., 2011). Thus, if the selection instruments are valid, they will have an impact on the decision to adopt agricultural innovations, but not on the quantity of production (yield) of farms that have not adopted agricultural innovations. In addition, to account for selection bias, we adopt an endogenous switching regression model of farm performance in which farms face two regimes: adopting agricultural innovations (Regime 1) and not adopting agricultural innovations (Regime 2). The regression model is defined as that of DiFalco et al. (2011), Abdulai and Huffman (2014) and Takam‐Fongang et al. (2019).

(4) $\{\begin{array}{c}Innovationadoptions:Y_{1i}=X_{1i}{\beta }_1+{\mu }_{1i}if{A}_i=1\left(a\right)\\ Non-adoptioninnovations:Y_{2i}=X_{2i}{\beta }_2+{\mu }_{2i}if{A}_i=0\left(b\right)\end{array}$(4)

Where $\mathrm{A}$ is the probability of adoption of agricultural innovations, Y_i represent the quantity of production in regimes 1 and 2, X_i represent the vector of explicative variables. ${\beta }_1$ and ${\beta }_2$ are the vectors of the parameters to be estimated, the error terms in the selection Equation 3(3) $A_i=\left\{\begin{array}{c}1if{A}_i^{\ast }>0\\ 0if{A}_i^{\ast }\le 0\end{array}\right.with{A}_i^{\ast }=Y\left(1,Z_i\right)-Y\left(0,Z_i\right)=Z_i\alpha +{\epsilon }_i$(3) and yield (productivity) (4) are assumed to have a “trivariate” normal distribution, with zero mean and the covariance matrix $\sum _{\hspace{0.17em}}$ namely $\left(\epsilon ,{\mu }_1{\mu }_2\right) N0,\sum$.

With

(5) ${\sum }^{=}\left(\begin{array}{ccc}{\sigma }_{\epsilon }^2& .& .\\ {\sigma }_{\epsilon {\mu }_1}^2& {\sigma }_{{\mu }_1}^2& .\\ {\sigma }_{\epsilon {\mu }_2}^2& .& {\sigma }_{{\mu }_2}^2\end{array}\right)$(5)

Where ${\sigma }_{\epsilon }^2$ is the variance of the error in the selection equation (3), which can be assumed to be equal to 1 since the coefficients can only be estimated up to a scaling factor (Maddala, 1983), ${\sigma }_{{\mu }_1}^{2}and{\sigma }_{{\mu }_2}^2$ are the variances of the error terms in the efficiency functions (4a) and (4b), ${\sigma }_{{\mu }_1\epsilon }and{\mu }_{2\epsilon }$ represent the covariance of $\epsilon ,{\mu }_{1}and{\mu }_2$. Since Y_1i and Y_2i they are not observed simultaneously, the covariance between ${\mu }_{1}and{\mu }_2$ is not defined. An important implication of the error structure is that, given the error term in selection equation (3), ${\epsilon }_i$ is correlated with the error terms in the performance functions (4a) and (4b) (${\mu }_{1}and{\mu }_2$). The expected values ${\mu }_{1}and{\mu }_2$ of the sample selection conditions are zero:

(6) $\left\{\begin{array}{c}E\left[{\mu }_{1i}\left|A_i=1\right.\right]={\sigma }_{{\mu }_1\epsilon }\frac{\varphi \left(Z_i\alpha \right)}{\mathrm{\Phi }\left(Z_i\alpha \right)}={\sigma }_{{\mu }_1\epsilon }{\lambda }_{1i}\hspace{0.17em}\\ \mathrm{a}\mathrm{n}\mathrm{d}\\ E\left[{\mu }_{2i}\left|A_i=0\right.\right]={\sigma }_{{\mu }_2\epsilon }\frac{\varphi \left(Z_i\alpha \right)}{1-\mathrm{\Phi }\left(Z_i\alpha \right)}={\sigma }_{{\mu }_2\epsilon }{\lambda }_{2i}\end{array}\right.\hspace{0.17em}$(6)

Where $\varphi \left(.\right)$is the density function of the standard normal probability, $\mathrm{\Phi }(.)$ the normal cumulative density function, ${\lambda }_{1i}=\frac{\varphi \left(Z_i\alpha \right)}{\mathrm{\Phi }\left(Z_i\alpha \right)}\hspace{0.17em}$ and ${\lambda }_{2i}=\frac{\varphi \left(Z_i\alpha \right)}{1-\mathrm{\Phi }\left(Z_i\alpha \right)}$. If the estimated covariances ${\sigma }_{{\mu }_1\epsilon }et{\sigma }_{{\mu }_2\epsilon }$are statistically significant, then the decision to adopt agricultural innovations and the quantity of production are correlated. This significance would be evidence of endogenous switching and would allow rejection of the hypothesis of no sample selectivity bias. Hence the use of this regression model with endogenous switching, whose most efficient method of estimation is the maximum probable estimate with complete information (Lee & Trost, 1978). Thus, we have:

(7) $ln{L}_i=\sum _{i=1}^N{A}_i\left[ln\mathrm{\Phi }\left(\frac{{\mu }_{1i}}{{\sigma }_{{\mu }_1}}\right)-ln{\sigma }_{{\mu }_1}+ln\mathrm{\Phi }\left({\theta }_{1i}\right)\right]+\left(1-A_i\right)\left[ln\mathrm{\Phi }\left(\frac{{\mu }_{2i}}{{\sigma }_{{\mu }_2}}\right)-ln{\sigma }_{{\mu }_2}+ln\left(1-\mathrm{\Phi }\left({\theta }_{2i}\right)\right)\right]\hspace{0.17em}$(7)

Where ${\theta }_{ji}=\frac{\left(Z_i\alpha +{\rho }_j{\mu }_{ji}/{\sigma }_j\right)}{\sqrt{1-{{\rho }_j}^2}}$ and J = 1,2; with ${\rho }_1=\frac{{\sigma }_{{\mu }_1\epsilon }^2}{{\sigma }_{{\mu }_1}^{\hspace{0.17em}}{\sigma }_{\epsilon }^{\hspace{0.17em}}}$ and ${\rho }_2=\frac{{\sigma }_{{\mu }_2\epsilon }^2}{{\sigma }_{{\mu }_2}^{\hspace{0.17em}}{\sigma }_{\epsilon }^{\hspace{0.17em}}}$ the correlation coefficient between the error term of the selection equation (3) and the error terms ${\mu }_{ij}$ of equations (4a) and (4b) respectively. The importance of this regression model is that it allows post-estimate analyses to compare the expected production performance of farms that have adopted an agricultural innovation (a) versus farms that have not adopted an agricultural innovation (b). In addition, the expected output performance can be assessed for the hypothetical counterfactuals (c) for farms that adopted agricultural innovations in case they had not adopted them, and for farms that did not adopt innovations (d) in case they had adopted them. These conditional farm performance expectations for the four cases are presented in Table 1 and defined as follows:

(8) $\left\{\begin{array}{c}E\left(Y_{1i}/A_i=1\right)=X_{1i}{\beta }_1+{\sigma }_{{\mu }_1\epsilon }{\lambda }_{1i},(\mathrm{a})\\ E\left(Y_{2i}/A_i=0\right)=X_{2i}{\beta }_2+{\sigma }_{{\mu }_2\epsilon }{\lambda }_{2i},(\mathrm{b})\\ E\left(Y_{2i}/A_i=1\right)=X_{1i}{\beta }_2+{\sigma }_{{\mu }_2\epsilon }{\lambda }_{1i},(\mathrm{c})\\ E\left(Y_{1i}/A_i=0\right)=X_{2i}{\beta }_1+{\sigma }_{{\mu }_1\epsilon }{\lambda }_{2i},(\mathrm{d})\end{array}\right.$(8)

Table 1. Conditional income expectations, treatment effect and heterogeneity effects

Sub-sample	Stages of the decision
	Adopted innovation	No adopted innovation	treatment effect
Farmers adopted innovation	(a) $E\left(Y_{1i}\left|A_i=1\right.\right)$	(c) $E\left(Y_{2i}\left|A_i=1\right.\right)$	TT
Farmers no adopted innovation	(d) $E\left(Y_{1i}\left|A_i=0\right.\right)$	(b) $E\left(Y_{2i}\left|A_i=0\right.\right)$	TU
Heterogeneity effects	$B{H}_1$	$B{H}_2$	$TH$

Source : DiFalco et al. (2011)

Cases (a) and (b) along the diagonal in Table 1 represent the actual expectations observed in the sample. Cases (c) and (d) represent the expected outcomes of the counterfactuals. In addition, following Heckman (2001), we calculate the effect of the “adoption of agricultural innovations” treatment on the treaty (ATT) as the difference between (a) and (c), which represents the effect of the adoption of agricultural innovations on the farm performance of operators who actually adopted an innovation on their farm

(9) $TT=E\left(Y_{1i}\left|A_i=1\right.\right)-E\left(Y_{2i}\left|A_i=1\right.\right)={\mathrm{X}}_{1i}\left({\beta }_1-{\beta }_2\right)+\left({\sigma }_{{\mu }_1\epsilon }-{\sigma }_{{\mu }_2\epsilon }\right){\lambda }_{2i}\hspace{0.17em}$(9)

Similarly, it is necessary to calculate the effect of treatment on untreated farmers (UTFs) who have not adopted innovations for their farm as difference (d) and (b).

(10) $TU=E\left(Y_{1i}\left|A_i=0\right.\right)-E\left(Y_{2i}\left|A_i=0\right.\right)={\mathrm{X}}_{1i}\left({\beta }_1-{\beta }_2\right)+\left({\sigma }_{{\mu }_1\epsilon }-{\sigma }_{{\mu }_2\epsilon }\right){\lambda }_{2i}$(10)

3.4. Propensity score matching

Since the results from the endogenous switching regression model can be sensitive to the selection of assumptions of the instrumental variables, the propensity score matching (PSM) is used to assess the robustness of the treatment effects. ${\mathrm{Y}}_1$ Since the performance of farms that adopted the innovations ($\mathrm{A}=1$) and the ${\mathrm{Y}}_2$ performance of farms that did not adopt the innovations ($\mathrm{A}=0$) it is possible to observe the outcome variable of farms that adopted the innovations ($\mathrm{E}({\mathrm{Y}}_1|\mathrm{A}=1)$), but not the outcome variable of non-adopters if they had adopted the innovations ($\mathrm{E}({\mathrm{Y}}_2|\mathrm{A}=1)$). Hence the estimation of the ATT using equation (9) may be biased. Estimation by the PSM linked to three assumptions (conditional independence, existence of a common support and the unit value of treatment) allows for sharper treatment effects (Fougère, 2010; Khandker et al., 2009; Rosenbaum & Rubin, 1983). The conditional independence hypothesis assumes that the conditional probabilities between the outcome variable in the absence of innovation adoption (${\mathrm{Y}}_2$) and the innovation adoption status (A) are statistically independent and define the innovation adoption propensity score as follows:

(11) $\mathrm{A}\left(\mathrm{X}\right)=\mathrm{P}\mathrm{r}\left(\mathrm{A}=1\right)|\mathrm{X}$(11)

In addition, the application of this matching method is possible if there are non-innovative holdings with identical characteristics to the innovative holdings. Thus, the farms that are compared have the same probabilities of adopting or not adopting innovations such as $0<\mathrm{A}\left(\mathrm{X}\right)<1$. Compliance with these assumptions leads to the specification of the ATT estimator by SHP as follows

(12) $\mathrm{A}\mathrm{T}\mathrm{T}=\mathrm{E}\left\{{\mathrm{Y}}_1|\mathrm{A}=1,\mathrm{A}(\mathrm{X})\right\}-\mathrm{E}\left\{{\mathrm{Y}}_2|\mathrm{A}=1,\mathrm{A}(\mathrm{X})\right\}$(12)

The estimation of this equation is done in several steps: first the probability of innovation adoption is estimated using a probit model that allows the propensity scores for each farm to be estimated. Subsequently, each innovative farm is matched with a non-innovative farm with a similar propensity score to estimate the value of the ATT. Using this technique, the SHP compares the difference between the outcome variables of innovative and non-innovative farms with similar characteristics (Kamdem, 2018) et (Takam‐Fongang et al., 2019).

4. Econometric results

4.1. Statistical analysis of explanatory variables

Table 2 presents descriptive statistics related to the socio-economic and demographic characteristics of operators, some of which show a significant difference in the status of adoption of new farming practices. The analyses in this table show that the average productivity of farms in the Sudano-Saharan zone is 2,962.5 kg/ha. Farmers have an average cultivated area of 2.25 ha (9 quarters of a hectare). This statistic is in line with the INS analysis (2015) showing that 80% of farms in Cameroon have an average area of two hectares with low productivity resulting in low incomes. Rainfall variations and agricultural shocks may be at the origin of this low productivity. With a view to improving this productivity, farmers adopt strategies for adapting to shocks. The productivity of the adopters of these strategies is 4,311 kg/ha while that of non-adopters is 1,224 kg/ha. This implies a priori that these strategies have an effect on farm productivity. In addition, about 56% of farmers are adopting at least one new agricultural strategy/practice in response to agricultural shocks (flooding, late rains, drought, crop destruction by wildlife, rising prices of agricultural inputs, climate change, falling prices). Some innovative activities to adapt to shocks include strategies such as: crop diversification (77%), use of improved seeds (60%), construction of bunds (50%) and reduction of arable land (35%). The high level of adoption of agricultural innovation in this area of Cameroon is linked to farmers’ perception of shocks. The main reason for non-adopters is that they do not perceive or have no information on innovation activities to be implemented. On average, the majority of farmers surveyed are male (90%) versus female (10%). This representativeness of men in the agricultural sphere is justified by the fact that farms are mostly coordinated by male heads of household. This is in line with the legal point of view showing that only male heads of household have the right to own land.

Table 2. Descriptive statistics

Variables	Designation	Innovation adoption	No adoption	Together
		Average	Average	Average
Capital	CPT	19018.12(14040.15)	22753.65(13370)	20650.1(13867)
Work	TRL	10364.2(11686.15)	11236.84(11432)	10745.4(11576)
Productivity in kg	Production/hectare	4311.1(34875.6)	1224.3(4328.04)	2962.5(26357)
Innovation Adoption	1=yes 2=no	…	…	0.5631(0.4963)
Types of innovation adoption	Diversification of culture	…	…	0.7698(.4213)
	Construction of bunds	…	…	0.5035(0.5003)
	Improved seeds	…	…	0.6033(0.4895)
	Decrease of the cultivated area	…	…	0.3467(0.4763)
Area	In hectare	8.872(8.002)	8.714(6.361)	8.803(7.326)
Sex of operator	Male	0.9064(0.2916)	0.888(0.3147)	0.8987(0.3018)
Age of operator	Under 30 years old	0.1280(0.3346)	0.2095(0.4076)	0.1636(0.3702)
	[30–50 years old]	0.5467(0.4984)	0.5333(0.4997)	0.5409(0.4986)
	Over 50 years old	0.3251(0.4689)	0.2571(0.4377)	0.2954(0.4565)
Education level	Primary	0.2906(0.4546)	0.3841(0.4871)	0.3315(0.4711)
	Secondary	0.2389(0.4269)	0.2888(0.4539)	0.2607(0.4393)
	Superior	0.0345(0.1827)	0.0254(0.1575)	0.0305(0.1721)
Size of manage in age	[6–14 years old]	1.997(1.822)	1.311(1.692)	1.6976(1.798)
	[15–21 years old]	1.953(2.17284	1.60 (1.727)	1.8016(1.997)
	[22–30 years old]	1.505(10.883)	1.285(1.612)	0.409(0.772)
	[31–50 years old]	1.887(20.978)	1.8444(1.8523)	0.868(2.547)
	Over 50 years old	0.3842(0.7504)	0.4095(0.7362)	0.3953(0.7438)
State support and advice to farmers	Appui conseil	0.2167(0.4125)	0.1460(0.3537)	0.1858(0.3892)
Household income in thousand FCFA	Less than 50	0.3448(0.4759)	0.2095(0.4076)	0.2857(70.326)
	[50–100]	0.1453(0.3528)	0.1619(0.3689)	0.1525(0.3598)
	Over 100	0.3546(0.4790)	0.4349(0.4965)	0.3897(0.4880)
Method of obtaining the exploited parcel of land	Purchase	0.1527(0.3601)	0.2381(0.4266)	0.1900(0.3926)
	Rental	0.1305(0.3373)	0.2000(0.4006)	0.1608(0.3676)
	Others	0.1330(0.3399)	0.1079(0.3108)	0.1220(0.3276)
Livestock farm	Breeding	0.2980(0.4579)	0.3301(0.4710)	0.3120(0.4636)
Intention to improve the future production system	Adaptability	0.9384(0.2407)	0.8793(0.3262)	0.9126(0.2826)
Belonging to an operator group	Member of a group	0.2438(0.4299)	0.1650(0.3718)	0.2094(0.4072)
Information on agricultural vulnerability	agricultural vulnerability	0.6650(0.4725)	0.8254(0.3802)	0.7351(0.4416)
The changes observed in recent years	Flood	0.1575(0.3647)	0.2629(0.4409)	0.2034(0.4028)
	Increase Season length	0.7350(0.4418)	0.7013(0.4584)	0.7203(0.4491)
	Price increase input	0.4875(0.5004)	0.5065(0.4584)	0.4957(0.5003)
	Decrease in yield	0.9750(0.1563)	0.8441(0.3633)	0.9180(0.2744)
The changes observed in recent years	Agricultural shocks	0.9950(0.2400)	0.9746(0.1576)	0.9861(0.1170)

With regard to the level of education, more than half (62%) are educated, with a majority in primary (33%) and a minority in secondary (16%) and higher education (3%). Similarly, the majority of the operators surveyed are between 30 and 50 years old (54%) with a minority in the under 30 (16%) and over 50 (30%) age group. Regardless of the size of the household headed by these operators, their incomes vary from less than 50,000 FCFA (29%) to more than 100,000 FCFA (39%). With these generally low incomes, these farmers often benefit from advisory support programs (16%) for capacity building and also participate in farmers’ groups (21%) for information and agricultural opportunities. These low statistics are similar to those obtained by Roussy et al. (2015) on access to agricultural vulnerability (20.1%) and membership in farmers’ organizations (16.34%) in Niger. Admittedly, these statistics remain low compared to some studies conducted in Africa (Yabi et al., 2016), but participation in these professional groups enabled 73.51% of the farmers surveyed to easily obtain information on agricultural vulnerability.

Moreover, easy access to this information enabled 94% of adopters to intend to adapt and improve the system for their future production. This ambition to improve the future production system is linked to the change observed in the past. On average 20% of the farmers observed flooding, 72% noticed an increase in the length of the season, 49.6% noticed an increase in input prices, 92% experienced a decrease in yield. These components of the agricultural shock show significant differences between adopters and non-adopters. The effect of these earlier changes on farms allowed 98.6% of farmers to perceive that agricultural shocks are a reality that requires the necessary means of adaptation. As the main reason why innovation activities are not adopted, this perception of shocks presents a significant difference between adopters of new adaptation practices and non-adopters.

4.2. Determinants of farm innovation adoption

The estimation of the endogenous switching regression model, which takes into account both observed and unobserved factors, makes it possible to have not only the impact of the adoption of innovations on farm productivity, but also the determinants of this adoption of innovations. The third column of Table 3 presents the results of the innovation adoption equation representing the determinants of the adoption of new farm practices. This so-called selection function (innovation adoption) helps us explain why some operators adopt new operating practices and others do not.

Table 3. Estimating innovation adoption and its impact on farms

Variables	Designation	Selection	Innovation adoption	No adoption of innovation
		Coefficient	Coefficient	Coefficient
Travail	TRL	0.0167	0.0272**	−0.00856
		(0.0124)	(0.0125)	(0.0152)
Capital	CPT	−0.0773***	−0.0734***	−0.0299
		(0.0278)	(0.0229)	(0.0391)
Cultivated area	SUPER	−0.108	−0.229**	−0.254**
		(0.0882)	(0.0892)	(0.117)
Gender of operator	Male	0.0570	0.252	0.902***
		(0.179)	(0.178)	(0.204)
Age of operator	Under 30 years old	−0.0787	0.166	0.0622
		(0.174)	(0.185)	(0.199)
	Between 30 et 50 years old	0.0196	0.342***	−0.0553
		(0.124)	(0.122)	(0.150)
Education level	Primary	−0.266**	−0.0941	0.198
		(0.123)	(0.124)	(0.150)
	Secondary	−0.201	0.0663	0.0487
		(0.141)	(0.141)	(0.173)
Size of manage in age group	[6–14 years old]	0.103***	0.125***	−0.00428
		(0.0293)	(0.0306)	(0.0413)
	[15–21 years old]	−0.0132	0.0269	0.0482
		(0.0278)	(0.0270)	(0.0362)
	[22–30 years old]	0.0228	0.0340	0.0218
		(0.0318)	(0.0301)	(0.0397)
	[31–50 years old]	0.00871	0.0912***	0.00858
		(0.0214)	(0.0190)	(0.0342)
State support and advice to farmers	Appui conseil	0.0259	−0.0684	0.0735
		(0.217)	(0.208)	(0.266)
Household income in thousand FCFA	Less than 50	0.0170	−0.524***	−0.0664
		(0.170)	(0.169)	(0.214)
	[50–100]	0.0136	−0.103	0.191
		(0.178)	(0.185)	(0.202)
	over 100	0.0358	−0.255*	0.122
		(0.144)	(0.149)	(0.165)
Method of obtaining the exploited parcel of land	Purchase	−0.139	0.304**	0.0455
		(0.139)	(0.149)	(0.157)
	Rental	−0.182	0.0791	−0.238
		(0.143)	(0.157)	(0.162)
Intention to improve the future production system	Adaptability	0.513***	0.123	0.253
		(0.193)	(0.213)	(0.198)
Member of an agricultural group	Member of an agricultural group	0.270	0.136	−0.440*
		(0.212)	(0.202)	(0.256)
Animal husbandry as a secondary activity	Breeding	−0.225*	−0.285**	0.160
		(0.116)	(0.118)	(0.140)
Information on agricultural vulnerabilit	Agricultural vulnérabilité	−0.562***	−0.0762	1.735***
		(0.150)	(0.137)	(0.231)
The changes observed in recent year	Flood	0.0315	−0.296**	−0.280*
		(0.135)	(0.147)	(0.149)
	Increasing the length of the season	0.106	0.155	0.405***
		(0.127)	(0.124)	(0.151)
	Rising input prices	0.306***	−0.0443	−0.175
		(0.116)	(0.126)	(0.134)
	Lower yields	0.901***	0.215	−0.325
		(0.221)	(0.300)	(0.211)
Perception of agricultural shocks	Agricultural shocks	1.015***	…	… .
		(0.382)
Constant	Constant	−1.310**	6.066***	4.069***
		(0.566)	(0.479)	(0.568)
Sigma	σ	…	1.063*** (0.0572)	1.092*** (0.1147)
Rho	ρ	…	0.7807 (0.0548)	−0.7319 (0.1412)
LR independence test Observation		16.57*** 721

Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.

The method of obtaining the farmed parcel, approximated by purchase or lease, negatively affects the propensity to adopt new farming practices. The low negative significance of the rental coefficient implies that the household that rents a plot of land pays the rent to the owner. The payment of this rent does not motivate the farmer to make sustainable investments in terms of new farming practices. The perception of production risks has a positive impact on the propensity to adopt new farming methods. This statistically significant at the 1% threshold implies that farmers intend to improve production systems for the future. Especially since the economic situation imposes on them a trade liberalism that they would have to be productive and competitive to cope with. Moreover, the coefficient of family size has a strong influence on the adoption of innovation on farms in the Sudano-Saharan zone. This effect is much greater when the majority of the farm household is made up of children under the age of 14. The strong positive significance at the 1% threshold of the effect of this household size bracket on the probability of innovation adoption suggests that farm households with large and young families (between 6 and 14 years old) are more likely to adopt new farming practices in response to agricultural shocks. With a large family, the farmer hopes to improve future yields to meet the family’s basic needs. In addition, younger people constitute the main labour force for farm households.

Information on agricultural vulnerability has a strong statistically significant and negative influence on the probability of innovation. This implies that farmers who do not have information on the future vulnerability of their farm are not motivated to adopt new practices (Table 3). Conversely, this shows that obtaining such information would increase the probability of adopting new farming practices to anticipate future shocks. This analysis is consistent with the findings of Bryan et al. (2010) on Ethiopia and of Khanal et al. (2018) on Nepal. In addition, household characteristics such as gender, age, education level, income, animal husbandry and membership in a farmers’ group influence the adoption of new practices. While some authors (Luiz Piva et al., 2013) find a very large effect of these variables, our results do not reveal a convincing impact of these variables on the adoption of agricultural innovations. This result corroborates with that of Huang et al. (2015) in China, Khanal et al. (2018) in Nepal and Issoufou et al. (2017) in Niger.

The instrumental variable (perception of agricultural shocks) has a positive influence on the adoption of innovations. Whether on the adoption of innovation in general or on the adoption of each new production strategy (Table A1 in the appendix), this statistical significance at the 1% threshold suggests that farmers who perceive agricultural shocks are more likely to adopt new operating practices. This result is justified by the fact that the perception of agricultural shocks motivates farmers to adopt coping mechanisms and may also increase the possibilities of future adaptation.

However, this instrumental variable is validated by three statistical tests. First, a homogeneity test on the pre-treatment characteristics of farms that do or do not perceive agricultural shocks is carried out. From the latter, it is noted that the sample is not unbalanced with respect to the observable variables. In addition, the results of the Fisher (F) statistic test show the significance at the 1% threshold of the instrumental variable in the selection model. This implies that the presence of this instrument in the model cannot be neglected. Given that this instrumental variable concerns the perception of agricultural shocks and that the selection variable is the adoption of agricultural innovations in response to these shocks, the impact of this instrumental variable on productivity should only be intuitive through its impact on the adoption of agricultural innovations. Furthermore, to show that there is no direct impact of the instrumental variable (IV) on farm productivity, a falsification test is performed. The falsification test determines whether the IV does not directly affect farm productivity, but has an indirect effect on productivity through its effect on the adoption of agricultural innovations. To do so, the productivity of farms that have not adopted new strategies is regressed on the IV with all other variables in the model. The insignificance of the IV coefficient suggests that there is no direct impact of IV on agricultural productivity.

5. Discussion

5.1. Farm productivity linked to innovation adoption

Columns 4 and 5 of Table 3 present the estimated coefficients of the productivity function for farmers who have adopted or not adopted the new farming practices. These estimates take into account the endogenous change in the farm productivity function. The correlation coefficients (p) are not significantly different from zero, implying that the hypothesis of no sample selectivity bias in the adoption of innovations cannot be rejected (DiFalco et al., 2011; Huang et al., 2015; Khanal et al., 2018). Since the correlation coefficient of innovation adoption is positive is significantly different from zero, suggesting that farmers who choose to adopt the new farming practices will be more productive than a random farmer in the sample, and those who do not adopt any agricultural innovations will not do better or worse in terms of productivity than a random farmer (Table 3 and Appendix Table A1). Furthermore, the likelihood ratio test (last row of Table 3 and Table A1 in the appendix) shows that there is a joint independence of the three analytical equations, which further justifies the use of this model. However, the differences in the coefficients of the farm productivity equation between adopters and non-adopters of innovations show the presence of heterogeneity in the sample. As the auxiliary parameters used in the maximum likelihood procedure, sigmas (σ) are the square roots of the variances of the residuals in the regression part of the model. Taking agricultural productivity per hectare as the outcome variable, analysis of the comparisons shows that there are positive differences between farms that adopt new farming practices and their non-adopting counterparts. This result is consistent with the theoretical view that inputs such as labour, farm characteristics and the economic environment are significantly associated with the productivity of farmers who adopt new practices to adapt to agricultural shocks.

Thus, the work approximated by the number of workers has an influence of 2.72% on the productivity of farms that have implemented an innovation. This effect remains negative and insignificant for farms that have not innovated. This implies that the adoption of innovations requires a skilled labour force, which has a significant impact on productivity. Similarly, farmers between 30 and 50 years of age contributed 34.2% of their farm’s output. This shows that the age of the farmer is a performance factor. As a result of all the failures and successes accumulated, the farmer is more aware of agricultural shocks and can anticipate the adoption of new adaptation practices that will have a significant effect on productivity.

In addition, the number of persons in the household, whose age is between 6 and 14 and between 31 and 50 years of age, affects the productivity of farms that have adopted agricultural innovations at 12.5% and 9.12% respectively. This is justified by the fact that the youngest generally follow their parents in their activities and the older ones, through endurance and life experience, contribute more to their parents’ activities while hoping for a future inheritance. The 15–21 age group remains insignificant. This age group is the transition between adolescence and maturity which is generally very disturbed by puberty, the symptoms of which push children to disengage from their parents’ activities or to want freedom. The method of obtaining the arable plot by purchase has a 30.4% impact on agricultural productivity. This statistical significance at the 5% threshold can be explained by the fact that the purchase of agricultural land stimulates the farmer’s confidence to invest better by adopting the agricultural innovations necessary to improve productivity. Agricultural shocks such as floods have a negative and statistically significant effect at the 5% and 10% thresholds respectively for adopters and nonadopters. These shocks generally linked to climate change have negative implications on farm production as shown (DiFalco et al., 2011).

Table 4 presents farm operator productivity for both adopters and non-adopters of innovation. Cells (a) and (b) represent the expected productivity of the farms observed in the sample. This productivity for innovation adopters is 1,274.44 kg and 1,162.1 kg for innovation non-adopters. This indicates that innovation adopters produced on average 9% (112.34 kg) more than no adopters of innovations. The fourth column of Table 4 presents the average treatment effects that show the effect of innovation adoption on farm productivity in the Sudano-Saharan zone of Cameroon. These treatment effects explain the selection bias resulting from the probability that adopters and non-adopters of agricultural innovations are systematically different (Abdulai & Huffman, 2014) Khanal et al. (2018). Cell (c) represents farm productivity per hectare for adopters of innovations if they had not adopted and cell (d) represents farm productivity per hectare for no adopters of innovations if they had adopted. Farms that actually adopted the new farming practices would have produced about 646.44 kg/ha less if they had not adopted the agricultural innovations.

Table 4. Impact of innovation adoption on farm productivity

Type of innovation	Designation	Innovation adoption	No-Adoption Of innovation	Treatment effect
Innovation adoption	Adoption	(a) 1274.44(790.62)	(c) 628(145.03)	ATT 646.44(563.54)
	No-adoption	(d) 1586.86(118.50)	(b) 1162.1(791.85)	ATU 424.72(309.89)
Types of agricultural innovations
Diversification of culture	Adoption	(a) 1154.61(616.15)	(c) 66.98(295.4)	ATT 1087.63(558.3)
	No-adoption	(d) 1594.8(1949.3)	(b) 587.96(75.93)	ATU 1006.84(313.7)
Construction of bunds	Adoption	(a) 1217.6(931.9)	(c) 179.44(145.69)	ATT 1038.16(697.2)
	No-adoption	(d) 1282.9(963.7)	(b) 245.57(135.96)	ATU 1037.33(439.9)
Diminution of surfaces area	Adoption	(a) 1265.03(897.1)	(c) 367.12(394.4)	ATT 897.91(751.65)
	No-adoption	(d) 1336.06(655.45)	(b) 1219.8(1061.4)	ATU 116.26(749.56)
User of improved seeds	Adoption	(a) 1133.25(695.2)	(c) 75.2(67.7)	ATT 1058.05(684.8)
	No-adoption	(d) 1812.3(781.6)	(b) 1459.02 (1417.6)	ATU 353.28(375.8)

Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%; (a) = adopters; (b) = no-adopters; (c) = adopters if they had not adopted; (d) = no-adopters if they had adopted; ATT = treatment effect (innovation adoption effect); ATU = adoption effect for farms that did not adopt an innovation.

In addition, farms that did not actually adopt an innovation would have produced about 424.72 kg/ha more if they had adopted it. These results imply that the adoption of new agricultural practices in the Sudano-Saharan zone of Cameroon, through the perception of agricultural shocks, considerably increases farm productivity. By observing the effect of each type of innovation, it can be seen that the effect is 1087 kg/ha for crop diversification, 1038 kg/ha when farmers build bunds to control runoff water in the irrigated area, 898 kg/ha for the strategy of reducing the cultivated area and 1058 kg/ha for the use of improved seeds (Table 4). These results are similar to those of Takam-Fongang et al. (2019) on maize production in Cameroon and Khanal et al. (2018) on rice production in Nepal showing that the adoption of improved seeds significantly increases farm yield. In fact, improved seeds are selected for their ability to produce crops with higher yields. They are often developed to resist disease, pests and harsh environmental conditions, enabling farmers to obtain more abundant harvests. These improved seeds, often developed to adapt to specific growing conditions (different soils, extreme temperatures or variable rainfall), enable farmers to grow crops in areas where they might not otherwise thrive, increasing overall agricultural productivity.

However, the transitional heterogeneity effect (TT—UT = 221.72 kg) is positive overall, meaning that the effect of adopting agricultural innovations is very large for those farms that have actually adopted the new practices for their farms than those that have not. This heterogeneity effect remains positive for each type of new strategy adopted, which shows that farms that have adopted a new production strategy tend to produce more than farms that have not adopted anything. Thus, the adoption of new farming practices would protect farmers against the risk of crop failure. These innovative farming practices enable farmers to increase the efficiency of their work, reduce losses and optimize production. For example, improved seeds that are more resistant to climatic variability produce higher yields and enable farmers to obtain more abundant, better-quality harvests.

The adoption of innovative practices in terms of reducing the cultivated area has a positive effect on agricultural productivity. In fact, with the vagaries of climate change, farmers are reducing cultivated areas, leading to a reduction in the availability of agricultural land and limiting their ability to produce crops in sufficient quantities to meet growing food demand. This reduction in cultivated areas leads to an intensification of the exploitation of the remaining lands, which leads to an increase in production in the short term, and can to a certain extent contribute to a reduction in productivity in the long term due to the degradation of the floors. The Sudano-Saharan zone being a wrinkled zone, the adoption of improved seeds is done according to their ability to resist diseases, pests and difficult environmental conditions (different soils, extreme temperatures or variable precipitation), which allows farmers to obtain crops with higher yields.

5.2. Robustness and comparative discussion

Factors explaining the adoption of innovations are examined by the probit model for estimating propensity scores. The results of this model are presented in Table A2 in the Appendix. The pseudo R² is 13.2%, the LR (chi2) of 127.74 significant at the 1% threshold showing that the model is globally adjusted and at least one of the explanatory variables explains the decision to adopt innovations on farms. Following these propensity scores, we performed matching by closest neighbours (Rubin (2002); Kassie et al., (2011); Bekele et al., (2014); Kamdem (2018)). Figure A1 shows the reduction of standardised biases before and after matching, while Figure A2 shows the common support of matched farms. According to Table A3 in the appendix, there are significant differences between farms that adopt agricultural innovations and those that do not. If no difference is observed before and after pairing between the means of the explanatory variables of farms that adopted innovations, there are large differences before and after pairing between the means of the explanatory variables of non-adopters. This raises the presumption of selection bias, whose co-matching between adopters and non-adopters allowed a reduction on average from 17.2% before pairing to 5.3% after pairing (Table A4 in the appendix). The total bias is thus reduced from 88% to 34% through the matching process. Furthermore, the p-values (0.000) of the maximum likelihood indicate the model’s significance after matching, while the pseudo-R² indicates the level of performance of the probit model. This pseudo-R² is reduced from 13.2% before matching to about 2.1% after matching confirming that after matching, there is no significant difference in the distribution of the two sub-populations (adopters and nonadopters). The low level of pseudo-R² (0.021), the low level of mean bias (5.3), the high level of bias reduction (34) and the non-significance of the maximum likelihood after matching indicate that the specification of the propensity score estimation process is successful in balancing the distribution of covariances between farms that adopted innovations and those that did not. This analysis joins that of Shiferaw et al. (2014), Khonje et al. (2015) and Kamdem (2018) on the case of Cameroon.

Thus, the results of estimating the average effects of agricultural innovation adoption estimated by Kernel-based matching methods and the nearest neighbour method are presented in Table 5. These results show that the adoption of new agricultural production strategies has a positive and significant impact on farm productivity in the Sudano-Saharan zone of Cameroon. This effect is respectively 104 Kg/ha by the nearest neighbour method and 281 kg/ha by the Kernel method. Regardless of the type of innovation adopted, the average effect of the two MMP techniques is significantly positive and similar to that of the “endogenous switching regression” model; which confirms the positive effect of the adoption of new production strategies on farm productivity. This result is in line with that of DiFalco et al. (2011); Khanal et al., (2018); Quan et al., (2019); Takam‐Fongang et al. (2019) showing the positive effect of the adoption of new strategies on agricultural productivity.

Table 5. Average effect of innovation adoption on farm productivity by propensity score matching method

Variables	Farm productivity	Innovation adoption	No-adoption of innovation	Treatment effect ATT
Innovation adoption	Nearest Neighbor	1114,8	1010,75	104,03(116,48)***
	Kernel	1114,8	833,51	281,29(114,08)***
Diversification of culture	Nearest Neighbor	1071,6	810,16	261,43(157,34)***
	Kernel	1071,6	1008,63	63(127,67)***
Construction of bunds	Nearest Neighbor	1059,4	884,71	186,88(119,48)***
	Kernel	1044,7	996,6	48,1(113,56)***
Decrease in surface area	Nearest Neighbor	1113,95	780,63	333,321(114,22)***
	Kernel	1113,95	878,006	235,944(126,87)***
Use of improved seeds	Nearest Neighbor	1045,26	1069,33	24,066(122,15)***
	Kernel	1045,91	1020,67	21,24 (101,97)***

Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%;

In the face of agricultural shocks, farmers are adopting new production practices such as: crop diversification, use of improved seeds, construction of bunds and reduction of cultivated areas to minimize risk. Crop diversification enhances the sustainability of the agricultural production system by reducing inputs, increasing heterogeneity, reducing yield losses due to frequent returns of the same species and thereby increasing farm productivity. Through continuous improvements and selection processes over time, improved seeds enhance the genetic potential of crop species and thus improve farm productivity. The adoption of the dyke technique (earth, stone, grass, etc.) to control runoff and sedimentation improves soil structure, water infiltration, water-soil-plant relationships and the process of plant mineralization, thus contributing to improved farm productivity. This explains the significant changes in the productivity of innovative farms compared to non-innovative ones estimated at 104 kg/ha (i.e. an increase of 9.33%) by the nearest neighbour technique and at 218.29 kg/ha (i.e. an increase of 25.23%) by the Kernel technique. The Sudano-Saharan zone of Cameroon has only a short rainy season, crop diversification allows farmers to diversify their sources of income. By growing different crops, they can take advantage of seasonal price variations and minimize the risks associated with a single crop. For example, if one crop is affected by a long dry season, income from other crops can offset losses. This can help stabilize farm income and improve the financial security of farmers. In addition, water being a limited resource in this region, the construction of bunds has a positive effect on agricultural productivity. These bunds built to retain rainwater and prevent its rapid flow, allow groundwater to be recharged, form water reservoirs and optimize the use of this precious resource. By conserving water in fields, bunds allow crops to be irrigated during periods of drought, promoting better plant growth and increased productivity. At times, these bunds prevent flooding and thus allow farmers to maintain their crops in good health and avoid significant losses. These analyses are in line with those of Chuma et al. (2022) and Karume et al. (2022) assessing the effect of agricultural innovations on the mitigation of climatic hazards on farms.

6. Conclusion

The objective of this paper was to assess the impact of the adoption of agricultural innovations on farm productivity in the Sudano-Saharan zone of Cameroon. To do so, the study compared the productivity of farms that adopt new agricultural practices and those that do not, based on survey data from 721 farms in the Sudano-Sahelian zone of Cameroon. These data are applied by the “endogenous switching regression” model, which takes into account both endogeneity and selection in the sample, and the propensity score method to verify the robustness of the results.

Analyses of this model show that the average productivity of farms that adopt new farming practices is significantly higher than that of farms that do not adopt these innovations. The effect of these farm innovation adoptions is approximately 646.44 kg/ha compared to those that did not adopt any innovations. This implies that the new agricultural practices (crop diversification, construction of crop bunds, use of improved seeds, reduction of arable land) adopted autonomously on farms play an important role in maintaining and increasing productivity in the context of agricultural shocks. The study also reveals that the perception of agricultural shocks, whether through the personal observation of the farmer or through agricultural information divulged on the radio, by extension workers or agricultural groups, plays an important role in the decision of farmers to adopt new agricultural practices. This shows the importance of awareness raising and capacity building activities initiated by the State and ONG in the Sudano-Saharan zone of Cameroon.

The study shows that about 57% of farmers have adopted at least one innovation to address agricultural shocks. This innovation adoption has a positive and statistically significant contribution to farm productivity. This shows that farmers are to some extent aware of agricultural shocks and have the knowledge and skills relating to the different types of innovation needed to maintain and increase productivity. Moreover, it appears that about 43% of farms in the Sudano-Saharan zone of Cameroon have not adopted any new agricultural practices despite the tenacity of agricultural shocks. As Cameroon’s Sahelian zone is characterized by arid climatic conditions, the government would benefit from implementing policies to promote adaptation to climatic constraints. This could be achieved through the development of water conservation techniques, the adoption of innovative and resilient agricultural practices, the promotion of crops adapted to climatic conditions and the sustainable management of natural resources. This policy should be accompanied by the promotion of irrigation, crop diversification, agricultural input extension and capacity-building policies and agricultural technical support adapted to the conditions of the Saharan zone.

As there are several types of agricultural innovation that still have a very significant effect on farm performance, future research may focus on the complementarity between these different types of agricultural innovations and their effect on farm productivity.

Acknowledgments

We would like to express our deep gratitude to AFRICALICS for its support through its “Africalis/Globelics Mentoring in Innovation and Development” program which aims to help mentees become more technically proficient in research and development. We also thank our mentor professor MANIR Abdullahi, Professor Fomba Kamga Bejamin and the anonymous reviewers who reviewed the paper and provided comments and suggestions that helped improve the overall quality of this paper. This paper was presented at the 17th International Globelics Conference under the theme “Innovation systems and sustainable development: new strategies for growth, social welfare and environmental sustainability”, November 3-5, 2021, Heredia, Costa Rica. We also thank the “Groupement Interpatronat du Cameroun (GICAM)” for its technical support during our consultation stay in 2023.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/23311886.2023.2282419

Additional information

Notes on contributors

André Dumas Tsambou

André Dumas Tsambou He is the principal investigator who will coordinate this research. He holds a Ph.D. in Economics and is a professor at the Faculty of Economics and Management of the University of Yaoundé II. As the principal investigator of this study, he has extensive experience in conducting large-scale projects, advocacy and translating research findings into economic policy proposals. In addition, he is the coordinator of the Observatory of Competition and Competitiveness of the inter-patronal group of Cameroon in terms of distortions of competition, which carries out economic and legal monitoring, sensitization and training of entrepreneurs, and carries out advocacy notes for institutional support to improve the competition environment in Cameroon. He has conducted several studies with the African Economic Research Consortium (AERC) and the International Fund for Agricultural Development (IFAD) as a senior researcher. Some of the studies conducted are on: ”Innovation Behavior and Productivity of Firms in Francophone Sub-Saharan Africa‘, ’Climate Change, Health and Gender Inequality in Human Capital Investment in Rural Senegal‘, ’Impact of Digital Technology Adoption on Employment in Senegal‘, ’Programs to Support Youth Employment and Employability in Growth Sectors in Senegal ”. He also has several publications in scientific journals, the most recent of which are on the adoption of innovations and productivity of firms in French-speaking Sub-Saharan Africa published in the journal of industrial economics and the adoption of environmental protection policies and performance of Cameroonian firms published in the Canadian Journal of Agronomy.

Notes

1. The Agro-system programme is an IRAD project that aims at the design, evaluation and dissemination of efficient and sustainable agro-systems in rural areas of the Far North of Cameroon. These actions will contribute to the sustainable improvement of the productivity of small polyculture farms in North Cameroon by preserving soil fertility through adapted and efficient technical proposals proposed to development structures.