Access to irrigation water-poverty nexus: Application of an

Endogenous Switching Regression in Ethiopia
 

F.A. Adela¹; J. Aurbacher²

 

1: Justus-Leibig University, Institute of Agribusiness Management and Food Economics,

Germany, 2: Justus-Leibig University, Institute of Agribusiness Management and Food
Economics, Germany
Corresponding author email: [email protected]
Abstract:
The lack of consensus on the role of the agricultural sector in poverty reduction and pitfalls in impact study
methodologies resulted in mixed findings on impact of irrigation. This study explores factors that determine
farmer’s decision to irrigate and whether access to irrigation water enhances livelihood of the farmers.
Cross-sectional data from a survey of 240 smallholder farmers in Wondo Genet, Ethiopia was used for the
analysis. Foster-Greer-Thorbecke indices indicated high poverty level among farmers without access to
irrigation. Further analysis was undertaken using an endogenous switching regression (ESR) model. The
correlation coefficient results proved the existence self-selection and endogeneity. Accordingly, variables
like scheme governance, level of water scarcity, and access to network found to be some of variables that
significantly affected the farmers’ decision to irrigate. Model estimates further indicated that access to
irrigation resulted in better life conditions when compared to counterfactual situation. Farm income of the
households has increased by 107% and 171% for irrigation users and non-users, respectively. Similarly,
per adult equivalent consumption expenditure has shown increase by 26% and 57% for irrigation users
and non-users. Key Words: Agriculture, Irrigation, smallholder, Wondo Genet, Ethiopia, Endogenous
Switching Regression, poverty, Farm income, Consumption
Acknowledegment:
JEL Codes: D01, O13
#1503 

 
Access to irrigation water-poverty nexus: Application of an Endogenous
Switching Regression in Ethiopia

Abstract
The lack of consensus on the role of the agricultural sector in poverty reduction and pitfalls in
impact study methodologies resulted in mixed findings on impact of irrigation. This study
explores factors that determine farmer’s decision to irrigate and whether access to irrigation
water enhances livelihood of the farmers. Cross-sectional data from a survey of 240 smallholder
farmers in Wondo Genet, Ethiopia was used for the analysis. Foster-Greer-Thorbecke indices
indicated high poverty level among farmers without access to irrigation. Further analysis was
undertaken using an endogenous switching regression (ESR) model. The correlation coefficient
results proved the existence self-selection and endogeneity. Accordingly, variables like scheme
governance, level of water scarcity, and access to network found to be some of variables that
significantly affected the farmers’ decision to irrigate. Model estimates further indicated that
access to irrigation resulted in better life conditions when compared to counterfactual situation.
Farm income of the households has increased by 107% and 171% for irrigation users and non-
users, respectively. Similarly, per adult equivalent consumption expenditure has shown increase
by 26% and 57% for irrigation users and non-users.

Key Words: Agriculture, Irrigation, smallholder, Wondo Genet, Ethiopia, Endogenous Switching
Regression, poverty, Farm income, Consumption

1
 1. Introduction
Agricultural water is usually the major constraint in agricultural production in developing
countries (Namara et al., 2010; Singh et al., 2009). And the lion’s share of the water poor
population lives in already poor regions of the world like South Asia and SSA. Moreover, the
average land holding size in developing countries is continually declining and is less than two
hectares (Singh et al., 2009). Hence, it has been indicated by several studies in the SSA in
general and Ethiopia in particular that, growth in food production only through area expansion is
no more possible. It should be complemented with adopting productivity enhancing alternatives.
However, use of productivity enhancing techniques and inputs is effective only when there is
enough agricultural water availability (Gebregziabher, Namara, & Holden, 2009).

Irrigation in general and small scale traditional irrigation in particular has been practiced in
Ethiopia since ancient times (Aberra, 2004; Bacha, et al., 2011). The country has a potentially
irrigable land of 5.3 million hectares out of which only 0.7 million hectares is currently under
irrigation (Awulachew & Ayana, 2011). Since recently, there has been implementation of
projects to develop and renovate significant number of small scale irrigation schemes with the
aim of improving the livelihood of the rural poor. Studies made to examine whether such
development of irrigation schemes are serving their purpose in the country are limited(Bacha et
al., 2011).

Impact studies exhibit difficulties because of the involved theoretical and empirical
complications. Studies so far used several approaches depending on the data availability and
objectives. These methods range from two-stage least square method (Amare et al., 2012;
Bravo-Ureta et al., 2006; Hanjra et al., 2009), Propensity Score Matching (PSM) (Kassie,
Shiferaw, & Muricho, 2010), and to econometric modeling like Heckman’s selectivity model
(Bacha et al., 2011). Econometric models including Heckman’s selection model, endogenous and
exogenous switching models were applied at different times to deal with drawbacks of impact
studies especially relating to selectivity bias and issues of endogeneity (Dutoit, 2007). A study
by Bacha et al., (2011) applied Heckman’s selectivity model using two stage estimation
technique to examine impact of small scale irrigation on household poverty in Ethiopia.
Simultaneous estimation of endogenous switching regression model using full information
maximum likelihood technique is currently gaining popularity because most research problems

2
involve endogenous switching and exhibit self-selection. Other studies undertaken in SSA used
endogenous switching regression to study impacts of soil and water conservation technology
(Abdulai & Huffman, 2014), adoption of improved maize varieties (Khonje, Manda, Alene, &
Kassie, 2015), effects of modern agricultural technology (Asfaw, 2010).

This study aims at examining factors that affect irrigation decision and the impact of irrigation
on welfare of small holder farm households in Wondo Genet, Ethiopia using ESR. It differs from
other studies which dealt with irrigation adoption and impact by the methodology and techniques
employed while those studies which used the same methods consider adoption of other
technologies.

2. The study area and Survey Design

2.1 The study area
Wondo Genet is located about 263 Km south of Addis Ababa in the south central rift valley of
Ethiopia. Geographically it is situated between 701’N38035’E latitude and 7.0170N 38.5830E
longitude. It covers an area with altitude ranging from 1600 to 2580 meters above sea level. The
mean annual temperature is between170 and 190. The climate is characterized as sub-humid with
bi-modal rainfall distribution. It receives an annual average rainfall of 1079.7 mm with
February– April being low rainfall months and June to September the main rainy season.

Small holder perennial crop farming dominates the agriculture in the area; the major crops being
enset, khat, and sugarcane (Dessie & Kinlund, 2008). Water from two major rivers -Worka and
Wosha- is used for irrigation during dry seasons based on allocation schedule set by water user
association committee members. The WUA lacks legal framework and mentioned for unfair
distribution of water. Mode of irrigation is only furrow and related institutional and
infrastructural arrangements are weak. Only part of the major canal which is used to divert the
river from the head work is constructed with concrete and cement lined; 1.648 km is lined out of
4.6 km long scheme. The rest of the water way is earthen canal dug by the farmers themselves.
Local markets are relatively not far compared to other areas in the region; the average distance of
the local market being 5.4 km away from the study area. The market systems, however, are
underdeveloped and most farmers sell outputs on farm with involvement of brokers. Access to
formal credit is very limited.

3
2.2 Data

2.2.3 Study design

Data from household survey of 240 small holder farmers located in Wotera Kechema Kebele
(Peasant Association) pertaining to 2014/15 cropping season was used for this study. Initially,
the Woreda (district) was selected purposively for its relative long years of practice of irrigation
and accessibility. There are two major rivers (Worka and Wosha) on which the majority of
irrigation is undertaken at the moment. Out of the two rivers the irrigation scheme on Worka
River is relatively old and covers wider area. It is also the scheme with high level of reported
scarcity as indicated by the Woreda Agriculture and Rural Development office. In terms of
coverage, it has a potential of irrigating a total area of 345 ha: 225 ha Wotera Kechema and 120
ha Wosha. Accordingly, Worka river irrigation scheme was selected for the study. A sampling
frame was prepared from the list of 742 farm households residing in Wotera Kechema Kebele in
three categories: Upstream, middle stream, and Downstream. Then from these categories 80
households each were selected using systematic random sampling.

The household survey was undertaken by five enumerators after undertaking an adequate level of
training and a pilot survey. The selected enumerators spoke the local language (Sidamu Afo)
fluently which helped to maintain the reliability of the collected data. A well designed and tested
questionnaire was used as an instrument for the survey. Additional data was also collected
through key informant interviews. Crop water use and river discharge data was collected by
experts in the field using the Area-Velocity method by float technique. Agro-ecological and
climate data for the study area is collected from the concerned offices and organization.

2.2.4. Descriptive summary

Table 1 presents descriptive statistics of the data for the relevant variables included in the
estimation of the model. For the descriptive analysis t-test was used to show the difference
between irrigation users and non-users with respect to relevant continuous variables. In addition,
Chi-square test was used to describe the binary variables.

4
Table 1 Descriptive summary of variables used in estimations
Item and unit Users Non- p-value
users
Household characteristics
Head male (1=male) 0.95 0.98 0.20[1.61]
Age of head (Year) 46.95 43.58 0.038**
Household size adult equivalent (#) 5.61 5.09 0.024***
Labor endowment in adult 3.86 3.26 0.003***
equivalent (#)
Education level completed by head 2.96 2.73 0.285
(years)
Highest education level completed 4.22 3.37 0.038**
by adult (years)
Farm Characteristics
Land holding per Plot (ha) 0.172 0.148 0.02***
Land covered by cash crop (%) 0.780 0.613 0.000***
Asset
Landholding (ha) 0.436 0.323 0.007***
Household asset value (ETB) 66708.67 26927.71 0.000***
Income and Consumption
Farm income per ha (ETB) 432506.60 101688.80 0.000***
Non-farm income (ETB) 3165.71 4033 0.230
Total per adult equivalent food 3964.75 2848.24 0.000***
consumption expenditure (ETB)
Access to market and modern technology
Distance of the nearest market 5.29 5.74 0.060**
(km)
Chemical fertilizer per ha (Kg) 929.47 621.81 0.000***
Insect/herbicide per ha (ml) 21.48 12.67 0.002***
Institutional and information access
related variables
Visit by extension officers (Yes=1) 0.88 0.87 0.776[0.08]
Level of scarcity (highly scarce=1) 0.97 0.50 0.000[52.41]***
Governance problem (Yes=1) 0.53 0.71 0.009[6.78]***
Own radio (Yes=1) 0.63 0.46 0.013[6.21]***
Own mobile phone (Yes=1) 0.662 0.35 0.000[21.12]***
***significant at 1% level, **significant at 5%level, t-values in square brackets
Source: own calculation from survey data
As it is presented on the table the irrigation users have significantly higher mean value for all of
the continuous variables except for few. As per the binary variables, the number of households
among the non-users who responded that there is high level of scarcity is significantly higher
than the user households. Similarly, households were asked several questions regarding the
current management and governance of the irrigation scheme on the river, how they are adapting

5
to the water scarcity problem they face, what they think is permanent solution to the problem,
and what should the concerned stakeholders do to solve the problem. Their responses to all the
questions were categorized in to households who indicated that there is governance problem and
those who do not. Accordingly, the number of households who indicated the existence of
governance problem is significantly higher among the non-user households than the users.
Ownership of radio and mobile phone is used to capture access to information and networking.

The descriptive statistic on Table 1, however, provides only the mean differences and cannot be
used to conclude about the factors that affect the farmer’s decision to irrigate or not. Further, the
above analysis does not account for other important unobservable characteristics of the
households. Results of further analysis using endogenous switching regression model will be
presented and discussed in section 5.

3. Conceptual Framework
3.1 Poverty
The concept of poverty has evolved from the original idea of inadequacy of income consumption
and wealth (O’Boyle, 1999; Watts, 1968) to Sen (1981) concept of capabilities and functioning
and further to include multidimensional aspects like socio-political rights, access to important
service and infrastructure, vulnerability etc. (Namara et al., 2010; Smith, 2004). Moreover,
absolute and relative poverty are commonly mentioned in poverty literatures. Absolute poverty
refers to the head count of households who are unable to afford certain standard of basic goods
and services. Relative poverty, on the other hand, measures the relative shortfall of a household's
income from the economy's average. Another concept related to, but wider than poverty is
equity. It refers to the level of equality in income and wealth distribution. Poverty is also
dynamic in that factors that affect poverty can change from time to time (Smith,
2004).Commonly, the Foster-Greer-Thorbecke (FGT) indices developed by Foster et al., (1984)
are used in poverty analysis.

3.2 Water poverty

Water poverty can relate to either physical or economic water scarcity. Most of the time poor
people do not have access to adequate quantity and quality of water because the water is
physically unavailable. In other cases people face water scarcity or they cannot access water

6
because of poor infrastructure, mismanagement, corruption etc. (Dudu & Chumi, 2008; Namara
et al., 2010; World Bank, 2016).

Water scarcity and incidence of poverty are not necessarily linked; access and control is more
crucial than endowment in several cases (Namara et al., 2010). One cannot deny, however, that
whatever the cause of poverty may be, increasing scarcity of and competition over water is major
challenge to poverty reduction efforts of any kind. It is highly likely that scarcity of water will
increase into the future mainly because of population growth, reallocation to competing uses like
industries, and climate change. This, in turn, more than proportionately affects the already poor
segment of the population (Rosegrant, Ringler, & Zhu, 2009).

3.3 Access to irrigation and poverty linkage

Studies show mixed results regarding the impact of investment in irrigation on poverty
alleviation. Jin et al., (2002) and Rosegrant & Evenson (1992), for instance, found no direct
relation between irrigation, productivity, and poverty reduction in Asia. Similar results were
indicated by studies in Ethiopia (Gebregziabher et al., 2009). Contrary to the above claims, other
studies indicate a positive linkage between irrigation, productivity, poverty reduction and food
security (Bacha et al., 2011; FAO, 2003; Hussain & Hanjra, 2004; Smith, 2004; Wichelns,
2014). This lack of agreement on the impact of irrigation on poverty reduction also follows on
the emerging debate on the impact of agriculture in growth and development as a whole
(Gebregziabher et al., 2009). The deductions made by this studies, however, are highly
influenced by the methodologies employed and the perspectives considered. Micro level studies
in this regard, mostly witnessed a strong and positive impact of irrigation on poverty reduction
unlike the macro level counterparts (Gebregziabher et al., 2009).

Irrigation impacts poverty reduction in a complex pathway including productivity, employment,

resilience and sustainability, consumption and nutrition, and indirect economic impacts on the
wider economy. For instance, the benefits of irrigation can be viewed from the perspective of its
impact on improved use of productivity enhancing inputs like fertilizer and improved varieties.
This use of productivity enhancing inputs in turn results in higher productivity and income to the
farmers. The wider economy can also benefit indirectly through backward linkage in the form of
income and employment (Gebregziabher et al., 2009; Namara et al., 2010; Smith, 2004). There
is, however, concern on the sustainability of the impacts and the long run scenarios. However,

7
‘with irrigation’ effort is for sure the most likely to help achieve poverty reduction and growth
objectives than the without scenario (Smith, 2004).

3.4 Issues in and methods of impact studies

The critical issue in impact study is acknowledging the potential biases. Most of the time two
sources of biases are mentioned. The first one relates to the possibility of significant difference
between the participants and non-participants due to observable farm and household
characteristics. These characteristics may have direct and significant impact on the outcome
variable. Secondly, unobservable factors like skill and attitude may result in difference among
households and may affect the behavior of the households towards deciding to participate.
Therefore it is crucial to recognize that difference between the participants and non-participants
may not be attributed only to the treatment but also to initial differences among them. Therefore,
the selected impact assessment model should either help to eliminate selection bias or be sound
enough to account for it (Bacha et al., 2011; Wooldridge, 2003; World Bank, 2010).

Several methods have been used so far in various impact studies. These methods differ in the
way they account for selection bias. Some of the methodologies, however, have major drawbacks
of ignoring the issue of self-selection and difference between adopters and non-adopters (Kassie
et al., 2010; World Bank, 2010). For instance, the simplest method would be using Ordinary
Least Square (OLS) through including the treatment as a dummy variable in the outcome
function. However, considering the systematic difference between users and non-users resulting
in unobserved selection bias, the results of OLS estimation are biased and inconsistent (Bacha et
al., 2011; Di Falco et al., 2011; World Bank, 2010). Such unobservable factors could not be
captured and cause correlation between the observed explanatory variables and the error term
(Abdulai & Huffman, 2014).

Another commonly used method is Propensity Score Matching (PSM). This method assesses
treatment effects between participants and matched individuals. The matching is undertaken only
on observed characteristics assuming that a selection bias occurs only due to observable
characteristics (World Bank, 2010). It, however, does not account for the possibility that there is
a latent variable that simultaneously influences selection and outcome (Ravallion, 2005).

8
Later econometric models like Heckman’s selection model, and endogenous and exogenous
switching models emerged. These models assume that the impact of explanatory variables is
different depending on which regime applies. There are basic differences on two crucial issues.
One relates to the concept that whether the regime is determined inside the model or outside.
Hence, the switching could be endogenous or exogenous. Secondly in some instances, both
regimes are observable while some others work with only one regime observed (Dutoit, 2007).
Comprehensive switching regression was considered by Goldfeld & Quandt (1972) with two
regimes. This switching regression model was exogenous because of the assumption made about
the error terms. The same exogenous switching regression model was extended by Goldfeld &
Quandt (1973) to simultaneous equation systems (Lee et al., 1982).

Maddala & Nelson (1975) extended the model to make it possible to deal with endogenous
switching. Studies that used Endogenous Switching Regression approached the modeling in two
stage estimation which requires cumbersome adjustment to produce consistent standard errors
(Kassie et al., 2010; Khonje et al., 2015). Full Information Maximum Likelihood technique to
simultaneously estimate Endogenous Switching Regression models are suggested as the most
efficient in this regard (Di Falco et al., 2011; Kassie et al., 2010; Lokshin & Sajaia, 2004).

3.5. Theoretical Model Specification

The decision to adopt a technology can be modeled in a random utility framework by expressing
the unobservable utility from adoption and non-adoption through observable variables (Khonje et
al., 2015). Accordingly, use of irrigation is modeled considering the assumption that small holder
farmers choose between irrigating and not irrigating. It is assumed that the farmers consider the
benefit from irrigation through the farm income derived from crop production to decide to
irrigate. The following model specifies the selection equation P* where P* is the latent variable
which is not observed. P* can, however, be expressed as a function of some observed farm,
household and institutional characteristics.

𝑃∗ = 𝛼𝑍𝑖 + 𝑢𝑖

𝐼𝑖 = 1 𝑖𝑓𝑃∗ > 0 𝑎𝑛𝑑 𝐼𝑖 = 0 𝑖𝑓 𝑃∗ ≤ 0 1

9
Ii is a binary variable which takes a value of 1 for farmers who irrigate and 0 for those who do
not irrigate. Zi represents factors that affect the irrigation decision. α denotes the vector of
parameters indicating the magnitude and direction of each explanatory variable’ s effect on the
decision to irrigate. The residual ui captures the unobserved factors and measurement errors.

The two regimes that the small holder farmers fall in to are represented by the following two
regression equations.

𝑅𝑒𝑔𝑖𝑚𝑒 1: 𝑌1𝑖 = 𝛽1 𝑋𝑖 + 𝜀1𝑖 𝑖𝑓𝐼𝑖 = 1 2𝑎

𝑅𝑒𝑔𝑖𝑚𝑒 2: 𝑌2𝑖 = 𝛽2 𝑋𝑖 + 𝜀2𝑖 𝑖𝑓 𝐼𝑖 = 0 2𝑏

Y1i and Y2i are the dependent outcome variables determined by the exogenous variables Xi, β1,
and β2, are parameters that show the direction and strength of the relation between the outcome
variable and the independent variables. 𝜀1𝑖 and 𝜀1𝑖 are error terms.

Several approaches are available for use in estimating the endogenous switching model. Two
step least square or maximum likelihood estimation can be used through estimating one equation
at a time (Lokshin & Sajaia, 2004). These approaches, however, are mentioned to be inefficient
and resulting in heteroskedastic residuals in that they need ‘cumbersome adjustments’ to drive
consistent standard errors (Abdulai & Huffman, 2014). This drawback can be tackled by
estimating the model using the Full Information Maximum Likelihood (FIML) technique.

4. Empirical Model Specification

Farmers decide to irrigate if they assume that the net benefits in the form of farm income from
irrigating is higher than that of not irrigating. Several types of unobservable factors also
determine the farmers’ decision to irrigate resulting in a selection bias. A selection bias arises if
unobservable factors affect both error terms in the selection equation (𝑢𝑖 ) and the outcome
equation (ε). This results in a correlation between the error terms of the selection and continuous
equation: corr (ε,𝑢𝑖 ) = ρ≠0. This correlation between the error terms witnesses the existence of
an endogenous switching (Maddala, 1986).

The unobservable factors may fall under personal, social or institutional characteristics. They can
include natural managerial and technical skills, the farmer to farmer networks and informal
10
associations to formal institutions like water user associations. They can also include transaction
costs incurred by the farmers because of poor infrastructure (Abdulai & Huffman, 2014).
Provided that different farm and farmers’ characteristics determine whether the farm household
decides to irrigate or not the following specification gives the outcome regression equations for
the two regimes:

𝑢𝑠𝑒𝑟𝑠: 𝑌1𝑖 = 𝛽1 𝑋𝑖 + 𝜀1𝑖 𝑖𝑓𝐼𝑖 = 1 3𝑎

𝑛𝑜𝑛 − 𝑢𝑠𝑒𝑟𝑠: 𝑌2𝑖 = 𝛽2 𝑋𝑖 + 𝜀2𝑖 𝑖𝑓 𝐼𝑖 = 0 3𝑏

Assume that the error terms ε1i, ε2i, and 𝑢𝑖 have a trivariate normal distribution, with mean vector
zero and covariance matrix (Lee et al., 1982),

𝜎𝑢2 . .
2 2
Cov(𝑢𝑖,𝜀1𝑖,𝜀2𝑖)= [𝜎𝜀1𝑢 𝜎𝜖1 . ] 4
2 2
𝜎𝜖2𝑢 . 𝜎𝜖2

2 2
Where 𝜎𝑢2 variance of the error term in the selection equation, 𝜎𝜖1 and 𝜎𝜀2 are variances of
2 2
the error terms in the continuous equations. 𝜎𝜀1𝑢 and 𝜎𝜖2𝑢 are covariance of 𝑢𝑖 and ε1i and ε2i
respectively. Since Y1i and Y2i are not observed simultaneously a covariance of the corresponding
error terms is not defined (Maddala, 1983). This structure of the error terms indicates that the
error terms of the outcome equation and the error term of the selection equation are correlated
which results in non-zero expected value of 𝜀1𝑖 and 𝜀2𝑖 given 𝑢𝑖 - error term of the selection
equation (Abdulai & Huffman, 2014). Therefore, the expected values of the truncated error terms
𝐸(𝜀1 | 𝐼 = 1) and 𝐸(𝜀2 | 𝐼 = 0) are given below:

𝐸(𝜀1 | 𝐼 = 1) = 𝐸(𝜀1 | 𝑢 > −𝑍𝛼)

𝑍𝛼
𝜑( 𝜎 )
= 𝜎𝜀1𝑢 ≡ 𝜎𝜀1𝑢 𝜆1 5𝑎
𝑍𝛼
Φ( 𝜎 )

11
And,

𝐸(𝜀 2 | 𝐼 = 0) = 𝐸(𝜀2 | 𝑢 ≤ −𝑍𝛼)

𝑍𝛼
−𝜑( 𝜎 )
= 𝜎𝜀2𝑢 ≡ 𝜎𝜀2𝑢 𝜆2 5𝑏
𝑍𝛼
1 − Φ( 𝜎 )

φ and Φ are the probability density and cumulative distribution function of the standard normal
distribution, respectively. The ratio of φ and Φ evaluated at Zα is referred to as the inverse Mills
2 2
ratio 𝜆1 and 𝜆2 (selectivity terms). If the estimated covariance 𝜎𝜀1𝑢 and 𝜎𝜀2𝑢 are significantly
different from 0 the decision to irrigate and the outcome variable (farm income) are correlated.
This implies endogenous switching and the presence of a sample selectivity bias (Maddala, 1986;
Maddala & Nelson, 1975).

Where 𝜌1 and 𝜌2 are correlation coefficients between the selection equation error term 𝑢𝑖 and
the error terms of the outcome equations 𝜀1 and 𝜀2 . Further, estimations of treatment effects were
made. Average Treatment effect on the Treated and Untreated (ATT and ATU) are computed
using the results for expected values of the dependent variable for users and non-users in actual
and counterfactual scenarios:

𝜑(𝑍𝛼)
𝐸(𝑌1𝑖 | 𝐼𝑖 = 1, 𝑋1𝑖 ) = 𝛽1 𝑋1𝑖 + 𝜎𝜖1𝑢 𝜌1 6
Φ(𝑍𝛼)

𝜑(𝑍𝛼)
𝐸(𝑌2𝑖 | 𝐼𝑖 = 0, 𝑋2𝑖 ) = 𝛽1𝑋2𝑖 − 𝜎𝜖2𝑢 𝜌1 7
(1 − Φ(𝑍𝛼))

𝜑(𝑍𝛼)
𝐸(𝑌2𝑖 | 𝐼𝑖 = 1, 𝑋1𝑖 ) = 𝛽2 𝑋1𝑖 + 𝜎𝜖2𝑢 𝜌2 8
Φ(𝑍𝛼)

𝜑(𝑍𝛼)
𝐸(𝑌1𝑖 | 𝐼𝑖 = 0, 𝑋2𝑖 ) = 𝛽2 𝑋2𝑖 − 𝜎𝜖1𝑢 𝜌2 9
(1 − Φ(𝑍𝛼))

ATT is the difference between the expected value of the outcome variable from equation 6 and 8.
It is the difference between the expected value of the dependent variable for users and if they had

12
not used. ATU is the difference between equations 7 and 9 estimating the difference between the
expected value of the outcome variable for non-users and if they had used the water.

5. Results and discussion

5.1. Poverty analysis
The level of poverty was tested between irrigation water users and non-users using Foster-Greer-
Thorbecke (FGT) indices. Based on the recommended daily energy requirement of 2100Kcal
Poverty line (Z) of Birr 3329.27 (=USD 123.30) per adult equivalent per year is used to estimate
the FGT indices of poverty. The poverty line was constructed using food and non-food per adult
equivalent consumption expenditure of the households. Table 2 below shows the results:

Table 2 FGT indices on consumption at α=0, 1, and 2 and Z= ETB 3746.77

Poverty estimates Groups

Users Nonusers
Incidence 0.28 0.67
Depth 0.03 0.18
Severity 0.008 0.06
Source: computation on own survey date

The incidence of poverty is measured by the head count index and shows that 67% of the
households who do not irrigate fall below the consumption based poverty line (Z) of ETB
3329.27. On the other hand, only 28 % of the farmers who irrigate are below the poverty line.
This result reinforces the claim by several studies that prevalence of poverty is higher in rain fed
areas than irrigated areas (Bacha et al., 2011; Hanjra et al., 2009; Wood et al., 2004). The depth
and severity of poverty is also higher among the non-users. The consumption expenditure of the
non-users should be pushed up by 18% of its current amount if they have to be lifted out of
poverty while it takes only 3% for users. These results are in line with most micro level empirical
studies on poverty and irrigation linkage. Bacha et al., (2011), for instance, found out that depth
of poverty among non-irrigators in Ambo district in Western Ethiopia is 21 % while it is only
10% among users.

13
 5.2. Results of switching regression analysis
As indicated by the descriptive and poverty analysis above there is significant difference in
several relevant variables and wellbeing indicators between the users and nonusers. These
differences could be due to several observable and unobservable factors in addition to access to
irrigation. Two outcome variables were used as proxy for welfare of the households for further
analysis: Farm income per ha and per adult equivalent food consumption expenditure. Table 3
presents the estimation results for the model with farm income per hectare as the outcome
variable.

Table 3: Full information maximum likelihood estimates of the switching regression model for
farm income per ha
Variables Model Estimates
Irrigation 1/0 Users Non-users

Highest Education level of 0.023(0.66) -0.020(1.73)* 0.008(0.37)

adult member

Age of the head of the 0.011(1.26) -0.003(0.99) 0.003(0.48)

household

Log of distance to nearest -0.011(0.24) -0.017(0.90) -0.045(1.37)*

market

Visit by extension workers 0.116(0.39) 0.185(1.49)* 0.200(0.93)

Log of Landholding 0.367(1.15) 0.412(3.68)*** 0.584(2.41)***

Log of landholding per plot -0.375(0.91) -0.264(1.92)** -0.655(2.27)***

Log of percentage of land 1.170(3.38)*** 0.417(3.02)*** -0.286(1.31)*

covered by cash crop

Log of Time endowment in -0.086(0.28) 0.038(0.36) 0.068(0.32)

adult equivalent units

Log of non-farm income -0.088(2.91)*** 0.002(0.26) 0.041(2.27)***

Log of chemical fertilizer 0.136(1.04) 0.198(3.67)*** 0.096(1.18)

applied

Log of pesticide applied 0.124(1.55)* 0.151(4.20)*** 0.143(2.61)***

14
Owned radio(1/0) 0.522(2.44)***

Owned mobile phone (1/0) 0.870(3.96)***

Scarcity level (high= 1/Low -2.177(5.64)***

=0)

Governance (1/0) -0.578(2.52)***

Constant 0.415(0.33) 11.41(27.08)*** 9.31(12.50)***

ρ1, ρ2 -0.734[0.107]** -0.478[0.263]*

Model diagnosis
Wald x2 109.28***
Log likelihood -268.58
LR test of independence 19.66***
Note: Absolute value of z statistics in parenthesis standard errors in square brackets
*Significant at the 10% level, ** Significant at the 5% level, ***Significant at the 1% level
Source: Own calculation using survey data

The second column of table 3 reports the estimates for the determinants of the decision to
irrigate. Education and age as explanatory variables have positive non-significant association
with the irrigation decision of the farmers. Generally, education tends to have positive
association with new technology adoption among farmers because of better access to and
comprehension of information on the technologies (Norris & Batie, 1987). Studies have
indicated a positive relationship between education and age and adoption of new technology
(Deressa, Hassan, Ringler, Alemu, & Yesuf, 2009; Lin, 1991; Nhemachena & Hassan, 2007)
while others found out negative association between the variables (Shiferaw & Holden, 1998).
Huffman (2001), however, argues that when an intervention has been there for relatively long
time education and experience may not significantly affect decision to participate. This
reinforces the above result considering irrigation has been practiced in the study area for at least
the last 30 years.

Land covered by cash crops has positive impact the decision to irrigate. The common cash crops
grown in the study area are sugarcane and khat on which irrigation is widely practiced.
Therefore, households with larger share of their land covered by cash crop are likely to irrigate
than the others. Land size per plot, on the other hand, has negative effect on the decision to

15
irrigate. This is possibly because farmers with smaller plots can increase their production only by
intensifying through adopting of technologies like irrigation. Previous studies indicated mixed
results on the association of land size and probability of adoption of technology (Bradshaw et al.,
2004; Deressa et al., 2009; Khonje et al., 2015). Similarly time endowment shows negative
association with the decision to irrigate. The negative relationship can be attributable to the fact
that the study area is one of the areas with high population density and resulting land
fragmentation. In southern Ethiopia the rural youth is forced to search for other livelihood
options because of scarcity of agricultural land (Bezu & Holden, 2014). The average land
holding in the country is 1.37 ha (Central Statistical Agency & World Bank, 2013). Some
studies in Ethiopia identified similar result (Tizale, 2007) while others showed positive
association between labor endowment and adoption decision (Deressa et al., 2009). Non-farm
income has negative and significant impact on practice of irrigation. This result is reinforced by
the findings of Wozniak (1984) that participation in non-farm activities may constrain the
amount of labor hour available for farm activities.

The variables representing amount of chemical fertilizer and pesticide applied show positive
relationship. This is because farmers who irrigate tend to use modern inputs to enhance
productivity. It is commonly argued that stable supply of agricultural water would encourage
farmers to invest on productivity enhancing inputs (Aberra, 2004).This relationship is supported
by findings of Gebregziabher et al., (2009), Namara et al., (2010) Smith, (2004) who indicated
that irrigation enhances the use of productivity boosting inputs.

Proper specification of the model requires the inclusion of at least one explanatory variable in the
selection equation which directly affects the irrigation decision but not the outcome variable
(Abdulai & Huffman, 2014; Khonje et al., 2015). Accordingly, proxy variables for access to
information and networks represented by ownership of a radio and mobile phone were used.
Estimates for both variables are positive and significantly different from zero. Bandiera & Rasul
(2006) found out similar result in their study that farmers’ decision to participate in an
intervention is influenced by their network with family and friends. The ‘scarcity level’ is
another variable used to represent the level of scarcity of water farmers are facing with negative
and highly significant result. This shows that scarcity of water is hindering the farmers from
irrigating. This is reinforced by the empirical findings of several studies indicating water as the

16
major constraint in agriculture in most developing countries in the world as well as Ethiopia
(Hanjra et al., 2009; Namara et al., 2010).

The variable governance is a dummy variable used to capture the institutional, scheme
governance and allocation of water issues in the area. The estimate for the proxy variable is
positive and significantly different from zero. The result is plausible knowing that one of the
major problems shared by all small-scale irrigation schemes in Sub-Saharan Africa is related to
the application of irrigation water attributable to excessive abstraction of water by upstream
users(Aberra, 2004). Studies pointed out that the common type of water scarcity in Sub-Saharan
Africa is economic scarcity which results from poor governance and mismanagement (CTA,
2011; Dudu & Chumi, 2008). The findings of a study by (Awulachew & Ayana, 2011) also
indicated proper management of the already developed schemes should be given equal attention
as developing new ones. If implementation of small scale irrigation scheme is to be successful
‘group organization and cohesion’ is crucial (Aberra, 2004) and ‘well organized’ water user
associations and farmers cooperatives need to be established (Setegn et al., 2011).

Another important finding is the sign and significance of the correlation coefficients ρ1 and ρ2.
The results show that the coefficients are statistically significant for both users and non-users
indicating the existence of self-selection. The estimate is also negative for both users and non-
users indicating positive selection bias such that farmers with above average farm income tend to
decide to irrigate. The likelihood ratio test is also significant indicating the existence of joint
dependence between the outcome and selection equation between users and non-users.

The model estimates of the variables against farm income per ha for users and non-users are
presented on the third and fourth column of table 3. Education shows negative and statistically
significant result for users. Di Falco et al., (2011) found similar results for literacy and
production per ha in Ethiopia. The adult members of the households with higher level of
education tend to get involved in non-farm employments. This in turn suppresses the farm
income because the time allocated to farm activities will be less. Distance to local markets
measured in kilometers also negatively and significantly affects the farm income per ha for both
the users and non-users. Advice and information from the agricultural extension workers as
measured by visit by the officers shows positive association with farm income for both users and

17
non-users. This is in line with the argument that farmers with better information and advice from
extension workers are likely to have better productivity (Abdulai & Huffman, 2014).

Land holding has positive and significant impact on the outcome variable for both users and non-
users. On the contrary, landholding per plot has negative and significant impact on farm income
per hectare for both users and non-users. The negative and significant impact of the farm size per
plot follows well the argument of the inverse farm size productivity relationship. Studies proved
that small farms are more productive than big farms (Abdulai & Huffman, 2014). A study in
Ethiopia also found out that land pressure is strongly associated with crop yield and income
proving the holding of Boserup’s hypothesis (Headey, Dereje, & Taffesse, 2014) The amount of
non-farm income earned by the non-user farm households has a positive and significant impact
on their farm income unlike the users. This is because the income from non-farm sources can be
invested to purchase productivity enhancing inputs like fertilizers and improved crop varieties
(Abdulai & Huffman, 2014). The amount of chemical fertilizer and insect/herbicide applied has
positive association with farm income of both users and non-users with highly significant impact
on farm income of only users.
The proportion of farm covered by cash crops also has significant positive and negative impact
for users and non-users, respectively. Users harvest Khat, one of the most traded cash crops in
the area, at least twice per year. The non-users on the other hand harvest this crop only once per
year because they produce using rainwater. Regarding sugarcane the non-users complain about
quality and yield compared to the one exposed to enough water. As the largest source of farm
income, therefore, as the share of farm covered by cash crops increase the farm income per ha of
the non-users decreases significantly while in increases for the users.

A model was also with the total per adult equivalent food consumption as dependent variable.
Table 4 below presents the model results:

18
Table 4: Full information maximum likelihood estimates of the switching regression model for
total per adult equivalent food consumption
Variables Model Estimates
Irrigation 1/0 Users Non-users

Log of farm income 1.562(6.69)*** 0.070(1.92)** 0.103(1.34)*

Highest education level of adult -0.010(0.24) 0.001(0.27) 0.002(0.24)

member

Age of the head of the household 0.009(0.75) 3.4 e-05 (0.02) 5.9e-05(0.02)

Log of landholding -0.870(2.47)** 0.035(0.68) 0.035(0.48)

Ratio of Log of land size 0.385(1.46)* 0.031(0.91) 0.073(1.40)

allocated to cash crop to food
crop

Log of adult equivalent -0.564(1.42)* -0.366(7.26)*** -0.593(5.72)**

household size

Ratio of Log of productive adult -0.071(0.10) -0.217(2.68)*** -0.187(1.08)

equivalent labor to total adult
equivalent household size

Log of non-farm income -0.047(1.24) -0.001(0.25) 0.016(2.07)**

Log of value of household asset 0.177(2.37)*** 0.034(3.06)*** 0.031(1.83)**

Log of number of visit by -0.175(1.38)

extension officers

Owned radio(1/0) 0.146(0.54)

Owned mobile phone (1/0) 0.316(1.08)

Scarcity level (high= 1/Low =0) -1.668(3.54)***

Perceived scheme governance -0.270(0.95)

problem (1/0)

Constant -18.56(5.93)*** 7.73(14.89)*** 7.62(8.84)***

ρ1, ρ2 0.24[0.266] 0.58[0.353]**

Model diagnosis
Wald x2 79.46***

19
Log likelihood -44.23
LR test of independence 1.66
Note: Absolute value of z statistics in parenthesis and standard errors in square brackets
*Significant at the 10% level, ** Significant at the 5% level, ***Significant at the 1% level
Source: Own calculation using survey data

The results for the selection equation of the ESR model with food consumption considered as
outcome variable is presented on the second column of Table 4. The estimates for the
determinants of the decision to irrigate, that are common with the previous model, are similar for
both models in direction with some variation on the significance level. New explanatory
variables like farm income and the value of household assets have positive and highly significant
impact on the farmers’ decision to irrigate. Bacha et al., (2011) and Deressa et al., (2009) found
the same result for value of asset and farm income versus probability of adoption of irrigation,
respectively.
The factors that affect the outcome variable, food consumption, are reported on the third and
fourth column of the table. The estimates for household size related variables are negative and
significantly different from zero for both users and non-users. This situation holds in most cases
because higher big household means less per- head consumption other things kept constant. The
same results are reported by Khonje et al., (2015) and Bacha et al., (2011) for impact studies in
Eastern Zambia and Western Ethiopia, respectively. The model estimates for the amount of non-
farm income show positive and significant impact on consumption of non-users. Mostly non-
farm income is used to augment the household income and cover for consumption expenditure
shortfall (Dorward et al., 2004) Similarly, the variable for the total value of household asset gives
positive and significant result for both groups of farmers which confirms with findings of (Bacha
et al., 2011).
Finally, the average treatment effect on the treated (ATT) and untreated (ATU) are presented in
the following tables.
Table 5: Impact of irrigation on farm income and consumption-ATT
Outcome variable Mean outcome ATT t-value
Log of farm income per ha 354 895.97 171320.26 183575.71 107% 10.80***
Log of total per adult equivalent 3818.30 3008.62 809.68 26% 10.42***
consumption
***significant at 1% level, Source: own calculation from survey data

20
Table 6: Impact of irrigation on farm income and consumption-ATU
Outcome variable Mean outcome ATU t-value
Log of farm income per ha 214490.34 79 082.74 135407.60 171% 11.78***
Log of total per adult equivalent 4232.32 2691.09 1541.23 57% 16.08***
consumption
***significant at 1% level
Source: own calculation from survey data
As presented in Table 5 and 6 above irrigation significantly affects both outcome variables for
both groups. This finding is in line with previous studies that argue the existence of direct
income and employment impact of irrigation (Bacha et al., 2011; Smith, 2004; Wichelns, 2014).
Generally, technology adoption results in reduced poverty and improved food security by
improving agricultural production and productivity (Khonje et al., 2015). The report of FAO
(2003) claiming that crop yields increased by 100-400% with irrigated crops as compared to rain
fed is another reinforcing evidence for the above result.

6. Conclusion and Implications

This study examined the determinants of farmers’ decision to irrigate and impact of irrigation on
welfare of households using ESR. The results indicate the existence of a selection bias among the
users and non-users as can be seen from the significant correlation coefficient between the error
terms of the selection equation and outcome equation. Variables relating to information and
social network, water scarcity level, and governance issues have significant impact on farmer’s
decision to irrigate. The ATT and ATU are positive and significant for both users and non-users
indicating that access to irrigation has resulted in significant positive impact on welfare of the
farmers.

Considering the findings of this study, several policy implications could be drawn. Significant
share of the water scarcity is created by poor infrastructure and resulting wastage of water.
Renovating and improving the conditions of the scheme canals is the best starting point to ensure
access of more people to irrigation water. Moreover, the current irrigation and production
technology is far from modern like elsewhere in the country. The government could work on
promoting the adoption and provision of better technologies. Moreover, water storage facilities
could significantly enhance the availability of water. Such facilities also make possible multiple

21
uses of water like fisheries which can be considered to enhance livelihood of households
especially of women and vulnerable groups with small landholdings. Related with this is a
promoting and developing alternative source of irrigation water.

Access to irrigation water by itself does not necessarily lead to better life conditions. Households
who have access to irrigation water but still below the poverty line witness the existence of
transaction costs involved in securing access to water, purchasing other inputs, and selling
outputs claiming huge share of the their income. This calls for intervention in relevant
institutional settings and markets to fully realize the potential of irrigation. WUA should be
supported by sound legal framework with proper monitoring and evaluation systems in place to
ensure fair allocation of water.

There is huge involvement of brokers and middlemen in the market especially for khat and
sugarcane. In this regard, it would be beneficial for the farmers if they could be organized in
producer cooperatives for collective marketing. Especially, with the highly developing sugar
industry in the country, there is huge market demand for sugarcane and the farmers could seize
this opportunity. Facilitating access to small and medium credit facility can also improve the
farmers’ capacity to invest in alternative irrigation water sources like shallow well and rainwater
harvesting facilities and modern agricultural inputs.

Last but no least is making sure that the interventions are integrated, pro-poor, and targeted. The
study area is one of the densely populated areas in the country with very small and fragmented
landholdings. The young generation is running out of land resource to stay in the agriculture
sector. The importance of non-farm sector is critical in such situations.

22
Reference

Abdulai, A., & Huffman, W. (2014). The Adoption and Impact of Soil and Water Conservation
Technology: An Endogenous Switching Regression Application. Land Economics, 90(1),
26–43. https://doi.org/10.1353/lde.2014.0009

Aberra, Y. (2004). Problems of the solution: Intervention into small-scale irrigation for drought
proofing in the Mekele Plateau of northern Ethiopia. Geographical Journal, 170(3), 226–
237. https://doi.org/10.1111/j.0016-7398.2004.00122.x

Amare, M., Asfaw, S., & Shiferaw, B. (2012). Welfare impacts of maize-pigeonpea
intensification in Tanzania. Agricultural Economics, 43(1), 27–43.
https://doi.org/10.1111/j.1574-0862.2011.00563.x

Asfaw, S. (2010). Estimating Welfare Effect of Modern Agricultural Technologies: A Micro-

Perspective from Tanzania and Ethiopia. Chronicpoverty.Org, 1–27. Retrieved from
http://www.chronicpoverty.org/uploads/publication_files/asfaw_agricultural_technologies.p
df

Awulachew, S. B., & Ayana, M. (2011). Performance of Irrigation: an Assessment At Different

Scales in Ethiopia. Experimental Agriculture, 47(S1), 57–69.
https://doi.org/10.1017/S0014479710000955

Bacha, D., Namara, R., Bogale, A., & Tesfaye, A. (2011). Impact of small-scale irrigation on
household poverty: empirical evidence from the Ambo district in Ethiopia. Irrigation and
Drainage, 60(1), 1–10. https://doi.org/10.1002/ird.550

Bandiera, O., & Rasul, I. (2006). Social Networks and Technology Adoption in Northern
Mozambique Author ( s ): Oriana Bandiera and Imran Rasul Published by : Wiley on behalf
of the Royal Economic Society Stable URL : http://www.jstor.org/stable/4121936,
116(514), 869–902. https://doi.org/10.1111/j.1468-0297.2006.01115.x

Bezu, S., & Holden, S. (2014). Are rural youth in ethiopia abandoning agriculture? World
Development, 64, 259–272. https://doi.org/10.1016/j.worlddev.2014.06.013

23
Bradshaw, B., Dolan, H., & Smit, B. (2004). Farm-level adaptation to climatic variability and
change: Crop diversification in the Canadian prairies. Climatic Change, 67(1), 119–141.
https://doi.org/10.1007/s10584-004-0710-z

Bravo-Ureta, B. E., Solís, D., Cocchi, H., & Quiroga, R. E. (2006). The impact of soil
conservation and output diversification on farm income in Central American hillside
farming. Agricultural Economics, 35(3), 267–276. https://doi.org/10.1111/j.1574-
0862.2006.00161.x

Central Statistical Agency, & World Bank. (2013). Ethiopia Rural Socioeconomic Survey
(ERSS) Survey Report, 1–64.

CTA. (2011). The water we eat – tackling scarcity in ACP countries.Technical Centre for
Agricultural and Rural Cooperation (ACP-EU). CTA Policy Brief, 2(June), 4.

Deressa, T. T., Hassan, R. M., Ringler, C., Alemu, T., & Yesuf, M. (2009). Determinants of
farmers’ choice of adaptation methods to climate change in the Nile Basin of Ethiopia.
Global Environmental Change, 19(2), 248–255.
https://doi.org/10.1016/j.gloenvcha.2009.01.002

Dessie, G., & Kinlund, P. (2008). Khat expansion and forest decline in wondo genet, Ethiopia.
Geografiska Annaler, Series B: Human Geography, 90(2), 187–203.
https://doi.org/10.1111/j.1468-0467.2008.00286.x

Di Falco, S., Veronesi, M., & Yesuf, M. (2011). Does adaptation to climate change provide food
security? A micro-perspective from Ethiopia. American Journal of Agricultural Economics,
93(3), 825–842. https://doi.org/10.1093/ajae/aar006

Dorward, A., Fan, S., Kydd, J., Lofgren, H., Morrison, J., Poulton, C., … Wobst, P. (2004).
Rethinking agricultural policies for pro-poor growth. O.D.I, 94(94), 1–4. Retrieved from
http://www.odi.org.uk/sites/odi.org.uk/files/odi-assets/publications-opinion-files/894.pdf

Dudu, H., & Chumi, S. (2008). Economics of Irrigation Water Management : A Literature
Survey with Focus on Partial and General Equilibrium Models.Policy Research Working
Paper 4556. The World Bank Development Research Group Sustainable Rural and Urban
24
 Development Team March 2008, (March).

Dutoit, L. (2007). Heckman’s Selection Model, Endogenous and Exogenous Switching Models:
A survey.

FAO. (2003). The State of Food Insecurity in the World: monitoring progress towards the World
Food Summit and Millennium Development Goals. Published in 2003 by the Food and
Agriculture Organization of the United Nations Viale delle Terme di Caracalla, 00100
Rome, Ita. Retrieved from https://www.presseportal.de/download/document/25159-fao-sofi-
all-v5.pdf

Foster, J., And, J. G., & Thorbecke, E. (1984). A Class of Decomposable Poverty Measures
Author ( s ): The Econometric Society Stable URL : http://www.jstor.org/stable/1913475,
52(3), 761–766.

Gebregziabher, G., Namara, R. E., & Holden, S. (2009). Poverty reduction with irrigation
investment: An empirical case study from Tigray, Ethiopia. Agricultural Water
Management, 96(12), 1837–1843. https://doi.org/10.1016/j.agwat.2009.08.004

Goldfeld, S. M. and, & Quandt, R. . (1972). Nonlinear methods in econometrics (North-Holland,

Amsterdam). Journal of Econometrics, 1(4), 399–401.

Goldfeld, S. M. and, & Quandt, R. E. (1973). A Markov model for switching regressions.
Journal of Econometrics, 1(1), 3–15. https://doi.org/10.1016/0304-4076(73)90002-X

Hanjra, M. A., Ferede, T., & Gutta, D. G. (2009). Reducing poverty in sub-Saharan Africa
through investments in water and other priorities. Agricultural Water Management, 96(7),
1062–1070. https://doi.org/10.1016/j.agwat.2009.03.001

Headey, D., Dereje, M., & Taffesse, A. S. (2014). Land constraints and agricultural
intensification in Ethiopia: A village-level analysis of high-potential areas. Food Policy, 48,
129–141. https://doi.org/10.1016/j.foodpol.2014.01.008

Huffman, W. E. (2001). Agricultural Production. Handbook of Agricultural Economics, 1, 333–

381. https://doi.org/10.1016/S1574-0072(01)10010-1

25
Hussain, I., & Hanjra, M. (2004). Irrigation and poverty alleviation: Review of the empirical
evidence. Irrigation and Drainage, 53(1), 1–15. https://doi.org/10.1002/ird.114

Jin, S., Huang, J., Hu, R. and, & Rozelle, S. (2002). T OTAL F ACTOR P RODUCTIVITY IN “
P HOENIX M IRACLES ”: Working Paper 0605. American Journal of Agricultural
Economics, 84(3), 916–930.

Kassie, M., Shiferaw, B., & Muricho, G. (2010). Adoption and impact of improved groundnut
varieties on rural poverty: evidence from rural Uganda. Environment for Development
Discussion Paper - Resources for the Future (RFF), (May), 10–11. Retrieved from
http://www.rff.org/RFF/Documents/EfD-DP-10-11.pdf

Khonje, M., Manda, J., Alene, A. D., & Kassie, M. (2015). Analysis of Adoption and Impacts of
Improved Maize Varieties in Eastern Zambia. World Development, 66, 695 706.
https://doi.org/10.1016/j.worlddev.2014.09.008

Lee, L.-F., Maddala, G. S., & Trost, R. P. (1982). TESTING FOR STRUCTURAL CHANGE
BY D-METHODS IN SWITCHING SIMULTANEOUS EQUATIONS MODELS.
CENTER FOR NAVAL ANALYSES. PROFESSIONAL PAPER 342.

Lin, J. Y. (1991). Education and Innovation Adoption in Agriculture: Evidence from Hybrid
Rice in China. Agricultural Economics, 73(3), 713–723. https://doi.org/10.2307/1242823

Lokshin, M. (1977). of Endogenous Switching Regression Models, 1–8.

Lokshin, M., & Sajaia, Z. (2004). Maximum likelihood estimation of endogenous switching
regression models. The Stata Journal, 4(3), 282–289. https://doi.org/The Stata Journal

Maddala, G. S. (1983). Limited-dependent and qualitative variables in econometrics (No. 3).

Cambridge university press.

Maddala, G. S. (1986). DISEQUILIBRIUM , SELF-SELECTION , AND SWITCHING

MODELS * 2 . Estimation of the switching regression model : Sample separation known 3 .
Estimation of the switching regression model : Sample separation unknown 4 . Estimation
of the switching regression mod. Science, III(1982).

26
Maddala, G. S., & Nelson, F. D. (1975). Switching regression models with exogenous and
endogenous switching. Proceedings of the American Statistical Association, 5, 423–426.

Namara, R. E., Hanjra, M. A., Castillo, G. E., Ravnborg, H. M., Smith, L., & Van Koppen, B.
(2010). Agricultural water management and poverty linkages. Agricultural Water
Management, 97(4), 520–527. https://doi.org/10.1016/j.agwat.2009.05.007

Nhemachena, C., & Hassan, R. (2007). Micro-level analysis of farmers adaption to climate
change in Southern Africa. Intl Food Policy Res Inst.

Norris, P. E., & Batie, S. S. (1987). Virginia farmers’ soil conservation decisions: an application
of the tobit analysis. Southern Journal of Agricultural Economics, 19, 79–90.
https://doi.org/10.1017/S0081305200017404

O’Boyle, E. J. (1999). Toward an improved definition of poverty. Review of Social Economy,

57(3), 281–301. https://doi.org/10.1080/00346769900000003

Ravallion, M. (2005). Evaluating Anti-Poverty Programs, 4(May), 1–90.

Rosegrant, M. W., & Evenson, R. E. (1992). Agricultural Productivity and Sources of Growth in
South Asia. Journal of Agricultural Economics, 74(3), 758–761.

Rosegrant, M. W., Ringler, C., & Zhu, T. (2009). Water for Agriculture: Maintaining Food
Security under Growing Scarcity. Annual Review of Environment and Resources, 34(1),
205–222. https://doi.org/10.1146/annurev.environ.030308.090351

Sen, A. (1981). Poverty and famines: an essay on entitlement and deprivation. Oxford university
press.

Setegn, S. G., Chowdary, V. M., Mal, B. C., Yohannes, F., & Kono, Y. (2011). Water Balance
Study and Irrigation Strategies for Sustainable Management of a Tropical Ethiopian Lake:
A Case Study of Lake Alemaya. Water Resources Management, 25(9), 2081–2107.
https://doi.org/10.1007/s11269-011-9797-y

Shiferaw, B., & Holden, S. T. (1998). Resource Degradation and Adoption of Land Conservation

27
 Technologies by Smallholders in the Ethiopian Highlands : A case study in Andit Tid,
North Shewa. Agricultural Economics, 18(1998), 233–247.

Shiferaw, B., Kassie, M., Jaleta, M., & Yirga, C. (2014). Adoption of improved wheat varieties
and impacts on household food security in Ethiopia. Food Policy, 44, 272–284.
https://doi.org/10.1016/j.foodpol.2013.09.012

Singh, A. K., Rahman, A., Sharma, S. P., Upadhyaya, A., & Sikka, A. K. (2009). Small holders’
irrigation - Problems and options. Water Resources Management, 23(2), 289–302.
https://doi.org/10.1007/s11269-008-9275-3

Smith, L. E. D. (2004). Assessment of the contribution of irrigation to poverty reduction and

sustainable livelihoods. International Journal of Water Resources Development, 20(2),
243–257. https://doi.org/10.1080/0790062042000206084

Tizale, C. Y. (2007). The dynamics of soil degradation and incentives for optimal management
in the Central Highlands of Ethiopia, (February), 263.
https://doi.org/https://repository.up.ac.za/bitstream/handle/2263/25333/Complete.pdf?seque
nce=6

Watts, H. W. (1968). An economic definition of poverty (pp. 316-329). Institute for Research on
Poverty.

Wichelns, D. (2014). Investing in small, private irrigation to increase production and enhance
livelihoods. Agricultural Water Management, 131, 163–166.
https://doi.org/10.1016/j.agwat.2013.09.003

Wood, S., You, L., & Zhang, X. (2004). Spatial patterns of crop yields in Latin America and the
Caribbean. Cuardernos de Economia / Latin American Journal of Economics, 41(124),
361–381. Retrieved from http://www.scielo.cl/pdf/cecon/v41n124/art03.pdf

Wooldridge, J. M. (2003). Introductory Econometrics: A Modern Approach. Economic Analysis,

2nd, 824. https://doi.org/10.1198/jasa.2006.s154

World Bank. (2010). Handbook on Impact evaluation: Quantitative methods and practices. The

28
 World Bank, Washington, D.C. World (Vol. 41).

World Bank. (2016). World Development Indicators, 46. https://doi.org/10.1596/978–1-4648–

0683–4

Wozniak, G. D. . (1984). The Adoption of Interrelated Innovations : A Human Capital Approach

Source : The Review of Economics and Statistics, 66(1), 70–79. Retrieved from
http://www.jstor.org/stable/1924697

29