Integrated Rainwater Harvesting Practices for Poverty Reduction Under Climate Change: Micro-Evidence from Ethiopia

Abstract

Rainwater harvesting (RWH) has been practiced and promoted to address the temporal and spatial variability of rainfall, thereby enhancing agriculture production in rainfed systems. The practices could also address the problem of land degradation. However, there is limited practice of approaching RWH from the perspective of managing both the water and land resources. Research on water productivity and impact for poverty reduction mainly focused on irrigated agriculture while it is potential to provide the water needed to produce food for rapidly growing population is the subject of intense debate these days. An important option is to upgrade rainfed agriculture through better land and water management that improves soil moisture conservation and rainwater harvesting that provides supplementary irrigation. In the meantime, studies on impact of agricultural water management focused more on unidimensional poverty while poverty is multidimensional. This study investigates the impact of integrated RWH practices (IRWHPs) on multidimensional poverty in Ethiopia. Results show that the use of IRWHPs has a significant negative impact on the probability that a household is multidimensionally poor. This study suggests that policies that enhance the promotion of IRWHPs would be central for the sustainable intensification of smallholder agriculture that simultaneously alleviate poverty and enhance resource sustainability.

Appendices

Appendix 1 Estimation of Endogenous Switching Probit Model

We consider three-equation models that describe a regime determination rule and two outcome equations to measure the determinants of use and impact of IRWHP for poverty reduction.

The regime determination rule is the decision whether to use IRWHP or not (R_i). A farmer will decide to use/adopt IRWHP if the expected benefit/utility from adopting is higher than from not using these practices. The decision rule is given as:

$$R_{i} = \left\{ {\begin{array}{*{20}l} 1 \hfill & {{\text{if}}\;\gamma Z_{i} + \mu_{i} > 0} \hfill \\ 0 \hfill & {\text{otherwise}} \hfill \\ \end{array} } \right.$$

(1)

The outcome equations ($Y_{i}$), the probability of multidimensional food security is defined:

$${\text{Regime }}1:Y_{1i}^{*} = X_{1i} \beta_{1} + \sigma_{1\mu } \hat{\lambda }_{1i} + \varepsilon_{1i} \quad {\text{if}}\;R_{i} = 1\;\;({\text{IRWHP users}})$$

(2)

$${\text{Regime }}2:Y_{0i}^{*} = X_{0i} \beta_{0} + \sigma_{0\mu } + \varepsilon_{0i} \quad {\text{if}}\;R_{i} = 0\;\;({\text{IRWHP non - users}})$$

(3)

where $Y_{1i}^{*}$ and $Y_{0i}^{*}$ are the latent variables for the observed binary outcomes Y₁ and Y₀; $Z_{i}$ and $X_{i}$ are vectors of observable factors influencing the decision to use or not to use IRWHP and the probability of multidimensional food security and poverty status. $\lambda_{1i} = \tfrac{{\varphi \left( {Z_{i} \alpha } \right)}}{{\Phi \left( {Z_{i} \alpha } \right)}}$ and $\hat{\lambda }_{0i} = \tfrac{{\varphi \left( {Z_{i} \alpha } \right)}}{{1 -\Phi \left( {Z_{i} \alpha } \right)}}$ are the inverse Mills ratios (IMR) computed from the selection equation (Eq. 1) to correct for selection bias in a two-step estimation procedure, i.e., endogenous switching regression. $\beta$ and $\sigma$ are parameters to be estimated, and $\varepsilon$ is an independently and identically distributed error term. The error terms ($\mu_{i} ,\varepsilon_{1i}$ and $\varepsilon_{0i}$) are assumed to be jointly normally distributed with mean-zero and correlation matrix expressed as:

$$\text{cov} (\mu ,\varepsilon_{1} ,\varepsilon_{0} ) = \left[ {\begin{array}{*{20}c} 1 & {\rho_{u0} } & {\rho_{u1} } \\ {} & 1 & {\rho_{ 1 0} } \\ {} & {} & 1 \\ \end{array} } \right]$$

(4)

where $\rho_{u0}$ and $\rho_{u1}$ are, respectively, the correlations between the error terms of selection (adoption) Eq. 1 and outcome Eqs. 2 and 3; $\rho_{10}$ is the correlation between $\varepsilon_{1}$ and $\varepsilon_{0}$. If the off-diagonal elements are nonzero, then the error terms of the adoption and the outcome equations are correlated, indicating evidence of endogenous switching or presence of sample selection bias. As we do not have a credible instrument, we base identification on functional forms. As argued by Lokshin and Glinskaya (2009), the system of Eqs. (1–3) is identified by nonlinearities even if the variables in $X$ and $Z$ overlap completely.

Conditional Expectations and Treatment Effects

We use the predicted probabilities from Eqs. (2) and (3) to compare the average treatment effect of IRWHP on users and non-users of the practice. Following Di Falco et al. (2011) and Kassie et al. (2015), the conditional expectations, treatment, and heterogeneity effects of the probability of multidimensional food secure and/or poor are defined as follows:

$$E\left( {\left. {Y_{i1} } \right|R = 1;x} \right) = x_{i1} \beta_{1} + \sigma_{1\varepsilon } \hat{\lambda }_{i1} \;({\text{Actual expected outcomes for users of IRWHP}})$$

(4a)

$$E\left( {\left. {Y_{i0} } \right|R = 0;x} \right) = x_{i0} \beta_{0} + \sigma_{0\varepsilon } \hat{\lambda }_{i0} \;({\text{Actual expected outcomes for non - users of IRWHP}})$$

(4b)

$$E\left( {\left. {Y_{i0} } \right|R = 1;x} \right) = x_{i1} \beta_{0} + \sigma_{0\varepsilon } \hat{\lambda }_{i1} \;({\text{Counterfactual expected outcomes if users had not used IRWHP)}}$$

(4c)

$$E\left( {\left. {Y_{i1} } \right|R = 0;x} \right) = x_{i0} \beta_{1} + \sigma_{1\varepsilon } \hat{\lambda }_{i0} \;({\text{Counterfactual expected outcomes if non - users had used IRWHP)}}$$

(4d)

Equations (4a) and (4b) represent the actual expected outcomes observed in the sample, while Eqs. (4c) and (4d) are the counterfactual expected outcomes. The counterfactual is the expected outcomes if the characteristics $(x_{1} )$ of users had the returns of the characteristics $(\beta_{0} )$ of non-users and vice versa. The effect of the IRWHP on household welfare is calculated using these conditional expectations. The expected change in the probability of multidimensional food secure and poor due to the use of IRWHP can be then calculated as the difference between Eqs. (4a) and (4c). These estimates are called as average treatment effect (ATT) in the impact assessment literature and are given as:

$${\text{ATT}} = \left( {\left. {E(Y_{i1} } \right|R = 1,x} \right) - \left. {E(Y_{i0} } \right|R = 1,x) = x_{i1} (\beta_{1} - \beta_{0} ) + \lambda_{1i} (\sigma_{1u} - \sigma_{0u} )$$

(5)

Similarly, we calculate the effect of IRWHP use decision for the non-user of the practice as the difference between Eqs. (4d) and (4b) and is given as average treatment effects on the untreated (ATU):

$${\text{ATU}} = \left( {\left. {E(Y_{i1} } \right|R = 0,x} \right) - \left. {E(Y_{i0} } \right|R = 0,x) = x_{i0} (\beta_{1} - \beta_{0} ) + \lambda_{0i} (\sigma_{1u} - \sigma_{0u} )$$

(6)