The primary goal of this study was to conduct a comprehensive analysis of the adoption of environmentally sustainable practices among rural households within the region, while simultaneously investigating the determinants shaping the intensity of their adoption. The study was conducted in Boukoumbe, Kouande and Toucountouna. These are municipalities where most households are farmers and partake in the combination of both farm and livestock production. These municipalities enjoy a Sudanian climate with two seasons (a rainy and a dry season). The rainy season starts from June to October while the dry season is from early November to May with an average of 1199.96 millimeters of rainfall yearly (Mouzoun, 2018). The issue of irregular rainfall in the communities has a serious effect on the farming systems.
For this study rural households of the three municipalities were of interest. So, a simple random sampling technique was used for the selection of three districts in each municipality. Simple random sampling was also used to select nine (27) rural communities (villages) within the three districts. These were: Dikounmini, Dimansouri, Dipoli, Abgonte, Koukongou, Koupagou, Didompe, Koukouankouangou, Koussoucoingou, Birni-Maro, Birni-Pebirou, Gorgoba, Chabi-Couma, Gantieco, Papatia, Danri, Fo-Tance, Kabare, Bouyagnindi, Kouarfa, Kouba, Dikokore, Kokota, Tampegre, Tectibayaou, Touvountouna I, and Boribansifa. To ensure reliable representation, allocation of sample sizes was done disproportionately to the population sizes of each municipality, based on the following formular:
Where: Sm is the sample size of the given municipality to be calculated, P is the population of the selected municipality; T: the sum of the population of the municipalities (Toukountouna, Kouande, Boukoumbe); and E is the desired total sample size. Toucountouna has being designated 17.01 percent, Boukombe 35.26 percent and Kouande 47.67 percent.
A total 405 rural household heads were interviewed for this study. Questionnaire administration, interviews, and observation were used to collect data in the study area, on the rural households’ assets and understanding of the ESPs.
2.1. Analytical techniques
Descriptive statistics and econometrics were used in achieving the objectives of this research. And so, the descriptive statistics includes Likert Scale to assess rural households’ knowledge on ESPs. Econometric techniques were also used to assess the determinants of the adoption of ESPs, the interdependence of various ESPs, and adoption’s intensity in the study area. Assessing the determinants of ESPs adoption can be done through Tobit, probit, logit and multinomial regressions. And so, scholars e.g., Kassie et al. (2015); cited in Ehiakpor et al. (2021) have undertaken empirical investigations into environmental sustainability practices among rural households. They found out that these models, fail to include important information such as interrelationships amongst technologies. In order words these methodologies fall short in fully capturing the interplay between the adopted practices. To address this gap, the study used the Multivariate Probit model (MPV) to simultaneously estimate both the determinants of adoption and the correlations among environmentally sustainable practices (ESPs), presenting a more holistic perspective on their interdependencies (i.e.: the correlations among ESPs). This constitutes the first approach. Concurrently, the second econometric technique focused on analyzing the intensity of ESP adoption in rural households. This involved the estimation of the Poisson regression model, to assess the intensity of the adopted ESPs among rural households
2.1.1. Determinants of ESP Adoption: The Multivariate Probit Model
Benin citizens confronts environmental challenges such as soil degradation and water scarcity. To combat these issues, adopting ESPs is pivotal. Within the context of rural areas, ESPs encompass techniques including water-efficient irrigation, nutrient management, pest control, sustainable land use practices among others.
The adoption of these practices does not only foster environmental well-being but also enhances productivity. In this study adoption is defined as: the process by which rural households actively and consistently engage in the implementation and the utilization of environmentally sustainable agricultural practices, as identified by the FAO in 1995 (Bruinsma, 2017). Adoption implies that households have deliberately chosen to include these practices into their activities in order to improve productivity as well as environmental well-being. Measuring the adoption of environmentally sustainable practices involves the collection of data reflecting the extent to which these have been incorporated into their rural activities. For this purpose, data can be collected through (1) questionnaire (focusing on whether respondents adopt these practices), (2) observations (by visually assessing the presence and utilization of ESPs on farm lands) and (3) interviews (to gather qualitative data about their motivation, challenges and benefits regarding adoption of the ESPs). These can later be triangulated to cross-validate the reliability of the adoption measurements.
Amidst the backdrop of deforestation and land degradation, agricultural producers are compelled to migrate towards more fertile soils for cultivation. Addressing this pressing environmental concern, the FAO (1995) defines ESPs as land management practices that are economically viable, ecologically friendly, socially acceptable, technically appropriate, non-degrading, resource-conserving, and preservative (Bruinsma, 2017). The literature in Africa, and notably in Benin (Hounnou et al., 2023; Houessou et al., 2022; Hounnou et al., 2019; Yabi et al., 2016), identifies several ESPs among rural households. These include but are not limited to water-efficient irrigation, integrated pest management (IPM), organic fertilization, reduced tillage, drip irrigation, native or drought-resistant varieties, conservation tillage, biopesticides, precision farming, agrobiodiversity conservation, crop rotation, composting, soil erosion menace control, nutrient cycling through green manure, and agroforestry practices. These techniques, endorsed by rural households, align with environmentally friendly principles and are geared towards improving agricultural production.
For analysis, focus will be on the eight most frequently adopted ESPs among rural households, while also engaging key informants to understand the low adoption rates of remaining technologies. However, all identified technologies will be modelled. The interplay between the determinants influencing the adoption of these selected ESPs will be explored using the Multivariate Probit Model. This model captures the influence of independent variables on the listed practices while accommodating the interdependence of error terms, both measurable and unobservable (Kassie et al., 2015a). In contrast to univariate models, the multivariate probit model demonstrates greater efficiency by considering the correlations between unobserved and unmeasured elements in ESP adoption studies. Neglecting these correlations, as seen in univariate probit models, leads to biased and inefficient estimates (Kassie et al., 2015a). The proposed multivariate probit model is as follows:
\({Y}_{ij}^{\text{*}}={Z}_{ij} {\alpha }_{ij} + {\in }_{ij}\) (with j = 1,2,3,4,5,6,7,8) (6)
$${Y}_{ij} =\left\{\begin{array}{c}1 if {Y}_{ij}^{\text{*}} >0\\ 0 if otherwhise\end{array}\right.$$
7
As presented in the model the environmentally sustainable practices (ESPs) will be captured with j = 1,2,3,4,5,6,7,8
In Eq. (1), \({Y}_{ij}^{\text{*}}\) designates the predicted variable of the environmental sustainability practices (ESPs) being used by the rural households of the region. It encodes both observed and unobserved factors that influence the adoption of a jth choice of ESP.
The explanatory vector’s variable in the equation is \({Z}_{ij}\). This variable is the one that influences the preferences and choices of a resident as to adopt or not a particular ESP practice. \({\alpha }_{ij}\) represents vector of parameters expected to be estimated. The error term also known as the unobserved disturbance in this equation is \({\in }_{ij}\). Eqs. (1) and (2) constitute specifically univariate probit models which does not indicate whether multiple ESP practices have been adopted by a rural resident. These models only model information on an ith rural resident adopting a given practice which does not necessarily influence the rural dweller’s probability of choosing another ESP practices. To be able to incorporate multiple ESP adopted practices, the assumption would be that the error terms in Eq. (1) would conform to a multivariate normal distribution concurrently, that is:
$$\left({\epsilon }_{i1}{\epsilon }_{i2}{\epsilon }_{i3}{\epsilon }_{i4}{\epsilon }_{i5}{\epsilon }_{i6}{\epsilon }_{i7}{\epsilon }_{i8}\right) ̴ MVN=\begin{array}{cccccccc}1& \rho 21& \rho 31& \rho 41& \rho 51& \rho 61& \rho 71& \rho 81\\ \rho 12& 1& \rho 32& \rho 42& \rho 52& \rho 62& \rho 72& \rho 82\\ \rho 13& \rho 23& 1& \rho 43& \rho 53& \rho 63& \rho 73& \rho 83\\ \rho 14& \rho 24& \rho 34& 1& \rho 54& \rho 64& \rho 74& \rho 84\\ \rho 15& \rho 25& \rho 35& \rho 45& 1& \rho 65& \rho 75& \rho 85\\ \rho 16& \rho 26& \rho 36& \rho 46& \rho 56& 1& \rho 76& \rho 86\\ \rho 17& \rho 27& \rho 37& \rho 47& \rho 57& \rho 67& 1& \rho 87\\ \rho 18& \rho 28& \rho 38& \rho 48& \rho 58& \rho 68& \rho 78& 1\end{array}$$
3
In this Eq. (3) the \(\rho\) denotes the error term coefficient in the estimation of any two of the ESP adoption’ equation. The Equation will later become an MVP model when there is correlation between two error terms. The sign of the \(\rho\) in this model delivers an important information as regards to whether there is a negative correlation between the two components or not. And so, a negative correlation will suggest sustainability between the ESP adopted practices. A positive correlation on the other hand will suggest that a given ESP practice may depend on another one, meaning complementarity between the two components.
2.1.2. Multivariate Probit Model for Interconnected Determinants
The analysis employed the Multivariate Probit Model to unravel the intricate relationships between determinants impacting the adoption of various Environmental Sustainability Practices (ESPs). This model is designed to accommodate the effects of independent variables on multiple practices, all while addressing associations between both observable and latent error terms. The efficiency of the Multivariate Probit Model shines through its capability to unearth correlations that exist between unobservable and unmeasured factors within ESP adoption studies. This consideration is essential to avoid skewed and inefficient estimations that can arise from neglecting these crucial interdependencies.
Capturing Adoption Intensity through Count Data Models
The study's objective is to quantify the intensity of ESP practice adoption within the region, specifically focusing on the number of ESP practices implemented by rural households. Count data models, extensively used for this purpose, have been previously explored by pioneering researchers such as Greene (2008) and Trivedi (2013). These models concentrate on adoption intensity, utilizing techniques like the Poisson Regression model, which is tailored to situations where the dependent variable signifies the count of a particular event or behavior. The Poisson Regression model operates on the assumption of equi-dispersion, implying that the conditional mean and variance are statistically equivalent.
The Poisson Regression Model
The Poisson Regression Model proposed by Greene (1997) is represented as followed:
\({P}_{r}(Y=y) =\frac{{e}^{-\lambda } {\lambda }^{y}}{y!}\) , y = 0, 1, 2 (1)
The assumption here is that the parameter \(\lambda\) is to be log-linearly as relates to the regressors (Xi, ). and so,
$$ln \left(\lambda \right)={\beta }^{1}{x}_{i}$$
2
This leads to the log-likelihood function given by:
$$ln L =\sum i=\text{1,2}, ...n [{-\lambda }_{i}+{y}_{i}{\beta }^{1}{x}_{i} - {lny}_{i}! ]$$
3
The following equation was then used to estimate the expected number of adopted practices by an ith individual.
$$E \left[{y}_{i}{Ix}_{i}\right]=var \left[{y}_{i}{Ix}_{i}\right] =exp [{\beta }^{1}{x}_{i} +{\mu }_{i} ]$$
4
Here, X = k*1 is a vector of independent variables, while β = k*1 is a vector of parameters. Thus the equation can be rewritten as:
$$E \left[{Y}_{i}\right]={exp }^{{\beta }_{1} {X}_{1i}} {exp }^{{\beta }_{2} {X}_{2i}}............ {exp }^{{\beta }_{k} {x}_{ki}}=exp \left[{\beta }_{i}{X}_{jn}\right] {C}_{i}$$
5
Here
j : [1, k] (can take a value from 1 to k)
\({C}_{i}\) is a constant. \({C}_{i}\)is the product of the outstanding exponential terms in Eq. 5.13
In the case of the dichotomous explanatory variable,
$${if, X}_{ji}=0,{E[Y}_{1}]={C}_{i} and when {X}_{ji}=1,{E[Y}_{1}]={{B}_{j}C}_{i}$$
Therefore,
Hence, the expression 100 x (exp(βj) − 1) computes the percentage change in the expected value of Y when the variable xj transitions from zero to one, across all observations (i). In a broader context, for independent variables assuming multiple integer values, the incremental change in the expected adoption level of improved technologies, as xj shifts from xj1 to xj2, can be determined as:
$$\frac{dy}{dx}=\frac{{exp}^{{\beta }_{jxj2}}- {exp}^{{\beta }_{jxji}} }{{exp}^{{\beta }_{jxji}}}$$
6
In this research, improved adopted practices was modelled. The empirical model aims at examining the rural household’s characteristics that influence their decision to adopt the improved adopted practices. The covariates are: age, gender, education, farming exp, household size, farm size, livestock, pests, constant erosion, flooding occurrence, contact ext. agents, dist. to-market, workshop, attendance, credit access.