2. Methods
2.1 Study area descriptions
Dhanusha district, which is in the southern part of Nepal and shares a border with India, was selected for this study. About 60% of the land comes under agriculture out of the total area of the district, 119000 ha. Data were collected from May through August 2015. Like other parts of Nepal, agriculture is the major economy of the district where about 90% of people are actively engaged in the cultivation of wheat, rice, and sugarcane (Dhakal et al, 2015). After the state federalization, the district now falls in the province no. 2. Located approximately 95 m above the sea level, the district is one of the hottest districts of Nepal with the average annual rainfall being 2199 mm. The meteorological data shows that April is the warmest month with the average temperature being 39.60 C (maximum) while January is the coldest with the average temperature of 21.40 C (maximum) (CBS, 2012).
Administratively the district consists of one sub-metropolitan city, eleven urban municipalities and six rural municipalities. The Terai Private Forest Development Association (TPFDA), a local NGO, has worked to promote a tree-based farming practice in then nine Village Development Committees (VDCs11Now, VDCs are a part of either urban or rural municipalities after restructuring the state.) covering 10,500 hectares (Figure 1). Therefore, these nine VDCs were selected as the study site. After the state is restructured, some parts of the study site fall in the urban municipality while most parts are still VDCs, now known as rural municipalities.
Figure 1: Study Area
2.2 Household survey
A two-stage sampling approach was adopted for this study. First, one ward22Ward is the lowest administrative unit. from each VDC was selected through purposive sampling. This means a total of nine wards were selected. Second, thirty-two households from each ward were selected randomly. This means 288 sample households were selected. In-person interviews were conducted with the head of the sample households using a structured questionnaire.
The questionnaire covered the household data- the demography and socio-economic conditions of the sampled households. Also included in the questionnaire were the institutional and biophysical variables which we hypothesized as adoption determinants and constraints. A total of 18 households were dropped out of the analysis since these households were practicing a combination of two or more agroforestry practices, agroforest/woodlot, boundary plantation and alley cropping.
2.3 Analytical model
Logistic regression model is the best fit when the outcome (dependent) variables are unordered and categorical. When there are only two outcome variables, the binary choice model (binary regression model) is the best fit. In our case, the outcome variable is different types of farming practices adopted by the study area farmers. Since we have more than two farming practices i.e. (i) agroforest/woodlot system (AFS), (ii) alley cropping system (ACS), (iii) combination of two or more AF practices, and (iv) conventional agricultural system (CAS), the binary choice model is not suitable. Out of the four practices, the third practice was dropped off from the analysis since the practice was very rarely practiced in the study area. Since we still have more than two practices even after dropping one practice off, we chose Multinomial Regression Model (MRM) for our analysis. Most commonly used multinomial regression models are probit model and logit model. We chose the Multinomial logit (MNL) model for our study since it gives more precise parameter estimation (Kropko, 2007). The MNL model estimates the likelihood of adoption of non-reference categories against a reference (base) category in terms of relative risk ratio (RRR) (Miheretu and Yimer, 2017). The other reason for choosing this model is that this has been more commonly used in recent studies (Lin et al., 2014; Luus et al., 2015; Paton et al., 2014; Miheretu and Yimer, 2017). Having three farming practices in place, farmers can choose the one they prefer the most from the three alternatives. That means their choice is mutually exclusive.
We assumed farmers follow the random utility theory, while making the choice out of the three farming practices available. Therefore, we used a random utility model while determining the farmers’ choice of farming practices, as given by Greene (2003).
\begin{equation} Y_{\text{ij}}=\beta_{j}X_{\text{ij}}+\varepsilon_{\text{ij}}\ldots\ldots\ldots\ldots\ldots\ldots.\ (1)\nonumber \\ \end{equation}
where Yij denotes the utility of farmer iobtained from farming choice j , Xijdenotes all the factors affecting farmers’ decision to adopt a farming practice j and βj is the parameter that reflects changes on Yij due to changes inXij. We assumed the error terms to have an independent and identical distribution (iid) (Cheng and Long, 2007). According to profit maximisation, farmer i will, thus, only choose a specific alternative j if Yij> Yik for all k ≠ j . This choice ofj depends on a number of predictor (independent) variables as denoted by Xij in the above equation. IfYi is a random choice that a farmer can make, the MNL model can be expressed as:
\(\text{Prob\ }\left(Yi=j\right)=\frac{e^{\beta_{j}x_{i}}}{\sum_{j=1}^{j}e^{\beta_{j}x_{\text{i\ }}}}\ldots\ldots\ldots\ldots\ldots(2)\)j = 0, 1, 2,…….., j
The above equation estimates probabilities for j+1 farming choices i.e. three practices for farmers with a number of independent variables, Xij. Here, we are to estimate the probabilities of two non-reference farming practices, agroforest system and Alley cropping system against the reference category i.e. conventional agriculture and this can be done by assuming β0 = 0 and expressed as follows:
\begin{equation} \text{Prob}\ \left(\text{Yi}=j\right)=\frac{e^{\beta_{j}x_{i}}}{1+\sum_{j=1}^{j}e^{\beta_{j}x_{i\ }}}\ldots\ldots\ldots\ldots\ldots(3)\nonumber \\ \end{equation}\begin{equation} \text{Prob}\ \left(\text{Yi}=0\right)=\frac{e^{\beta_{j}x_{i}}}{1+\sum_{j=1}^{j}e^{\beta_{j}x_{i\ }}}\ldots\ldots\ldots\ldots\ldots(4)\nonumber \\ \end{equation}
2.4 Variables defined
The dependent variable is the adoption of farming practices by farmers as denoted byYi . For MNL model, the outcome (dependent) variable was denoted as:
Yi = 0 if a household adopts conventional agriculture system (CAS) -reference category- (j = 0);
Yi =1 if a household adopts agroforest system (AFS)- non-reference category- (j = 1);
Yi = 2 if a household adopts alley cropping system (ACS)- non-reference category-( (j = 2).
Before the model is run, all the hypothesized independent variables were tested for multicollinearity using the variance inflation factor (VIF). We found the VIFs of the independent variables below 10 (1.09– 2.03), indicating there is no issue of multicollinearity.
The estimation of the MNL model for this study was undertaken by selecting CAS as the base category. The odds of two other farming systems namely AFS and ACS against the CAS were estimated in this study. Since the CAS was the base category, it was hypothesized that most predictor variables will positively impact the adoption of the tree-based farming practices i.e. one unit increase in an“ independent variable will increase the likelihood of AFS and ACS adoption.
2.5 Variables used in the model
The three farming practices were the dependent variables out of which farmers chose the one they preferred the most. We extensively reviewed the contemporary literature on adoption to identify and determine independent (explanatory) variables. The explanatory variables included socio-economic, biophysical and institutional characteristics (Table 1). However, some variables were excluded in the model. The variable ‘farmers’ perception on agroforestry’ was dropped off the model because studies suggest it had no relationship with adoption (Alavalapati et al., 1995; Anley et al., 2007; Carlson et al., 1994; Thangata and Alavalapati, 2003) and also the methodological challenge we faced to precisely measure the perception made us drop this variable off the model (Roberts et al., 1999). The ‘slope gradient’ is another variable we ignored because of little altitudinal variation across the sampled households. The third variable ‘access to credit facility’ was also excluded because of no financial guarantee from the financial institutions for agroforestry promotion in the study area.