Off-Farm Income and Technical

Efficiency of Smallholder Farmers in

Ethiopia
- A Stochastic Frontier Analysis

Getahun Gemechu Abebe

Master’s thesis · 30 hec · Advanced level

European Erasmus Mundus Master Program: Agricultural Food and Environmental
Policy Analysis (AFEPA)
Degree thesis No 862 · ISSN 1401-4084
Uppsala 2014
iiii
Off-farm Income and Technical Efficiency of Smallholder Farmers in Ethiopia
- A Stochastic Frontier Analysis

Getahun Gemechu Abebe

Supervisor: Sebastian Hess

Swedish University of Agricultural Sciences (SLU)
Department of Economics

Assistant Supervisor Thomas Heckelei

University of Bonn
Institute of Food and Resource Economics

Examiner: Ing-Marie Gren

Swedish University of Agricultural Sciences (SLU)
Department of Economics

Credits: 30 hec
Level: A2E
Course title: Independent Project/ Degree Project in Economics E
Course code: EX 0537
Programme/Education: European Erasmus Mundus Master Program: Agricultural Food
and Environmental Policy Analysis (AFEPA)
Faculty: Faculty of Natural Resources and Agricultural Sciences

Place of publication: Uppsala

Year of publication: 2014
Name of Series: Degree project/SLU, Department of Economics
No: 862
ISSN 1401-4084
Online publication: http://stud.epsilon.slu.se

Key words: Cobb-Douglas, Maximum Likelihood Estimation, Off-farm income,

Smallholder Farmers, Technical Efficiency
Acknowledgements
First, I would like to thank the European Commission for granting me the prestigious
Erasmus Mundus scholarship. Next, I would like to thank my supervisor Dr. Sebastian Hess
for his constructive comments and suggestion in this study. I am also thankful to Professor
Ing-Marie Gren for her comments and follow-ups during the seminar sessions. Furthermore,
my gratitude also goes to Professor Thomas Heckelei for his motivation to be my thesis
examinor at University of Bonn. I am also thankful to AFEPA program coordinators
Professor Bruno Henry de Frahan, Henrich Brunke, Dr. Ralf Nolten and Kristina Jansson for
their help on practical matters. My special gratitude also goes to Wondemagne Tafesse for his
continuous advises and motivations in writing this thesis. Friends, Tesfaye B. Birmaji, Wagari
Shore and Selam Assefa deserve much more thank for their valuable advises. Finally, I would
like to thank AFEPAians and other friends I have met during my stay in Germany and
Sweden.

iii
Abstract
As in most developing countries, agricultural production in Ethiopia is dominated by
subsistence-based smallholder farmers, whose production and incomes from the sector are
constrained by socio-economic, institutional, resource and environmental factors. These
factors generally attribute for lower productivity of the sector, which in turn forces farmers to
participate in off-farm activities in order to diversify their sources of income. However,
participation in off-farm activities has direct and indirect influence on agricultural production.
The purpose of this study is to examine the impact of off-farm income on technical efficiency
and farm output of smallholders in Ethiopia. The study used data from the 2009 Ethiopian
rural household survey conducted by International Food Policy Research Institute. A
Stochastic frontier model is used to address the objectives of the research. Results confirm the
appropriateness of the Cobb-Douglas form of the production function over the Translog and
Stochastic Frontier Analysis over the Ordinary Least Squares.
An instrumental variable regression framework is used to address the endogeneity of off-farm
income in determining technical efficiency and farm output of farmers. The estimation results
show that size of farm land, household size, off-farm income, gender and education of the
household head are the most significant variables determining the value of farm output.
The average technical efficiency of farmers is only 53 percent, implying the existence of
wider scope for improvement of their efficiency. In addition, maximum likelihood estimation
result indicates that household size, education of the head, soil conservation, extension
services and off-farm income are major factors for differences in technical efficiency among
farmers. Particularly, the effect of off-farm income on farm output and technical efficiency is
positive showing the spillover effects of income from off-farm activities on farm productions.
Hence, policy makers should focus on increasing opportunities and access of off-farm
activities to enhance production, productivity and overall wellbeing of the rural societies.

iv
Abbreviations
CD Cobb-Douglas
DEA Data Envelopment Analysis
EHRS Ethiopian Rural Household Survey
ETB Ethiopian Birr
GDP Gross Domestic Product
Ha Hectare
HH Household
IFPRI International Food Policy Research Institute
IV Instrumental Variable
Kgs Kilograms
LR Log-likelihood Ratio
MLE Maximum Likelihood Estimation
OLS Ordinary Least Squares
TE Technical Efficiency
TL Translog
SFA Stochastic Frontier Analysis
SNNPRS Southern Nation’s Nationalities and People’s Regional State
2SLS Two stage Least Squares

v
Table of Contents
Acknowledgements ................................................................................................................... iii

Abstract ..................................................................................................................................... iv

Abbreviations ............................................................................................................................. v

Table of contents ....................................................................................................................... vi

List of figures .......................................................................................................................... viii

List of tables ............................................................................................................................ viii

1. Introduction............................................................................................................................ 1

1.1. Background of the study .................................................................................................. 1

1.2. Objective of the study ...................................................................................................... 3
1.3. Research Questions .......................................................................................................... 3
1.4. Scope of the study............................................................................................................ 4
1.5. Relevance of the study ..................................................................................................... 4
1.6. Organization of the study................................................................................................. 4

2. Theoretical Approach ............................................................................................................ 5

2.1. Overview of production function and efficiency ........................................................... 5

2.2. Measures of technical efficiency .................................................................................... 7

3. Literature Review .................................................................................................................. 9

3.1. Empirical studies on determinants of technincal efficiency........................................... 9

3.2. Determinants of participation in off-farm activities ..................................................... 11
3.3. Empirical studies on agricultural production and off-farm income ............................. 12

4. Methodology of the Study ................................................................................................... 15

4.1. Descrtiption of study areas and data ............................................................................ 15

4.2. Empirical model specification ...................................................................................... 16
4.2.1. Production frontier and technical efficiency....................................................... 16
4.2.2. Off-farm income and farm output........................................................................ 22
4.2.3. Estimation Technique .......................................................................................... 23

vi
5. Result and Discussion .......................................................................................................... 24

5.1. Descriptive Statistics ................................................................................................. 24

5.2. Econometric Results .................................................................................................. 26
5.2.1. OLS Estimates of Production Frontier ................................................................ 26
5.2.2. Test for Model Specifications .............................................................................. 27
5.2.3. MLE Estimates of Production Frontier................................................................ 29
5.2.4. Distribution of technical efficiency ..................................................................... 30
5.2.5. Elasticity of factors of production ....................................................................... 31
5.2.6. The effect of off-farm income on technical efficiency ........................................ 31
5.2.6.1.Sensitivity Analysis ....................................................................................... 34
5.2.7.Effects of off-farm income on farm output ........................................................... 35

6. Conclusion and Policy Recommendation ............................................................................ 39

Appendices ............................................................................................................................... 41

References ................................................................................................................................ 48

vii
List of Figures
Figure 1: Graphical representation of technical and allocative efficiencies .............................. 6
Figure 2: Graphical illustration of stochastic frontier production function ............................... 8
Figure 3: Types of off-farm income generating activities........................................................ 16
Figure 4: Distribution of output per worker ............................................................................. 25
Figure 5: Distribution of output per hectare ............................................................................. 26
Figure 6: Distribution of technical efficiency .......................................................................... 31

List of Tables
Table 1: Summary of selected empirical studies on technical efficiency in Ethiopia.............. 10
Table 2:Summary statistics of model variables........................................................................ 24
Table 3: OLS estimates of CD production function: The dependent is log of value of output 27
Table 4: MLE estimates of production function in half-normal, exponential and truncated
models ................................................................................................................................ 30
Table 5: MLE estimates of determinants of technical inefficiency: The dependent variable is
technical inefficiency......................................................................................................... 33
Table 6: Production frontiers of farmers with and without off-farm income........................... 35
Table 7: First-stage estimates of off-farm income: the dependent variable is off-farm income
........................................................................................................................................... 36
Table 8:2SLS estimates of farm output: the dependent variable is value of farm output ........ 37

viii
1. Introduction
1.1. Background of the study

Agriculture is still an important pillar of our planet’s economy. The sector plays important
roles in the development process of any nation by supplying food items, industrial inputs,
generating foreign exchange, creating employment opportunities, contributing to gross
domestic product (GDP) and expanding markets for industrial outputs. According to World
Bank (2007), agriculture is the major source of income and employment for about 70
percent of the world’s rural poor societies and 32 percent in the growth of GDP in these
countries.

The sector is also the mainstay of the Ethiopian economy, constituting 45.6 percent of the
economy in GDP, 90 percent of foreign exchange, 85 percent of total employment
opportunities and 70 percent of industrial raw materials. It is also the major supplier of food
stuffs for the entire population and hence plays a crucial role in contributing to country’s
food security programs. Besides, it is this sector that is expected to play an important role in
creating surplus capital for the economic development processes in the country (Deressa,
2007).
In spite of its great importance to the country’s economy, agricultural productions are
however, subsistence-based and dominated by smallholder farm households that operate on
farms of less than one hectare (Gebre-Selassie, 2004). Smallholder farming represents for
about 90 percent of agricultural outputs and 95 percent of land area under crop production.
In general, about 98 percent of coffee, the country’s leading cash crop and 94 percent of
food crops are produced by smallholders, while only 2 percent of coffee and 6 percent of
crop production are produced by private and state commercial farms. Even though the
present government has given higher priority to the agricultural sector, its productivity is
however at its lowest level because of different interrelated socio-economic and climatic
problems such as inappropriate use of farm land, over grazing, over cultivation, population
growth, tenure insecurity, weak extension services, inadequacy of infrastructure, low
access to fertilizer and pesticides (Deressa, 2007). According Rahman (2007), smallholder
productions are generally characterized by low access to improved technologies, financial
services, modern inputs, agricultural markets and irrigations services that attributed to
variability of earnings from the farming sector. As a result, farmers are forced to participate
in off-farm activities to overcome these obstacles.

1
Several studies suggest that participation in off-farm activities are initiated by two
conventional factors, namely pull and push factors. The “push” factors are mandatory
factors that force farmers to participate in off-farm activities in order to manage income risk
and in this case off-farm income is used as a coping mechanism. These factors include
shortfalls of agricultural production resulting from temporary failures due to, unexpected
drought or long term factors such as shortages of farm land, absence of crop insurance,
failures in input and credit markets and others (Reardon ,et al., 1998). On the other hand,
the “pull” factors are incentives that attract households to non-farming sectors when non-
farm activities offer higher return than the farm activities (Barrett ,et al., 2001).

According to Reardon ,et al. (1998), household’s participation in off-farm activities may
also differ depending on their level of wealth. Conventionally, poorer households have
more averse types of negative shocks to their production and they have less capacity to cope
up with these shocks. Thus, they are expected to diversify their income sources more than
richer households.

Off-farm income constitutes for about 25 to 40 percent in the total income of households in
Ethiopia.Woldehanna (2000) indicated that off-farm income constitutes for about 35
percent in the total income of households from his study northern part of Ethiopia.
Similarly, Beyene (2008) reported that 57.3 percent of farm households are participants in
off-farm activities. Latest studies such as Bedemo ,et al. (2013) reported that about 73.5
percent of households participate in off-farm activities in their studies in three districts of
the country; Guto Gida, Gida Ayana and Jima Arjo.

Unskilled wage works in others’ farm, unskilled non-farm 1 works, skilled and professional
works as teacher, clerical, government and health worker, trader, driver, weaver, tailor and
paid developmental works (for instance, food-for-work) are major off-farm activities in the
country (Woldehanna, 2002). In general, farm households participate in one or more types
of these activities to diversify their sources of income.

Even though households’ participation in off-farm activities have considerable effects on

the productivity of the farming sector, the possible link is not clear (Yue & Sonoda, 2012).
As documented in different studies, it can be positive or negative or nil depending on where
income from off-farm activities are invested. If incomes generated from off-farm activities

1
Non-farm income and Off-farm income are different in the way that non-farm excludes wages earned in the
agricultural sector; whereas, off-farm income is broader and includes all incomes generated out of own farm.

2
are spent as investment in the farming sector in the form of purchase of modern inputs and
adoption of new technology, then the sector’s production will be enhanced. On the other
hand, the absence of labor in the agricultural sector negatively affect the productivity of the
sector if incomes generated in the off-farm activities are spent for consumption or used as a
means to leave the agricultural sector (Feng, 2008; Pfeiffer ,et al., 2009; Babatunde, 2013).

Using panel data on Slovenian farms, Bojnec & Ferto (2011) for example studied the
impact of off-farm income on farm’s technical efficiency. Their finding indicates positive
association between off-farm income and technical efficiency. Similarly, Yue & Sonoda
(2012) also confirmed that farmers with off-farm wage are more efficient than those
without off-farm wage in their studies on Chinese farmers. On the other hand, Kumbhakar
,et al. (1989) found negative correlation between the level of technical efficiency and off-
farm income from their studies on Utah dairy farm households and Chang & Wen (2011)
confirmed differential impacts of off-farm income on technical efficiency of Taiwanese
farm households.
In developing countries, like Ethiopia, where income from farm activities varies
considerably, farm households usually participate in off-farm activities to supplement their
agricultural income (Beyene, 2008). Hence, off-farm income is expected to enhance their
production and productivity in farming. However, it is not known to what extent households
with off-farm income are better off than those without off-farm income and whether there
exists variability in the level of technical efficiency among the two groups of households in
the study areas. Therefore, this study intends to analyze the effects of off-farm income on
technical efficiency and aims to fill the gap in this area.

1.2. Objective of the study

The general objective of this study is to examine the potential effect of off-farm income on
technical efficiency of farmers in Ethiopia. Specifically, the study intends to

• identify factors determining technical efficiency of farmers,

• estimate the effect of off-farm income on the value of farm output and
• identify factors determining off-farm income.

1.3. Research Questions

This study aims to answer the following research questions:
• What is the existing mean efficiency of households?
• What determines efficiency and productivity of smallholders’ production?

3
 • What is the link between off-farm incomes and the observed level of technical
efficiency for households?

1.4. Scope of the study

This study mainly focuses on productivity of smallholder farmers in Ethiopia. It estimates
their technical efficiency and assesses the impact of off-farm income on the level of
efficiency. It uses the 2009 rural household survey data obtained from International Food
Policy Research Institute (IFPRI) to achieve the objectives. The data however, cover crop
production in major regions of the country more explicitly while details on livestock
subsectors and peripheral regions are not covered.

1.5. Relevance of the study

In countries in which growth and prosperity of the agricultural sector determine the fate of
the entire economy, increasing the sector’s productivity plays a greater role in increasing
economic growth. Understanding the possible link between on-farm and off-farm activities
plays a vital role in enhancing productivity of the farming sector and gives relevant policy
information for the betterment of off-farm sectors. Thus, results from this study will provide
information for stakeholders in government and extension services on how to increase the
productivity of the farming and non-farming sector. Besides, it adds to the existing
literature in the area of efficiency and off-farm income.

1.6. Organization of the study

The rest of the thesis is structured as follows. Chapter two presents the theoretical
approaches of the study. It mainly covers microeconomic theories of production and
efficiency measures. Chapter three focuses on literature review. It covers a broader range of
empirical literatures on determinants of efficiency and the linkage between off-farm income
and technical efficiency. Chapter four discusses methodological framework of the study. It
begins with the overall descriptions of the study areas followed by the specification of the
empirical model, description of model variables and finally gives brief explanation of
estimation techniques. Chapter five presents the descriptive and econometric results.The
last chapter summarizes the empirical findings of the study and draws appropriate policy
recommendations.

4
2. Theoretical Approach
2.1. Overview of production function and efficiency
Estimation of production functions and technical efficiency is one of the most popular areas
of research. In microeconomic theory, production is defined as the process of transforming
inputs (raw materials) into outputs. A production function represents technological
relationships between inputs and outputs. In particular, it shows the maximum level of
output the firm can produce combining the existing inputs (Besanko & Braeutigam, 2005).
A particular production function can be specified as:

f ( xi ) = max{y i : T ( xi , y i )} (1)

Where y i the maximum level of output (frontier output) the firm can produce, xi is the

quantity of various inputs employed and T ( xi , yi ) is the technological relationships between

inputs and outputs. Given the existing level of inputs, three assumptions are usually made
on frontier productions: (1) any production on the frontier output is attainable and efficient,
(2) any production possibilities below the frontier level are attainable, but technically
inefficient and (3) any production points above the frontier level are unattainable.

In general, the level of output can be increased in several ways. Firstly, by expanding the
level of inputs used in production. This approach is called “horizontal expansion”.
However, increasing use of inputs is only possible if either the price of inputs decrease or
the price of output increases. Secondly, output can be increased by enhancing efficiency in
production. This approach is termed as “improvement approach” and requires the
improvement of socio-economic, institutional and environmental constraints to enhance
production using the existing inputs. Thirdly, output can be also increased by improving the
technology in production. This includes use of improved techniques of production,
improved seeds, modern fertilizer and chemicals. This approach is termed the
“transformation approach” (Alene, 2003).

Most often, different studies use the terms productivity and efficiency interchangeably,
though they are not exactly the same. Productivity refers to the ratio of output(s) to input(s)
while; efficiency is the highest productivity level from each input level (Coelli & Rao,
1998). Farrell (1957) classified efficiency as technical (physical), allocative (price) and
economic (overall) efficiency. Technical efficiency shows the ablity of farmers to produce
maximum amount of output using the existing level of inputs. On the other hand, allocative

5
efficiency measures the ability of farmers to use inputs in an optimal proprtion, given the
price of inputs and outputs. A firm is economically (overall) efficient if it achieves both
technical and allocative efficiencies.
For a given firm which uses two inputs ( X 1 and X 2 ) to produce a single output ( q ) under a
constant return to scale, Farrell (1957) illustrated the three types of efficiency using the
following figure. The isoquant SS’ represents the different combinations of the two inputs
that the firm uses to produce a given amount of output and deviations from the isoquant
implies technical inefficiency of the firm. thus, if the firm for example uses inputs at point
P to produce a unique output on the isoquant; technical inefficiency of a firm is represnted
by the segment QP , which shows the amount by which all inputs could be proportionally
reduced without a reduction in the level of output. This can be expressed in percentage
terms by the ratio of QP / 0 P (Coelli ,et al., 2005).

Figure 1: Graphical representation of Technical and allocative efficiencies

Thus, the technical efficiency of a firm is one minus the ratio of QP / 0 P as shown in
equation (2). On the other hand, allocative efficiency is measured by the ratio of input
prices represented by the slope of isocost line AA’, whereas economic (overall efficiency)
is the product of technical and allocative efficiencies (Coelli ,et al., 2005).
Technical efficiency TE = 0Q / 0 P = 1 − (QP / 0 P ) (2)
Allocative efficiency AE = 0 R / 0Q (3)

Economic efficiency EE = TE x AE = (0Q / 0 P ) x(0 R / 0Q ) = (0 R / 0 P ) (4)

6
2.2. Measures of technical efficiency
In frontier models, technical efficiency is measured based on the performance of an
individual producer compared to the most efficient producer in the industry. Various
approaches have been used to measure efficiency of a producer. The most widely used
approaches are Stochastic Frontier Analysis (SFA) and Data Envelopment Analysis (DEA).
The DEA is a non-parametric approach that involves mathematical programming; whereas
the SFA is a parametric approach that uses econometric methods. The DEA approach
assumes that all deviations from the frontier output (for example, due to bad weather, strike
and shortage of inputs) are due to technical inefficiency; whereas, SFA approach considers
both an inefficiency component and a random error.
The SFA approach is usually preferred to estimate efficiencies of production systems. This
is mainly because of two reasons: (1) the very nature of agricultural productions depends
on climatic conditions and is affected by measurement errors that attribute for statistical
noise in data sets and (2) stochastic frontier models allows decomposition of error terms
between statistical noises and inefficiencies measure that enables statistical tests on the
validity of model specification (Gelaw, 2004; Chen, 2007).
Battese & Coelli (1993) explains stochastic frontier production as follows:
“The stochastic frontier production function postulates the existence of technical
inefficiencies of production of firms involved in producing a particular output. For a
given combination of input levels, it is assumed that the realized production of a firm is
bounded above by the sum of a parametric function of known inputs, involving
unknown parameters, and a random error, associated with measurement error of the
level of production or other factors, such as the effects of weather, strikes, damaged
product, etc. The greater the amount by which the realized production falls short of
this stochastic frontier production, the greater the level of technical inefficiency”.
Given a Cobb-Douglas form of stochastic frontier production function expressed in
equation (5),
ln (qi ) = β i X i + vi − u i (5 )

Where ln (qi ) is the logarthm of output of the ith farm household, β i is a (k x 1 ) vector of

unknown parameters to be estimated, X i is a (1 x k ) vector of inputs used in the production

of the ith output, vi is a random error measuring statistical noise and ui is a non-negative
error term measuing inefficiency effects, figure(2) shows the graphic illustration of the
production function.

7
 Figure 2: Graphical illustration of stochastic frontier production function

Source: (Neumann ,et al., 2010)

Assuming two farm housholds, A and B , production inputs are shown in the horizonal axis
and outputs are shown in the vertical axis. Farmer A and B use X A and X B level of

inputs to produce q A and q B level of output. The points marked by × shows the observed
values whereas, ¤ shows the frontier values. If there is no inefficiency effects in the
production (i.e. u A = 0 and u B = 0 ) then the frontie level of outputs would be
ln (q A ) = β i X A + v A and ln (q B ) = β i X B + v B for farmer A and B respectively. Further,
the frontier level of output for farmer A lies above the determinstic level of output because
of postive noise effects ( i.e. v A > 0 ) whereas that of farmer B lies below the determinstic

level of output because of negative noise effects ( i.e. v B < 0 )(Coelli ,et al., 2005).

8
3. Literature Review
3.1. Empirical studies on determinants of technincal efficiency

Most empirical studies on productivity and efficiency of farmers indicated that

demographic, socio-economic, institutional, environmental and resource factors are the
major determinants of efficiency differentials among farmers (Battese & Coelli, 1995;
Bravo‐Ureta & Pinheiro, 1997; Obwona, 2006; Nyagaka ,et al., 2010).

For instance, in their analysis on technical efficiency of smallholder farmers in Girawa

district of Ethiopia, Ahmed ,et al. (2013) confirmed that tehcnical efficiency of farmers is
postively associated with education, extension services, livestock holdings and use of
irrigation. Thus, education and extension services increases efficiency of a farmer by
increasing awareness and ablity on the proper use of farm inputs, control of pest and crop
disesases and overall mangement of farm productions. Livestock enhances efficiency
directly through their use in farming operation; and indirectly by financing farm income in
bad production years. Similarly, Asefa (2012) and Khai & Yabe (2011) also confirmed the
importance of education, extension services and irrigation in improving technical efficency
of farmers in their respective studies on Ethiopian smallholder farmers and Vetnamise rice
producers.

Besides demographic and socio-economic factors, environmental conservation also plays

key role in enhancing efficiency of farmers. For instance, in his study on the link between
technical efficiency and environmental conservations in Ghana, Nkegbe (2012) found that
those farmers adopting conservation practices are more technically efficient than non
adopters. Similarly, Solis ,et al. (2007) and Jara-Rojas ,et al. (2012) also confirmed the role
of soil, water and environmental conservations in enhancing technical efficiency of farmers.

The following table gives an overview of the findings of selected empirical studies on
technical efficiency of farmers in Ethiopia. Most of these studies found that education and
trainings of the household, extension services, farm size, off-farm income, access to credit
and other socio-economic variables are the major determinants of technical efficiency of
farmers.

9
Table 1: Summary of selected empirical studies on technical efficiency in Ethiopia

Author/s ( Year ) Product Sample size Mean TE (%) Determinants of Technical efficiency
Ahmed ,et al. (2013) Crops 200 81.5 Education, extension services, farm land, livestock holdings, farmers
training and participation in irrigation
Ahmed ,et al. (2014) Maize 138 88 Family size, extension Services, access to credit and distance from
the nearest market
Alene & Hassan (2003a) Maize 60 76 Education, farm size, access to credit and timely delivery of modern
inputs
Alene & Zeller (2005) Crops 53 79 Improved technologies, education, extension and credit systems

Alemu ,et al. (2004) Crops 254 75.68 Education, access to credit and proximity to markets
Fita ,et al. (2013) Dairy 240 65 Education , exposure to media,Training and experience in dairy
farming
Geta ,et al. (2013) Maize 385 40 Farm size, number of oxen, agro-ecology and use of high yielding
maize varieties
Haji (2007) Vegetables, 150 91 Household asset, family size, farm size, off farm income and
crops and extension services
livestock
Kebede & Adenew (2011)* Wheat 32 82 Distance from main road, managerial capacities and experience,
Value of self-owned machineries
Tirkaso (2013) Crops 562 40.2 Education, commercialization level of the farmer and access for
communication devices such as cellphone and Radio
*Their study is on Commercial Farms

10
3.2. Determinants of participation in off-farm activities
Income from off-farm activities plays a greater role in the livelihood of rural societies
especially for subsistence-oriented households. Off-farm income directly contributes to
households’ income and indirectly influences agricultural productions with potential
implications to policy makers (Kuiper ,et al., 2007). According to Woldehanna (2000), off-
farm income can help farmers to purchease modern inputs, hire labour and reduce the
varablity of farm income and smooth consumption.
Most studies in the area of off-farm business indicated that, demographic characteristic and
financial and resource bases of the household are considered as main factors determining
the decision of participation in off-farm activities. For example, using data on 200
households selected from 40 villages of Southeast Nigeria, Ibekwe ,et al. (2010) examined
factors determining non-farm income. Their findings shows that age of the household,
education level, farm size and hours spent on farm activities are the most significant
variables determining both farm and off-farm income. Specially, the size of farm land is
positively associated with farm income whereas negatively correlated with off-farm
income. This indicates that increases in the size of farm land increases farmers’ willingness
to operate in farm activities than participating in off-farm activities. This may further show
the fact that small-sized farmers are driven out of farm businesses in the study areas.
Besides, they also found positive association between household size and farm income and
negative correlation between age of the household and off-farm income, implying less
participation of older farmers in off-farm activities.
Bedemo ,et al. (2013) studied factors determining the decsions to participate in off-farm
work in western Ethiopia. The finding of their study shows that variables on household
characteristics, access for credit and size of farm land are major determinats of decisions to
participate in off-farm activities. They also noted the importance of off-farm income in
reducing the problems of low agricultural productivity in the study area.
Zahonogo (2011) also examined factors determining participation non-farm activities in
Burkina Faso and results of his study indicate that participation in non-farm activities
mainly related to farm income, technologies in farm production, age and education of the
household head, the number of working individuals in the household and the amount of
rainfall. Accordingly, income from farm activities was found to have a negative effect on
participation in the non-farm activities; whereas, other variables were found to have
positive effects on participation in non-farm activities.

11
On his analysis on factors determining decisions to participate in off-farm activities in
Ethiopia, Beyene (2008) estimated separate models for male and female members of a
given farm households. His result indicates that training on off-farm activities and health
status of the participant has considerable impact on their participation in off-farm activities.
Besides the human capital variables, access to credit and income transfers were also found
to have positive impact on off-farm activity participation. Accordingly, trained farmers are
more likely to participate in off-farm business. In addition, transfer income and credit have
a positive effect on participation in off-farm activities. He also confirmed that female
members of households are less likely to participate in off-farm activities because of
cultural factors and influences of the household head. His study however does not indicate
the impact of off-farm activity participation on farm activities.

3.3. Empirical studies on agricultural production and off-farm income

Studies indicate that the impact of off-farm income on production and efficiency of farmer
is not well understood yet. Even though most studies found positive relationships between
the two variables, some authors also found negative or differential relations (Kumbhakar ,et
al., 1989; Alene & Hassan, 2003b; Bojnec & Ferto, 2011; Chang & Wen, 2011). This
section summarizes different empirical works on the link between off-farm income and
farm output and technical efficiency.
From panel data on Slovenian farmers, Bojnec & Ferto (2011) examined the impact of off-
farm income on technical efficiency. The finding of their study indicated a positive
relationship between off-farm income and technical efficiency. Their result also revealed
that off-farm income increases efficiency of framers over time, showing spillover effects of
income from off-farm activities on farm activities. Furthermore, they also investigated the
relationship between the size of farm and technical efficiency and found positive
correlation.
Using Taiwanese national survey on rice farmers, Chang & Wen (2011) studied the
differences in production, efficiency and risks among farmers with and without off-farm
income. They estimated separate stochastic production functions for each group of farmers.
Their result indicates differential relationships between off-farm incomes and technical
efficiency of farmers. For instance, they found an increasing impact of off-farm income on
technical efficiency for lower percentiles of the technical efficiency distribution. The
authors also argue that lower scores in technical efficiency do not necessarily relate to
participation in off-farm activities because of differences in the use of resources among the

12
two groups of farmers. Accordingly, those farmers without off-farm activities were found to
be more productive in the use of farm inputs than those with off-farm activities. This
implies that farmers without off-farm activities have better knowledge in use of farm inputs
that emanates from their concentration on farm activities. The authors also argue that higher
production risks are associated with farmers participating in off-farm activities.
From the 2002 Chinese household income survey, Yue & Sonoda (2012) analyzed the
impact of off-farm work on technical efficiency of farmers in three regions of the country.
Their result confirms that farm household’s without off-farm wage were found to be
technically more efficient than those with off-farm wage. This is because households with
off-farm wage do not invest their income on farm productions. Furthermore, their findings
also show regional difference of the impact of off-farm wage on technical efficiency.
Similarly, Babatunde (2013), also studied the link between on-farm works and off-farm
works in rural Nigeria. In his study, he examined the impact of off-farm income on farm
level output, purchased inputs and technical efficiency of farmers. The finding of his
research confirmed positive associations between off-farm income and output, purchased
inputs and technical efficiency.
Using Instrumental variable estimation, Woldehanna (2002) examined the linkages between
farm and non-farm activities in northern Ethiopia. His result indicates that non-farm income
plays a vital role in reducing credit constraints of poor farmers. However, the impact of
non-farm income on farm input was found negative due to unfavorable conditions in the
study areas. The author also found that poor households and large families are more likeliy
to participate in off-farm activities than rich households and small sized families. In
general, the findings of his study indicate that non-farm income has a positive and
significant effect on investments in equipment, livestock and buildings, whereas negative
effects on improved seeds and fertilizer.
Iheke ,et al. (2013) examined the impact of remittance on technical efficiency of farmers in
Nigeria. They estimated separate production frontiers for farmers with and without
remittance. The finding of their analysis shows that remittance- receiving farmers are less
technical efficient than non-remittance receiving farmers. Remittance- receivers were found
to have a mean technical efficiency of 42 percent, whereas non- receivers have a mean
technical efficiency of 53 percent. They also indicated that efficiency of non-recivers is
related to educational level of the head of household, size of household, size of farm land
and farming experiances, whereas that of remittance- receivers mainly relates to age of the
head of household and size of farm land.

13
Pfeiffer ,et al. (2009) studied the impact of off-farm income on agricultural production in
Mexico; specifically, they anlyzed the impact of off-farm income on household production,
input demand and technical efficiency. They regressed farm production against off-farm
income and other explanatory variables. The authors also adopt an instrumental variable
approach to handle the problem of endogeneity of off-farm income. The findings of their
study indicate that off-farm income is negatively associated with value of agricultural
output and family labour used in agricultural production, whereas positively associated with
technical efficiency and purchased inputs. Further, the authors also indicated that mean
technical efficiency of farmers with off-farm income is higher than those without off-farm
income.

14
4. Methodology of the Study
4.1. Descrtiption of study areas and data
With an estimated population of more than 90 million, Ethiopia is the second most
populous country in Africa. The country is located in the horn of the continent covering an
area of land of 112.3 million hectares. Agriculture is the pillar of its economy, accounting
for 46.3 percent in gross domestic product (GDP). Out of total land area, 16.4 million
hectares are adequate for production of perennial and annual crops (Deressa, 2007).
According to Dorosh and Gemessa (2013), wheat, teff 2, maize, barley and sorghum
production constitutes the major food crops in the country, accounting for three-fourth of
total area of land under cultivation and 14 percent of GDP. Coffee, pulses, hides, skins,
oilseeds, tea, honey and beeswax are the major agricultural exports of the country.
The study uses Ethiopian Rural Household Survey (ERHS) data conducted by the
Economics department of Addis Ababa University in collaboration with International Food
Policy Research Institute (IFPRI) and Centre for the Study of African Economies,
University of Oxford in 2009. The comprehensive survey was undertaken in four major
regions of the country; Tigray, Amhara, Oromia and Southern Nation’s Nationalities and
People’s Regional State (SNNPRS) covering larger number of peasant associations,
districts, and villages in rural part of the country.

These regions represent for an estimated total land area of 52 percent and a population of 90
percent in the country. The data set mainly covers households’ demographics, asset
holdings, access to credit, expenditures, off-farm income and poverty aspects of the rural
societies (Dercon & Hoddinott, 2004). Based on the availability of data on relevant
variables, 1360 households are involved in this study. Out of the whole sample about 63.6
percent of the households participate in one or more types of off-farm income generating
activities whereas the remaining 36.4 percent are non-participants. As shown figure (3),
about 29 percent of farmers receive remittances, 23 percent engage in traditional labour
sharing, 19 percent participates in unskilled non-farm work and 15 percent participate in
paid farm works. The remaining groups participate in skilled works, paid developmental
activities (such as food-for-work), professional works and other activities.

2
Teff is an annual grass-type cereal grown in Ethiopia.

15
Figure 3: Types of off-farm income generating activities

Types of Off-farm Activities in percent

Domestic Worker (Servant)

Driver
0.58 0.69
Farm Work(for pay)

18.73 14.80 Food- For-Work

Other Activities
4.86
0.92 Professional Worker
3.12
Remittance and Other
transfers
23.35 Skilled Worker

Soldier
29.02
Traditional Labour Sharing

0.35 Unskilled Non-Farm Worker

3.58

Source: Own Computation

4.2. Empirical model specification

4.2.1. Production frontier and technical efficiency
The first main objective of this study is estimating a production frontier and the
corresponding technical efficiency. Hence, the empirical analysis begins with estimating a
production function and efficiency scores of individual households, using a stochastic
frontier model. One of the main advantages of th is model is that the error term captures
measurement errors, statistical noises, exogenous shocks, and contains an inefficiency
component used in the statistical test for the degree of technical inefficiency of farm
household.

The stochastic frontier production function can be specified as:

yi = f ( xi ; β )e vi TEi , (i = 1,2,.......,1360) ( 6)

Where y i is output of i th household, xi is vector of inputs used in the production process by

i th household, f ( xi ; β ) is production frontier, β is a vector of frontier parameters to be

estimated, e vi measures random shocks and TEi is the technical efficiency of the i th farm
household.

16
According to Coelli ,et al. (2005) technical efficiency of an individual farm is the ratio of
observed (actual) output to the corresponding frontier (potential) output. Therefore,
technical inefficiency measures the amount by which the actual level of output falls below
the frontier level. The value of technical efficiency varies between zero and one. If technical
efficiency is exactly equal to one, the actual output yi achieves its potential

level f ( xi ; β ) exp(vi ) . On the other hand, if technical efficiency is less than one, it implies
the presence of technical inefficiency. Technical efficiency therefore can be re-written as;

yi
TEi = , where 0 ≤ TE i ≤ 1 (7 )
f ( xi ; β ) * e vi

Most stochastic frontier studies use either a Cobb-Douglas (CD) or Translog (TL)
functional form for the production functions. In this study, both CD and TL models are
specified and the most appropriate model is selected based on log-likelihood ratio tests 3.
Given different factors of production, the two alternative stochastic production frontiers are
specified respectively as follows:

5
Cobb-Douglas: n( y i ) = β 0 + β i ∑ n( X i ) + vi − u i (8)
i =1
5
1 5 5
Translog: n( yi ) = β 0 + ∑ β k n( X ik ) +
k =1
∑∑ β ij n( X ik )* n(X ij ) + vi − ui
2 k =1 j =1
(9)

Where n is the natural logarithms, β ' s are coefficients of parameters to be estimated, y i

is the total value output in Ethiopian Birr (ETB). In fact the value of output was calculated
in steps. Since all productions were measured in local units, it was first converted to
standard unit, kilograms (kgs). Then, using price conversion, quantity of output was
converted to ETB. X i ' s are factors of prductions, vi is the idiosyncratic error that arises

from measurement errors in input use and/or yield of production and u i is the non-negative
random variables in measuring the technical inefficiency of individual household.
The non-negative error term u i assumes different distributional forms. The most commonly
used distributions are half-normal, exponential, truncated-normal, and gamma distributions,
each having their own assumptions and characteristics. Since there is no prior
justification/reasoning in choosing one distribution over the other, all except the gamma

3
Log-likelihood ratio test results are reported in sub section(5.2.2)

17
distributions are considered in this study. The gamma distribution is not included because of
numerical difficulties that arise when estimating the model parameters.
For example, if it is assumed that the non-negative error term is half-normally distributed,
then the marginal distribution of the composed error ε i = vi − u i is given as;

∞ 2  u2 (ε + u ) 2 
f (ε i ) = ∫ 0
f (u , ε ) du =
2πσ uσ v
exp −
 2σ u
2
−
2σ v2 


2 ε   ελ 
= φ  1 − Φ  for − ∞ < ε i < ∞ (10)
σ σ   σ 
σu
Where σ 2 = σ u2 + σ v2 and λ= are variance parameters and φ (.) is the standard
σv
probablity density function and Φ (.) is the standard cummulative density function.

According to Kumbhakar & Lovell (2003), λ in equation (10) shows the relative

contribution of u i and vi , respectively, to ε i . Hence, as λ → 0 either σ v2 → +∞

or σ u2 → 0 . In this case vi (the symmetric error) dominates u i (one–sided error). On the

other hand, as λ → +∞ either σ v2 → 0 or σ u2 → +∞ , u i dominates vi in determining ε i .

Given the marginal distribution of the composed error in equation (10), the next step is
forming and maximizing the likelihood function with respect to β , σ 2 , λ to obtain the

estimate of unknown parameters and consequently the measure of inefficiency, u i .

However, it is impossible to decompose the estimates of the error term, εˆi , into v̂i and û i .

The solution is to take the conditional distribution of u i given ε i as it was first proposed by

Jondrow ,et al. (1982). The conditional distribution of u i given ε i is therefore given as
follows:

E (ui | ε i ) = uˆi =
σλ  φ ελ

( σ  ελ) 

1 + λ2 1 − Φ − ελ

(σ
−
σ ) 


(11)

Finally, given the point estimates of ui in equation (11), the technical efficiency of the
individual household is:

TE i = exp(−uˆ i ) (12)

18
Several empirical studies on productivity and efficiency argue that demographic, socio-
economic, institutional and environmental factors attribute to efficiency differentials among
farmers (Aigner ,et al., 1977; Battese & Coelli, 1995; Bravo‐Ureta & Pinheiro, 1997;
Obwona, 2006; Nyagaka ,et al., 2010). Following these studies and availability of data, ten
variables have been used to explain efficiency differentials among households.
Accordingly, demographic factors (household size, age, gender and education level of the
household head), institutional factors (access to extension services, use of irrigation, manure
and soil conservations practices) and resource factors (number of livestock owned and
income generated from off-farm activities) are used as explanatory variables in the
inefficiency model specified in equation (13) :

10
U i = δ 0 + ∑δ i Z i + ϕi (13)
i =1

Where U i is technical inefficiency, δ i ' s are the parameters to be estimated, Z i ' s represent
a set of explanatory variables explaining technical inefficiency. These includes, household
size, age, gender and education level of the household head, soil conservation practices, use
of extension services, irrigation and manure, total off-farm income of the household and ϕ i

( )
is the random error term ϕ i ~ N (0, σ ϕ2 ) .

There are in general two methodological approaches in the estimation of inefficiency model
in stochastic frontier analysis. The first approach is a one-stage procedure in which
production function and sources of technical efficiency are estimated simultaneously. The
second approach a two-stage estimation technique in which the stochastic production
frontier and efficiency scores are estimated in the first stage and the derived efficiency
scores from the first stage are regressed on a set of explanatory variables using OLS.
However, this approach has been criticized because farmers’ knowledge of its level of
inefficiency may affect choice of inputs (Chirwa, 2007). Given these facts, estimation of
technical efficiency scores and determinants of inefficiency will be estimated based on one-
stage estimation technique.

19
List of explanatory variables and expected signs
Labour ( X 1 ): is total number of family and hired labourers used in different stages of
production such as land prerpartion, planting and general cultivation and harvesting.

Land ( X 2 ): is total area of land used for main crop production in hectare (ha). Land is the
most important input in smallholder production. In this study, land refers to the total area of
farm land used for main crop production. It doesn’t include marginal, set aside lands and
grazing lands.

Fertilizer ( X 3 ): is the total amount of modern fertilizers used in kgs.

Farm instruments ( X 4 ): is number farm equipment such as ploughs, hoes, sickles and other
farm instruments owned by the household.

Oxen ( X 5 ): is total number of oxen owned by the household.

In general, several studies have shown positive association between inputs and output in
production. This implies that households with more inputs are expected to produce more
output. Therefore, positive association between the dependent variable (value of output) and
all explanatory variables (inputs) is expected.

Age (Z 1 ) : is an indication of experience and capacity of the household in agricultural

operations. Emprical studies for example Tan ,et al. (2010) and Etim and Okon (2013)
argue that older households are more experienced than younger ones. Therefore, age is
expected to have positive impact on farm output and the level of technical efficiency of the
farmer.

Household Size (Z 2 ) : represents the number of household members. In smallholder

production the size of household is a means to have more supply of labour. Hence, the
larger the household size the higher level of production they produce and the higher
technical efficient they will be. Hence, size of the household is expected to have positive
association between production and technical efficiency of the farmer.

Gender of HH Head (Z 3 ) : is a dummy variable representing the gender of household head

taking a value of 1 for male headed households and 0 for female headed households. Even
though women play a substantial role in agricultural activities, there are still tasks that are

20
not done by women for example ploughing. Hence, male headed households are expected to
be more efficient than female headed households. Besides, female headed households may
also have to perform additional tasks such as taking care of children and therefore they may
have to allocate their time between these tasks and actual farm activities.

Education (Z 4 ) : indicates the years of schooling of the household head. Education is usually
an indication for quality of labour. It is argued that educated (skilled) farmers to have better
skills of managing farm operations and understand new technologies that increase their
production. Hence, education is expected to be positively associated with farm output and
the level of technical inefficiency of the farmer.

Livestock owned (Z 5 ) : agricultural productivity can also be affected by the number of

livestock owned by the farmer. Besides their use in different stages of agricultural activities,
revenues from selling livestock supplement farm income in case of a crop failure. Hence,
those farmers owning more livestock are expected to be more technically efficient.

Soil conservation (Z 6 ) : households practice different soil conservation mechanisms to

sustain the productivity their farm land. It is obvious that conserved farm land produces
more output than non-conserved lands. Hence, those households practicing soil
conservation are expected to be more technically efficient. Soil conservation is a dummy
variable that assumes 1 if the farmer conserve soil and 0 otherwise.

Extension Services (Z 7 ) : is a dummy variable taking values of 1 if an extension expert visits

the farmer and 0 otherwise. Those households who are visited by extension experts are
expected to have more information which enables them to produce efficiently. Hence, the
impact of extension services on the level of efficiency is expected to be positive.

Use of irrigation (Z 8 ) : is a dummy variable that assuming 1 if the farmer uses irrigation and
0 otherwise. Basically, the majority of the farmers in Ethiopian practice rain-fed agriculture
and productions are usually once a year in most parts of the country. Hence the use of
irrigation increases total output produced in a given year. Irrigation is therefore expected to
have a positive impact on production and technical efficiency of farmers.

Use of Manure (Z 9 ) : is a dummy variable assuming 1 if the household uses manure and 0
otherwise. Studies indicate that, those households using manure are more efficient in

21
production than those who are not using manure. Hence, it is hypothesized that farmers
using manure are more efficient than others.

Off-farm Income (Z 10 ) : off-farm income is the variable in the main focus in this study. It
represents total income generated from any type off-farm activities including remittances.
As explained in earlier sections, the relationship between off-farm income and level of
technical efficiency is quite unclear. Off-farm income enhances the production and
productivity of farmers if incomes from off farm activities are spent as investment in
farming sector, whereas participation in off-farm activities have negative effects on farm
activities when off-farm incomes are used for consumption or investment in other sectors.
Existing evidence such as Beyene (2008) reveals that Ethiopian farmers usually participate
in off-farm activities mainly to overcome their liquidity constraints. Hence, Off-farm
income is expected to positively impact on their farm production and on the level of
technical efficiency.
4.2.2. Off-farm income and Farm output
Off-farm income has direct and indirect effects on agricultural production. The direct effect
is related to the absence of labour (lost-labor effect) in farm operations; whereas, the
indirect effect is related to the investments of off-farm income on farm activities (López-
Feldman ,et al., 2007).
The impact of off-farm income on the value of farm outcome can be analyzed by comparing
output of recipents of off-farm income and non-recipents, while controlling other
factors.Thus, systematic difference among the two groups of farmers are captured through
the inclusion of explanatory variables on household characteristics (Kilic ,et al., 2009).
Accordingly, following, López-Feldman ,et al. (2007), Kilic ,et al. (2009) and Babatunde
(2013), the value of farm output is regressed against off-farm income and other explanatory
variables and the model for farm outcome is specified as follows:
6
Yi = β 0 + β 1OI i + β i ∑ X i + ε i (14)
i =2

Where Yi the value of farm output is, β ' s are parameters to be estimated, OI i is off-farm

income and X ' s are a set of variables on household characteristics and resource factors
mentioned above in section (4.2.1) . This includes, household size, age, gender, education

level and marital status of the household head, size of farm land, and ε i is the error term.
The impact of off-farm income on farm outcome is therefore determined based on the
estimated coefficient of off-farm income in equation (14) . However, studies on the linkage

22
of farm and off-farm income, for example Babatunde (2013), indicates the possibilities
where productivity of farming sector depends on off-farm income and vice versa. Thus,
there might be reverse causality problem that leads to endogeneity bias. In fact, endogeneity
test confirms the presence of endogeneity problems in the model. As a result, the OLS
method results in biased estimates for unknown parameters. Therefore, instrumental
variable regression framework is introduced to overcome this problem. Accordingly, off-
farm income is instrumented by other proxy variables such as access to credit, radio and
distance from the nearest town.
As evidenced by some studies for example Woldehanna (2002),Beyene (2008), Babatunde
,et al. (2010) and Babatunde (2013), income from off-farm activities is also determined by
the availability of credit, access of new information and distance to the nearest town besides
household characteristics and resource factors. Theoretically, access to credit and radio or
related telecommunication technologies should have a positive effect on off-farm activity
participation, while distance from the nearest town has a negative impact. Therefore,
positive association between off-farm income and access to credit and radio is expected;
whereas negative association between distance from the nearest town and off-farm income
is expected.
4.2.3. Estimation Technique

Because of the non-negativity assumption of the inefficiency term ( u i ), the underlying

distribution is non-normal and the error terms are therefore asymmetrically distributed.
Hence, the ordinary least squares (OLS) estimator is inefficient. Coelli (1995) argues that
there are two main benefits of using frontier functions as the ones in equation (8) and (9),
rather than OLS (average functions). First, frontier functions are based on best performing
producers and therefore can show the technology used in production, whereas average
fucntions only shows technology of an average producer. Secondly, frontier functions show
the best-practicing technology, whereas average functions show efficiency of producers
within the whole group. Schmidt & Knox Lovell (1979) also argued that MLE technique
provides more efficient estimates than the OLS, besides its guarantee of non-negativity
assumptions on the error terms. Thus, unknown parameters of frontier models are estimated
using the maximum likelihood estimation (MLE) technique

23
5. Result and Discussion
5.1. Descriptive Statistics
As shown in the summary statistics in table (2), the average age of household in the whole
sample is 52 years. Male headed households’ accounts for 67 percent of the whole sample. This
implies that on average about 67 percent control and management of agricultural resources and
activities are male’s responsibility. Besides, 69 percent of the surveyed households are married.

Table 2: Summary statistics of model variables

Variables Mean Sta.Dev.

Age ( Years) 52.212 14.71

Male headed HH (Dummy, 1 if male & 0 otherwise) 0.67 0.47

Household size (Number) 5.71 2.55

Education (Years of schooling) 1.94 2.89

Married (Dummy, 1 if married& 0 otherwise) 0.69 0.46

Output (ETB) 6086.24 6480.57

Land(Ha) 0.56 0.647

Labour( Number) 14.69 19.42

Oxen( Number) 1.13 1.34

Livestock( Number) 5.34 5.76

Farm equipment (Number) 11.85 8.92

Fertilizer(Kg) 74.68 125.93

Off-farm income 1418 3836

Use of irrigation(Dummy, 1 if s/he uses& 0 otherwise) 0.10 0.31

Soil conservation(Dummy, 1 if s/he conserves& 0 otherwise) 0.504 0.50

Access to extension services(Dummy, 1 if s/he have & 0 otherwise) 0.48 0.50

Use of manure(Dummy,1 if s/he uses& 0 otherwise) 0.07 0.26

Access to Credit(Dummy,1 if s/he have& 0 otherwise) 0.57 0.49

Association membership(Dummy,1 if s/he is & 0 otherwise) 0.15 0.35

Access to radio(Dummy,1 if s/he has& 0 otherwise) 0.58 0.49

Access to cellphone(Dummy, 1 if s/he has& 0 otherwise) 0.11 0.31

Distance from the nearest market/town (kilometers) 10.25 5.68

Source: Own Computation

24
The average schooling years of the farm household is 2 years. The mean value of output is
6086.24 ETB. Regarding farm inputs, the average values are 0.56 hectares of land, 15
Labourers, 12 units of farm equipment, 5 livestock and 75 kilograms of fertilizers. The
average off-farm income of the household is 1418 ETB. Use of extension services,
irrigations and manure among farmers accounts for 48, 10 and 7 percent respectively.
About 50 percent households in the survey sites practice soil conservation. On the other
hand, 57 percent of households have access to credit and 51 percent have access to either
radio or tape whereas only 11 percent of farmers have access to cellphone. In addition, 15
percent of the households are members of one or more association. The average distance
from the nearest town/market is about 10 kilometers.
As presented in table 2, the highest value standard deviation of value of output (i.e. 6480.57)
shows the degree of variation in output level in the sample. In fact, the distributions of
output per worker and output per hectare also confirm the variability of output. As shown in
figure 4, about 24 percent of sampled farmers have output per worker less than 100. About
72 percent of farmers have output per worker between 100 and 1000; whereas, the
remaining 4 percent of sampled farmers have output per worker greater than 1000.

Figure 4: Distribution of output per worker

Source: own computation

25
Similarly, output per hectare also shows substantial variation across the sampled farmers.
As presented in figure 5, about 13 percent of the farmers have output per hectare less than
1000. The majority of the farmers (i.e. 84 percent) have output per hectare between 1000
and 10000. The remaining 3 percent of farmers have output-per hectare greater than 10000.

Figure 5: Distribution of output per hectare

Value of output per hectare

16.00
14.00
12.00
10.00
Percentages

8.00
6.00
4.00
2.00
0.00

Value of output per hectare category

Source: own computation

5.2. Econometric Results

5.2.1. OLS Estimates of Production Frontier

The production function in equation (8) is first estimated using OLS to identify potential
inputs that will be included in maximum likelihood estimation of production function. The
estimation result indicates that 80 percent of variation in output is explained by inputs of
production; land, labour, fertilizer, oxen and farm equipment.
As reported in table (3) land, fertilizer, labour and oxen are significant at 1 percent
significance level and use of farm equipment is significant at 5 percent. Besides, the
presence of multicollinearity and heteroskedasticity is checked using variance inflation
factors (VIF) and the Breusch-Pagan test, respectively. The test result confirms that there
are no multicollinearity and heteroskedasticity problems in the model. Hence, all inputs are
included in the maximum likelihood estimation of production function.

26
Table 3: OLS estimates of CD production function: The dependent is log of Value of Output

Variables Coefficients P>|z|

Constant 6.63* 0.000

n (labour ) 0.10* 0.005

n (land ) 0.73* 0.000

n ( fertilizer ) 0.13 * 0.000

n ( farm equpment ) 0.17** 0.012

n (oxen) 0.29* 0.000

R-squared /Adj. R-squared 0.8015 / 0.7947

No. of Observation 1360

* p< 0.01, ** p < 0.05, Source: Own Computation

5.2.2. Test for Model Specifications
Before estimating model parameters using the MLE method, testing specification and
validity of the model is a vital task. Thus, the following hypotheses are tested to identify the
appropriate functional form and estimation method for the dataset in scope. The first
hypothesis aims at testing the adequate functional form. Accordingly, both CD and TL4
production functions in (8) and (9 ) are estimated using maximum likelihood estimation
method followed by a log-likelihood ratio test (LR test) to choose the most appropriate
production function. The generalized LR test statistic is given as:
λ = −2og [L( H 0 ) /( H 1 )] = −2{og [L( H 0 )] − og [L( H 1 )]}
Where og [L( H 0 )] and og [L( H1 )] are values of likelihood functions under null

hypothesis H 0 (CD function) and alternative hypothesis H 1 (TL function) respectively. In

this test, we reject the null hypothesis when the value of the calculated likelihood ratio
( )
exceeds the critical likelihood ratio χC 2 . The value of calculated LR equals to 17.08,
15.93 and 16.50 in half-normal, exponential and truncated-normal respectively; while the
(
critical LR at upper 5 percent level of significance and 15 degrees of freedom χC 2 0.05,15 is)
equal to 24.99 as published in the article of Kodde & Palm (1986). Given these values, we
cannot reject the null hypothesis that states that coefficient values of all interaction and

4
MLE estimate of TL function is presented in appendix (2).

27
second order terms of the TL production function are equal to zero. Hence, the Cobb-
Douglas production function is more appropriate in representing the data. This study is
therefore based on estimation results from the CD form of the production function. One of
the main advantage of the CD form is the ease at which returns to scale can be measured.
Thus, the return to scale can be constant, decreasing or increasing based on the sum of
exponents (Neumann ,et al., 2010).
The second hypothesis is checking whether the stochastic production frontier is more
appropriate than a conventional production function, i.e. testing whether there exists
technical inefficiency in the production process or not. In this test the null hypothesis is
given as H 0 : γ = δ 1 = δ 2 = ......... = δ 10 = 0 where γ = σ u2 / σ 2 and 0 ≤ γ ≤ 1 . The closer

the value of γ is to zero, the more likely OLS represent the data and the closer the value of
γ is to one, the more likely a stochastic frontier model represent the data best (Piesse &
Thirtle, 2000). Accordingly, the value of γ in the CD function is 0.878, 0.667 and 0.985 in
half-normal, exponential and truncated-normal models respectively. This implies that
farmer’s technical inefficiency accounts for 88 percent, 67 percent and 99 percent in the
variations of actual output from its potential level in the respective models. Likewise, the
value of γ in the TL function is 0.851, 0.615 and 0.844 in half-normal, exponential and
truncated-normal distributions respectively. Given these values the null hypothesis is
rejected confirming the presence of technical inefficiency in the data set.
On the other hand, the choice of the estimation method to be adopted can also be checked
based on the skewness of residuals from the OLS regression. Thus, if the third moment is
negative, then OLS residuals are negatively skewed, indicating the presence of technical
inefficiency. In contrast, positive skewness of the third moment suggests the absence of
inefficiency which in turn implies the appropriateness of the OLS method in the estimation
of model parameters (Waldman, 1982; Kumbhakar & Lovell, 2003; Van Keilegom &
Wilson, 2012). Given these facts, the kernel density estimates of residuals of the OLS
regression presented in appendix (3) showed negative skewness. This implies that the
maximum likelihood estimation method provides more consistent estimates than ordinary
least squares.
Besides, identifying the most representative model among the three distributions (i.e. half-
normal, exponential and truncated-normal distributions) is the basic task. According to
Greene (2005), a distribution with relatively smaller variance is more representative. Thus,

28
comparing the kernel density5 plots of efficiency scores of the three models, exponential
model is more representative than the other two models. Half-normal and truncate-normal
models cannot represent the data because of higher variance in the half-normal model and
statistically insignificance of variance component ( σ u ) and inefficiency indicator lambda

( λ ), in truncated model as shown in table 4. Hence, estimation of determinants of technical

efficiencies is based on exponential model.
5.2.3. MLE Estimates of Production Frontier

The MLE estimates of CD production function based on the three distributional

assumptions is presented in table (4). The result shows that all inputs are positive and
statistically significant at 1 percent in all the three models. The log likelihood values of the
distributions are also similar. Lambda ( λ ), an indicator of the presence of inefficiency, is
greater than zero and significant at 1 percent in half-normal and exponential models and
insignificant in truncated model. Besides, variance components σ u and σ v are also
significant at 1 percent significance level in half-normal and exponential models.

However, the values of σ u and µ are found insignificant in truncated-normal. This implies
that deviation of output from its frontier output is because of statistical noise. In reality
however, there is always a deviation of output from its potential level in any production
process, which entails the contribution of the inefficiency term error ( σ u ) in the total error.

On the other hand, the mean technical efficiency is closely similar in exponential and
truncated-normal distributions whereas relatively smaller in the half-normal model. The
mean technical efficiency of households is 42.7, 52.9 and 52.1 in half-normal, exponential
and truncated normal, respectively. In fact, Technical efficiency is a relative concept and
these values imply that farmers are on average 43, 53 and 52 percent efficient compared to
the most efficient farmer in the respective models. In general, the values are consistent with
the study by Asefa (2012), Geta ,et al. (2013) and Tirkaso (2013).

5
kernel density plots of the three distributions are presented in appendix (3)

29
Table 4: MLE estimates of production function in half-normal, exponential and truncated models

Variables Half-Normal Exponential Truncated-Normal

Coeff. P>|z| Coeff. P>|z| Coeff. P>|z|

Constant 7.510* 0.000 7.19* 0.000 7.222* 0.000

n (labour ) 0.085* 0.007 0.078* 0.006 0.078* 0.006

n (land ) 0.824* 0.000 0.951* 0.000 0.937* 0.000

n (fertilizer ) 0.108* 0.000 0.111* 0.001 0.111* 0.001

n ( farm equpment ) 0.152* 0.005 0.145* 0.004 0.145* 0.003

n (oxen) 0.227* 0.000 0.207* 0.000 0.209* 0.000

sigma_u 1.314* 0.001 0.781* 0.003 4.389 0.130

sigma_v 0.488* 0.000 0.551* 0.000 0.544* 0.000

lambda 2.683* 0.004 1.416* 0.000 8.066 0.110

Mu - - -22.453 0.160

Log-Likelihood -1732.545 -1723.195 -1723.010

Mean TE 0.4271 0.5299 0.5208

No. of Observation 1360 1360 1360

* p< 0.01 , Source: Own Computation

5.2.4. Distribution of technical efficiency

The estimated result shows a higher variation in technical efficiency score among farm
households. As presented in the summary of technical efficiency in appendix (5), the
technical efficiency of farm households ranges from a minimum score of 14 percent to a
mximum score of 89 percent. As shown in figure (6), about 12 percent of the households
have an efficiency score less than 20 percent. Whereas, 25 percent of farm households have
efficiency score between 20 and 50 percent.The majority of the household (i.e. about 60
percent) have efficiency score between 50 and 80 pecent. Only 3 percent of farm
households have efficiency score gretare than 80 percent. In general, such a wider variation
in efficiency scores indicates farmers’ inefficiency in utlilizing their resources; which
further implies the existence of wider scope for improvement of their efficiency.

30
Figure 6: Distribution of technical efficiency

Distribution of technical efficiency

30

25

20
Percentages

15

10

0
Less than 21-30 31-40 41-50 51-60 61-70 71-80 Greater
20 than 81
Range of TE Scores in percentages

Source: Own Computation

5.2.5. Elasticity of factors of production
The estimated elasticity6 of output with respect to labour, land, fertilizer, farm equipment
and oxen in this model is 0.08, 0.95, 0.11, 0.15 and 0.21 respectively. This implies that,
keeping all othr inputs constant, a 1 percent increase in hectares of land, the number of
labour, farm equipment, oxen and kilograms of fertilizer leads to 0.08, 0.95, 0.11, 0.15 and
0.21 percent in the level of output. The result confirms that land is the most important input
followed by number of oxen owned by the household.
Furthermore, the sum of elasticity of all inputs is greater than one, indicating that
production function exhibits increasing returns to scale. Thus, a proportional increase in all
inputs would increase output more than the proportional increase in the inputs. Such an
increasing return to scale is also confirmed by Chirwa (2007) and Asefa (2012). In other
words: Increasing inputs beyond the current level is still beneficial which implies that the
farmers in the sample should grow and use all inputs more intensively.
5.2.6. The effect of off-farm income on technical efficiency
This section presents estimation results of the inefficiency model specified in equation (13).
As explained in section 4.2.1, it is not possible to use instruments for endogenous variables
(off-farm income) in the single-stage estimation that simultaneously estimates the stochastic
production frontier and the efficiency model. Therefore, predicted value of off-farm income

6
The estimated elasticities are based on exponential model presented in table 4.

31
from the first-stage estimation of off-farm income regression in section (5.2.7) is used to
address problem of endogeneity in the estimation of determinants of technical efficiency.
The MLE results of the inefficiency model (as presented in table 5) show that all
explanatory variables except use of irrigation and manure have the expected sign.
However, age and number of livestock owned are not statistically significant implying their
small/nil importance in improving the technical efficiency of households. All other
variables are in line with prior hypothesis discussed in the third chapter. Accordingly,
household size, education of the head, off-farm income, soil conservation and extension
services are significant at 5 percent while gender of the household head is significant at 10
percent.

Besides, the coefficient of household size also shows a positive association between
household size and technical efficiency. This implies that households with larger family size
are more efficient; indicating their higher possibilities of having larger labour supply during
peak agricultural seasons. This result is consistence with Obwona (2006), Msuya ,et al.
(2008) and Al-hassan (2012).
Educational level of the household head is also positively associated technical efficiency. In
fact, education is usually considered as an indication of higher possibilities of literate
households in having better managerial skills, access and understanding of information on
improved methods to their farm operations. Thus, households with more schooling years are
more technically efficient. Such association between education and technical efficiency is
also confirmed by Battese & Coelli (1995), Liu & Zhuang (2000), Khai & Yabe (2011) and
Tirkaso (2013).
Gender of the household head has a positive impact on the level of technical efficiency.
This means that male headed households are more technically efficient than female headed
households. Msuya ,et al. (2008) also found the same correlation between gender of the
household head and technical efficiency.
The coefficient of soil conservation reveals positive association between conservation
practices and technical efficiency. According to Jara-Rojas ,et al. (2012), soil conservation
practices not only increase production and productivity of farm land, but also enhances
environmental sustainability. Similarly, Solis ,et al. (2007), also found positive correlation
between soil conservation practices and technical efficiency in their studies on the level of
technical efficiency under different degrees of soil conservation practices in Central
America.

32
Table 5: MLE estimates of determinants of technical inefficiency: The dependent variable is
technical inefficiency

Variables Coeff. 7 P>|z|

Constant -1.38* 0.001

Z 1 ( Age) -0.17 0.180

Z 2 ( HH Size) -0.23** 0.041

Z 3 (Gender ) -0.14*** 0.072

Z 4 ( Education) -0.16** 0.035

Z 5 ( Livestock ) -0.15 0.130

Z 6 ( Soil Conservation) -0.36** 0.023

Z 7 ( Extension Services) -0.09** 0.037

Z 8 ( Irrigation) 0.35 0.240

Z 9 ( Manure ) 0.04 0.160

Z 10 (Off − farm Income) -0.48** 0.029

Log likelihood -1455.2437

No of Observation 1360

* p< 0.01, ** p < 0.05 , *** p < 0. 10 , Source: Own Computation

Extension services provided to households plays crucial role in creating capacities to

improve overall performances of farm productions through access to better information on
new technologies. The estimated coefficient of extension services in this study also
confirms the positive impact of extension services on technical efficiency of farmers.
Similar effects of extension on technical efficiency of farmers are also reported in other studies
by Obwona (2006), Binam ,et al. (2008) and Nyagaka ,et al. (2010).

Off-farm income is the most primary variable of analysis of this study. According to Amare
(2005),off-farm incomes can have positive effect on the households’ efficiency by
alleviating financial constraints in terms of timely purchase of farm inputs. On the other

7
A negative value of estimated coefficients implies an increase in the level technical efficiency.

33
hand, participating in off-farm activities might be at the expense of own farm activities in
terms of less labour and time causing for a negative relationship between technical
efficiency and participation in off-farm activities. In this study, the estimated coefficient of
off-farm income shows a positive and significant association between off-farm income
technical efficiency. This implies that an increase in off-farm income increases the technical
efficiency of farmers. Such a positive association between the two variables is also reported
in other studies by Rizov ,et al. (2001), Alene & Hassan (2003b) and Bojnec & Ferto
(2011). However, it is against the studies by Kumbhakar ,et al. (1989), Singh ,et al. (2009)
and Geta ,et al. (2013) who found negative correlations; and Chang & Wen (2011) who
confirmed differential effects of participation in off-farm activity on technical efficiency of
farm household.
5.2.6.1. Sensitivity Analysis
Separate production frontiers for farmers with off-farm and without off-farm income are
estimated to further explore the effects of off-farm income on the technical efficiency. The
result indicates that farmers with off-farm income have higher mean technical efficiency
than those without off-farm income. As shown in table 6, the mean technical efficiency is
58 percent for farmers with off-farm income and 51 percent for those without off-farm
income. The estimated elasticities of factors of production are slightly higher for farmers
with off-farm income than those without off-farm income. Specifically, the elasticity of
output with respect to fertilizer and farm equipments is for those farmers with off-farm
income.This may show the inverstment of income from off-farm income on farm inputs.
In addition, the mean technical efficiency of farmers increases with the increases in off-
farm income. Accordingly, the first 25% least efficient farmers have a mean technical
efficiency of 38 percent with an average off-farm income of 1739 ETB. On the other hand,
the second 25% least efficient groups have a mean technical efficiency of 49 percent with
an average of off-farm income of 2258 ETB. The top 25 percent most efficient farmers have
a mean technical efficiency of 79 percent with a mean off-farm income of 2616 ETB.

34
Table 6: Production frontiers 8 of farmers with and without off-farm income

With Off-farm Income Without off-farm Income

Variables
Coeff. P>|z| Coeff. P>|z|

Constant 6.672* 0.000 7.405* 0.000

n (labour ) 0.152** 0.039 0.046** 0.036

n (land ) 0.978* 0.000 0.920* 0.000

n ( fertilizer ) 0.138* 0.000 0.096* 0.000

n ( farm equpment ) 0.182** 0.023 0.152** 0.039

n (oxen) 0.202** 0.015 0.201** 0.027

sigma_u 0.705* 0.000 0.839* 0.000

sigma_v 0.528* 0.000 0.556* 0.000

lambda 1.333* 0.000 1.509* 0.000

Log-Likelihood -858.894 -890.575

Mean TE 0.5786 0.5104

No of Observation 698 662

* p< 0.01, ** p < 0.05 , Source: Own Computation

5.2.7. Effects of off-farm income on farm output

Before estimating the effects of off-farm income on farm output, endogeneity of off-farm
income is checked using Durbin-Wu-Hausman test and the test statistic 9confirmed
endoginity of off-farm income. Therefore, testing the relevance and exogeneity of
instruments is an important task in IV regression. Firstly, instruments are said to be relevant
if they are correlated with the endogenous variable and at least one of the instruments have
a coefficient value significantly different from zero. The following table presents first-stage
estimation results of off-farm income specified in equation (14) to show the relevance of
instruments. The estimation result shows, all instruments (Access to credit, radio and
distance from the nearest town) are statistically significant in determining off-farm income.

8
The production frontiers are based on exponential distribution and estimation results of half-normal and
truncated-normal distributions are presented in appendix 6 and 7 for both groups of farmers respectively.
9
The Durbin-Wu-Hausman test statistic χ2 test has a value of 0.762 with p-value of 0.006.

35
Table 7: First-stage estimates of off-farm income: the dependent variable is off-farm income

Variables Coefficient P>|z|

Constant -468.10** 0.018

Household size 102.26** 0.026

Age 17.88 0.160

Gender 570.39 0.220

Educational level 134.32*** 0.076

Martial status 60.89 0.127

Farm land -301.38** 0.015

Access to credit 406.05** 0.022

Access to radio 289.66*** 0.089

Distance from the nearest market -439.47* 0.000

R-Squared/Adj. R-Squared 0.8422 / 0.8301

Number of Observation 1360

* p< 0.01, ** p < 0.05 , *** p < 0. 10 , Source: Own Computation

Accordingly, accesses to credit and radio have positive impact in determining income from
off-farm activities while; distance from the nearest town has negative impact on off-farm
income. On the other hand, other variables, such as size of the household, the level of
education of the head and farm land are found statistically significant with prior expectation
while age and marital status are found insignificant.
As shown in table (7), larger families tend to generate more income from off-farm activities
than small-size families. Thus, keeping all other variables constant, having an additional
family member increases off-farm income by 102 ETB. Similarly, an additional years of
schooling increases off-farm income by 134 ETB. Access to credit and access to radio also
increases off-farm income by 406 and 290 ETB respectively. Size of farm land and distance
from the nearest town however decreases off-farm income by 301 and 439 ETB
respectively. The argument for these is that, those households owning larger farm lands are
more likely to concentrate on farm activities than participating in off-farm activities.
Besides, households with longer distances from the nearest town are less likely to earn off-
farm income because of lower possibilities of participation in off-farm activities. In general,

36
these results are in line with prior expectation and consistent with other studies by
Babatunde (2013), Beyene (2008) and Ibekwe ,et al. (2010).
Secondly, instruments are said to be exogenous if they are uncorrelated with the error term.
The Sargan test for over identification is used to test the validity of instruments. In this test
the null hypothesis states that over identifying restrictions are valid and therefore rejection
of the null shows invalidity of the instruments. Given these facts, the Sargan χ2 has a value
of 4.27 with p-value of 0.48 implying the insignificance of over identifying restrictions.
Thus, the null hypothesis cannot be rejected and the instruments are therefore not correlated
with the error term.
Table (8) presents the two stages least square (2SLS) estimates of farm output. The
estimation result indicates that gender, level of education, farm land and off-farm income
are statistically significant in determining farm output. Accordingly, the positive coefficient
of gender indicates that male headed households produce higher output than female headed
households. Consequently, keeping all other variables constant; being male household head
increases farm output by 119.21 ETB.

Table 8: 2SLS estimates of farm output: the dependent variable is value of farm output

Variables Coefficient P>|z|

Constant 6103.05* 0.000

Household size 119.21 0.173

Age 42.35 0.152

Gender 97.53** 0.014

Educational level 12.21*** 0.069

Martial status 74.27 0.193

Farm land 1442.21** 0.025

Off-farm Income 14.25** 0.019

R-Squared/Adj. R-Squared 0.5622/0.5501

Number of Observation 1360

Instrumented: Off-farm Income

Instruments : Household size, Age ,Gender, Educational level, Martial status, Farm land,
Access to credit, Access to radio and Distance from the nearest town

* p< 0.01, ** p < 0.05 , Source: Own Computation

37
The coefficient of education is also found positive and statistically significant in
determining the value of farm output. This indicates the fact that educated farmers have
better capacities of using modern inputs in their productions. Therefore, every additional
year of schooling will increases farmer’s output by 12.21 ETB. Furthermore, the effect of
off-farm income on farm output is also positive and every additional ETB of off-farm
income increases farm output by 14.25 ETB. In addition, the coefficient of land showed a
positive and significant effect on farm output. In fact, land is the most important input in
smallholder’s production. Thus, an additional hectare increase in the size of farm land
increases farm output by 1442.21 ETB.

38
6. Conclusion and Policy Recommendation
This study mainly focused on the analysis of the relationship between off-farm income and
technical efficiency of smallholders in Ethiopia. The study used a stochastic frontier model
in the derivation of individual efficiency scores and estimation of factors determining
technical efficiency in smallholder farming. The Cobb-Douglas form of the production
function was found to be more appropriate in representing the data than the translog.
Positive elasticities of output with respect to land, labour, fertilizer, farm equipment and the
number of oxen revealed the importance of these inputs in smallholders’ production.
Three distributions, namely half-normal, exponential and truncated-normal distributions are
considered to represent the inefficiency error term. The estimation results show that the
mean technical efficiency of farmers is 43 percent, 53 percent and 52 percent in each
model, respectively. In fact, these figures are in line with some studies in the area of
technical efficiencies in the country. Besides, the test for inefficiency effects reveals the
existence of inefficiency among the surveyed farmers; that necessities the use of frontier
models.
Variables on demographic, institutional and resource factors were used to explain technical
efficiency differentials among farmers. An instrumental variable regression framework was
used to address the problem of endogeneity of off-farm income in determining the farm-
output and the degree of technical inefficiency respectively. Accordingly, off-farm income
is instrumented by proxy variables such as access to credit, radio and distance from the
nearest town.
Maximum likelihood estimation results show that all explanatory variables except use of
irrigation and manure have the expected sign. However, age of the household head and
number of livestock owned are not statistically significant. All other variables are in line
with prior hypothesis. Thus, efficiency of the farmers mainly related to demographic
characteristics such as household size, years of schooling and gender of the household head;
extension services, soil conservation practices and off-farm income. Accordingly, the size of
the household is positively associated with technical efficiency.This implies that
households’ with larger families are more efficient; indicating their higher possibilities of
having larger labour supply during peak agricultural seasons.
Education of the household head and extension services are also positively associated with
technical efficiency of farmers.

39
The effect of off-farm income on technical efficiency and farm output is positive. The result
of empirical analysis shows that a one ETB increase in off-farm income increases farm
output by 14.25 ETB. Similarly, the MLE estimation result of the inefficiency model also
confirms the importance of off-farm income in improving the level of technical efficiency
of farmers. Accordingly, the mean technical efficiency of farmers with off-farm income is
58 percent whereas, those without off-farm income is 51 percent. This shows spillover
effects of off-farm income on on-farm activities; possibly as investment on modern inputs
that enhances farm production.
In general, the technical efficiency of farmers varies from 14 to 89 percent. This indicates
that there is wider room/scope of increasing the productivity of the farrmers.Therefore,
stakeholders in governmental and non-governmental organizations should (1) invest more
on farmer’s education and extension services; (2) put strong efforts towards soil
conservation practices and (3) diversification of off-farm activities to improve production,
efficiency and wellbeing of farmers. Finally, further research that considers the effect of
off-farm income on farm operations over time is needed.

40
Appendices
Appendix (1): Sample size distribution across survey sites and regions

Survey Sites (Pa) Region Observations Percent

Adado SNNPRS 125 9.19

Adele Oromia 84 6.18

Aze Deboa SNNPRS 73 5.37

Bako Tibe Oromia 72 5.29

Debre Berhan Amhara 157 11.54

Dinki Amhara 64 4.71

Doma SNNPRS 56 4.12

Gara Godo SNNPRS 84 6.18

Geblen Tigray 19 1.40

Haresa Tigray 26 1.96

Imdibir SNNPRS 60 4.41

Korode Oromia 103 7.57

Oda Dawata Oromia 43 3.16

Shumsha Amhara 106 7.79

Sirban Oromia 72 5.25

Somodo Oromia 74 5.44

Triruf Oromia 92 6.76

Yetmen Amhara 50 3.68

Total 1360 100

Source: ERHS Data & Own Computation

41
Appendix (2): MLE estimates Translog production function

Half-Normal Exponential Truncated-Normal

Variables Coeff. P>|z| Coeff. P>|z| Coeff. P>|z|

Constant 7.068 0.000 6.775 0.000 6.837 0.000

n (labour ) 0. 987 0.042 0.997 0.037 0.988 0.035

n (land ) 0.697 0.001 0.688 0.000 0.643 0.000

n ( fertilizer ) 0.417 0.370 0.456 0.002 0.445 0.001

n ( farm equpment ) 0.510 0.132 0.524 0.121 0.504 0.121

n (oxen) 0.065 0.143 0.068 0.142 0.069 0.142

(
n labour 2 ) 0.536 0.046 0.493 0.034 0.488 0.037

(
n land 2 ) -0.066 0.132 -0.015 0.129 -0.006 0.129

(
n fertilizer 2 ) 0.781 0.000 0.803 0.000 0.797 0.000

(
n farm equpment 2 ) 0.314 0.173 0.326 0.175 0.313 0.173

(
n oxen 2 ) 0.247 0.089 0.245 0.081 0.248 0.080

n (labour ) * n (land ) -0.398 0.001 -0.370 0.000 -0.376 0.000

n (labour ) * n ( fertilizer ) 0.005 0.120 0.001 0.118 0.001 0.118

n (labour ) * n ( farm equp ) 0.140 0.037 0.135 0.005 0.137 0.005

n (labour ) * n (oxen) 0.129 0.069 0.106 0.118 0.109 0.112

n (land ) * n ( fertilizer ) 0.023 0.162 0.044 0.159 0.040 0.002

n (land ) * n ( farm equp ) -0.234 0.045 -0.266 0.032 -0.254 0.004

n (land ) * n (oxen) -0.303 0.023 -0.389 0.003 -0.377 0.002

n ( fertilizer ) * n ( farm equ ) -0.040 0.074 -0.044 0.018 -0.043 0. 017

42
 n ( fertilizer ) * n (oxen) 0.050 0.036 0.046 0.000 0.047 0.091

n ( farm equ ) * n (oxen) -0.126 0.160 -0.073 0.156 -0.083 0.158

sigma_u 1.203 0.000 0.709 0.000 2.988 0.063

sigma_v 0.503 0.000 0.560 0.000 0.546 0.000

lambda 2.391 0.002 1.265 0.001 5.471 0.149

Mu - - -10.50 13.51

Log-Likelihood -1724.00 -1715.23 -1714.76

No of Observation 1360 1360 1360

Source: Own Computation

Appendix (3): Kernel density estimate of OLS residuals

Kernel density estimate

.4
.3
Density
.2.1
0

0 2 4 6 8 10
Residuals
kernel = epanechnikov, bandwidth = 0.1945

43
Appendix (4): Kernel density estimate of Half-normal, Exponential and Truncated models

Kernel density estimate

2
1.5
Density
1 .5
0

0 .2 .4 .6 .8 1
TE_half_Normal
kernel = epanechnikov, bandwidth = 0.0435

Kernel density estimate

2.5
2 1.5
Density
1
.5
0

0 .2 .4 .6 .8 1
TE_Exponential
kernel = epanechnikov, bandwidth = 0.0447

44
 Kernel density estimate
2.5
2 1.5
Density
1
.5
0

0 .2 .4 .6 .8 1
TE_Truncated_Normal
kernel = epanechnikov, bandwidth = 0.0448

Appendix (5): Distribution and Summary of technical efficiency in the half-normal, exponential
and truncated models
Half-Normal Exponential Truncated-Normal
Distributions of TE Scores
0.00-0.20 18.68 11.62 12.35
0.21-0.30 10.96 6.99 6.62
0.31-0.40 13.90 7.94 8.31
0.41-0.50 16.69 9.85 10.88
0.51-0.60 19.71 18.60 18.97
0.61-0.70 12.43 26.54 25.66
0.71-0.80 5.96 15.44 14.56
0.81-1.00 1.69 3.01 2.65
Summary of TE Scores
Mean 0. 4271 0. 5299 0. 5208
Std.Dev. 0. 2145 0. 2108 0. 2108
Minimum 0.1094 0. 1392 0. 1392
Maximum 0.8852 0. 8904 0. 8904
Observation 1360 1360 1360
Source: Own Computation

45
Appendix (6): MLE estimates Production function of farm households with Off-farm income

Half-Normal Truncated-Normal
Exponential model
model model

Variables Coeff. P>|z| Coeff. P>|z| Coeff. P>|z|

Constant 6.975 0.000 6.672 0.000 6.673 0.000

n (labour ) 0.160 0.032 0.152 0.039 0.152 0.042

n (land ) 0.805 0.001 0.978 0.000 0.978 0.000

n ( fertilizer ) 0.142 0.000 0.138 0.000 0.138 0.000

n ( farm equpment ) 0.193 0.001 0.182 0.023 0 .183 0.000

n (oxen) 0.223 0.016 0.202 0.015 0.203 0.016

sigma_u 1.187 0.000 0.705 0.000 0.880 0.263

sigma_v 0.489 0.000 0.528 0.000 0.306 0.000

lambda 2.425 0.000 1.333 0.000 2.874 0.163

Mu - - -0.461 0.187

Log-Likelihood -871.225 -858.894 -858.906

Mean TE 0.4755 0.5786 0.5783

No of Observation 698 698 698

Source: Own Computation

46
Appendix (7): MLE estimates Production function of farm households without Off-farm
income

Truncated-Normal
Half-Normal model Exponential model
Variables model

Coeff. P>|z| Coeff. P>|z| Coeff. P>|z|

Constant 7.771 0.000 7.405 0.000 7.547 0.000

n (labour ) 0.055 0.135 0.046 0.036 0.049 0.036

n (land ) 0 .832 0.000 0.920 0.000 0.889 0.000

n ( fertilizer ) 0.088 0.000 0.096 0.000 0.094 0.000

n ( farm equpment ) 0.150 0.039 0.152 0.039 0.151 0.006

n (oxen) 0.225 0.078 0.201 0.027 0.207 0.001

sigma_u 1.398 0.000 0.839 0.000 2. 509 0.024

sigma_v 0.481 0.000 0.556 0.000 0.529 0.000

lambda 2.909 0.000 1.509 0.000 4.738 0.000

Mu - - -5.210 0.205

Log-Likelihood -893.5507 -890.5749 -889.5798

Mean TE 0.4105 0.5104 0.5057

No of Observation 662 662 662

Source: Own Computation

47
References
Ahmed, B., Haji, J., & Geta, E. (2013). Analysis of farm households' technical efficiency in
production of smallholder farmers: The case of Girawa district, Ethiopia. American-
Eurasian Journal of Agricultural & Environmental Sciences, 13(12), 1615-1621.
Ahmed, M. H., Lemma, Z., & Endrias, G. (2014). Technical efficiency of maize producing
farmers in Arsi Negelle, Central rift valley of Ethiopia: Stochastic frontier approach.
Agriculture & Forestry/Poljoprivreda i Sumarstvo, 60(1).
Aigner, D., Lovell, C. A., amp, & Schmidt, P. (1977). Formulation and estimation of
stochastic frontier production function models. Journal of Econometrics, 6(1), 21-
37.
Al-hassan, S. (2012). Technical efficiency in smallholder paddy farms in Ghana: An
analysis based on different farming systems and gender. Journal of Economics and
Sustainable Development, 3(5), 91-105.
Alemu, B. A., Nuppenau, E., & Boland, H. (2004). Technical efficiency of farming systems
across agro-ecological zones in Ethiopia: An application of stochastic frontier
analysis. PhD dissertation presented to Department of Environmental Management
and Market research,Justus-Liebig University of Giessen, Germany.
Alene, A., & Hassan, R. (2003a). The determinants of farm-level technical efficiency
among adopters of improved maize production technology in Western Ethiopia.
Agrekon, 42(1), 1-14.
Alene, A. D. (2003). Improved production technology and efficiency of smallholder
farmers in Ethiopia: Extended parametric and non-parametric approaches to
production efficiency analysis. PhD Thesis, University of Pretoria, South Africa.
Alene, A. D., & Hassan, R. M. (2003b). Measuring the impact of ethiopia’s new extension
program on the productive efficiency of farmers. Paper presented at the Proceedings
of the 25th International Conference of Agricultural Economists (IAAE).
Alene, A. D., & Zeller, M. (2005). Technology adoption and farmer efficiency in multiple
crops production in Eastern Ethiopia: A comparison of parametric and non-
parametric distance functions. Agricultural Economics Review, 6(1), 5.
Amare, H. (2005). Analysis of technical efficiency in sorghum production: The case of
smallholder farmers in Raya-Azebo district, Southern Tigray, Ethiopia. An M.Sc.
Thesis presented to the School of Graduate Studies of Haramaya University,
Ethiopia.
Asefa, S. (2012). Who is technically efficient in crop production in Tigray region, Ethiopia?
Stochastic frontier approach. Global Advanced Research Journal of Agricultural
Science, 1(7), 191-200.
Babatunde, R. (2013). On-farm and off-farm works: Complement or substitute? Evidence
from rural nigeria. Paper presented at the 4th International Conference of the
African Association of Agricultural Economists, September 22-25, 2013,
Hammamet, Tunisia.
Babatunde, R., Olagunju, F., Fakayode, S., & Adejobi, A. (2010). Determinants of
participation in off-farm employment among small-holder farming households in
Kwara state, Nigeria. Production Agriculture and Technology, 6(2), 1-14.

48
Barrett, C. B., Reardon, T., & Webb, P. (2001). Nonfarm income diversification and
household livelihood strategies in rural Africa: Concepts, Dynamics, and Policy
implications. Food policy, 26(4), 315-331.
Battese, G. E., & Coelli, T. J. (1993). A stochastic frontier production function
incorporating a model for technical inefficiency effects (Vol. 69): Department of
Econometrics, University of New England Armidale.
Battese, G. E., & Coelli, T. J. (1995). A model for technical inefficiency effects in a
stochastic frontier production function for panel data. Empirical Economics, 20(2),
325-332.
Bedemo, A., Getnet, K., Kassa, B., & Chaurasia, S. (2013). Off-farm labor supply decision
of adults in rural Ethiopia: Double hurdle approach. Journal of Agricultural
Economics and Development, 2(4), 154-165.
Besanko, D., & Braeutigam, R. R. (2005). Microeconomics. Hoboken: NJ: John Wiley &
Sons.
Beyene, A. D. (2008). Determinants of off-farm participation decision of farm households
in Ethiopia. Agrekon, 47(1), 140-161.
Binam, J. N., Gockowski, J., & Nkamleu, G. B. (2008). Technical efficiency and
productivity potential of cocoa farmers in West African countries. The Developing
Economies, 46(3), 242-263.
Bojnec, S., & Ferto, I. (2011). Impact of off-farm income on farm efficiency in Slovenia.
Paper presented at the 2011 International Congress, August 30-September 2, 2011,
Zurich, Switzerland.
Bravo‐Ureta, B. E., & Pinheiro, A. E. (1997). Technical, economic, and allocative
efficiency in peasant farming: Evidence from the Dominican Republic. The
Developing Economies, 35(1), 48-67.
Chang, H. H., & Wen, F. I. (2011). Off‐farm work, technical efficiency, and rice production
risk in Taiwan. Agricultural Economics, 42(2), 269-278.
Chen, C.-F. (2007). Applying the stochastic frontier approach to measure hotel managerial
efficiency in Taiwan. Tourism Management, 28(3), 696-702.
Chirwa, E. W. (2007). Sources of technical efficiency among smallholder maize farmers in
Southern Malawi. AERC research paper 172. African Economic Research
Consortium, 2007, Nairobi, Kenya.
Coelli, T., & Rao, D., G. Battese (1998) An introduction to efficiency and productivity
analysis: Kulwar Academic Publishers, Massachusetts, USA.
Coelli, T. J. (1995). Recent developments in frontier modelling and efficiency
measurement. Australian Journal of Agricultural and Resource Economics, 39(3),
219-245.
Coelli, T. J., Rao, D. S. P., O'Donnell, C. J., & Battese, G. E. (2005). An introduction to
efficiency and productivity analysis: Springer.
Dercon, S., & Hoddinott, J. (2004). The Ethiopian rural household surveys: Introduction.
International Food Policy Research Institute, Washington, DC Photocopy.
Deressa, T. T. (2007). Measuring the economic impact of climate change on Ethiopian
agriculture: Ricardian approach. World Bank Policy Research Working
Paper(4342).

49
Farrell, M. J. (1957). The measurement of productive efficiency. Journal of the Royal
Statistical Society. Series A (General), 253-290.
Feng, S. (2008). Land rental, off-farm employment and technical efficiency of farm
households in Jiangxi province, China. NJAS-Wageningen Journal of Life Sciences,
55(4), 363-378.
Fita, L., Trivedi, M., Patel, A., Tassew, B., & Joshi, C. (2013). Determinants of technical
efficiency of the dairy farmers in Ada’a district of Oromia state, Ethiopia. Iranian
Journal of Applied Animal Science, 3(1), 59-65.
Gebre-Selassie, S. (2004). The roles of agriculture in the development process: Recent
experiences and lessons from Ethiopia. Paper presented at the 2004 Inaugural
Symposium, December 6-8, 2004, Nairobi, Kenya.
Gelaw, F. (2004). Analysis of technical efficiency of wheat production: A study in
Machakel woreda. An M.Sc. Thesis presented to the School of Graduate Studies of
Haramaya University, Ethiopia.
Geta, E., Bogale, A., Kassa, B., & Elias, E. (2013). Productivity and efficiency analysis of
smallholder maize producers in Southern Ethiopia. Journal of Human Ecology,
41(1), 67-75.
Greene, W. (2005). Fixed and random effects in stochastic frontier models. Journal of
Productivity Analysis, 23(1), 7-32.
Haji, J. (2007). Production efficiency of smallholders' vegetable-dominated mixed farming
system in Eastern Ethiopia: A non-parametric approach. Journal of African
Economies, 16(1), 1-27.
Ibekwe, U., Eze, C., Onyemauwa, C., Henri-Ukoha, A., Korie, O., & Nwaiwu, I. (2010).
Determinants of farm and off-farm income among farm households in South East
Nigeria. Academia Arena, 2(10), 58-61.
Iheke, O., Nwaru, J., & Onyenweaku, C. (2013). The impact of migrant remittances on the
technical efficiency of arable crop farm households in Southeastern Nigeria.Paper
presented at the 4th international conference of the African Association of
Agricultural Economists, September 22-25, 2013, Hammamet, Tunisia
Jara-Rojas, R., Bravo-Ureta, B. E., Moreira, V. H., & Diaz, J. (2012). Natural resource
conservation and technical efficiency from small-scale farmers in Central Chile.
Paper presented at the 2012 Conference, August 18-24, 2012, Foz do Iguacu, Brazil.
Jondrow, J., Knox Lovell, C., Materov, I. S., & Schmidt, P. (1982). On the estimation of
technical inefficiency in the stochastic frontier production function model. Journal
of Econometrics, 19(2), 233-238.
Kebede, K., & Adenew, B. (2011). Analysis of technical efficiency: Lessons and
implications for wheat producing commercial farms in Ethiopia. Journal of
Economics and Sustainable Development, 2(8), 39-47.
Khai, H. V., & Yabe, M. (2011). Technical efficiency analysis of rice production
inVietnam. Journal of ISSAAS, 17(1), 135-146.
Kilic, T., Carletto, C., Miluka, J., & Savastano, S. (2009). Rural nonfarm income and its
impact on agriculture: Evidence from Albania. Agricultural Economics, 40(2), 139-
160.

50
Kodde, D. A., & Palm, F. С. (1986). Notes and comments wald criteria for jointly testing
equality and inequality. Econometrica, 54(5), 1243-1248.
Kuiper, M., Meijerink, G., & Eaton, D. (2007). Rural livelihoods: Interplay between farm
activities, non-farm activities and the resource base Science for agriculture and rural
development in low-income countries (pp. 77-95): Springer.
Kumbhakar, S. C., Biswas, B., & Bailey, D. (1989). A study of economic efficiency of
Utah dairy farmers: A System Approach. The Review of Economics and Statistics,
71(4), 595-604.
Kumbhakar, S. C., & Lovell, C. K. (2003). Stochastic frontier analysis: Cambridge
University Press.
Liu, Z., & Zhuang, J. (2000). Determinants of technical efficiency in post-collective
Chinese agriculture: Evidence from farm-level data. Journal of Comparative
Economics, 28(3), 545-564.
López-Feldman, A., Pfeiffer, L., & Taylor, J. (2007). Is off-farm income reforming the
farm? Evidence from Mexico: Draft Paper, Department of Agricultural and
Resource Economics, University of California, Davis, USA.
Msuya, E. E., Hisano, S., & Nariu, T. (2008). "Explaining productivity variation among
smallholder maize farmers in Tanzania",proceedings of xii world congress of Rural
Sociology of International Rural Association, Goyang, Korea.
Neumann, K., Verburg, P. H., Stehfest, E., & Müller, C. (2010). The yield gap of global
grain production: A Spatial Analysis. Agricultural Systems, 103(5), 316-326.
Nkegbe, P. K. (2012). Technical efficiency in crop production and environmental resource
management practices in Northern Ghana. Environmental Economics, 3(4), 43-51.
Nyagaka, D. O., Obare, G. A., Omiti, J. M., & Nguyo, W. (2010). Technical efficiency in
resource use: Evidence from smallholder Irish potato farmers in Nyandarua North
district, Kenya. African Journal of Agricultural Research, 5(11), 1179-1186.
Obwona, M. (2006). Determinants of technical efficiency differentials amongst small and
medium scale farmers in Uganda: A case of tobacco growers (Vol. 152): African
Economic Research Consortium, Nairobi, Kenya.
Pfeiffer, L., López‐Feldman, A., & Taylor, J. E. (2009). Is off‐farm income reforming the
farm? Evidence from Mexico. Agricultural Economics, 40(2), 125-138.
Piesse, J., & Thirtle, C. (2000). A stochastic frontier approach to firm level efficiency,
technological change, and productivity during the early transition in Hungary.
Journal of Comparative Economics, 28(3), 473-501.
Rahman, M. (2007). Socio-economic determinants of off-farm activity participation in
Bangladesh. Russian Journal of Agricultural and Socio-Economic Sciences, 1, 13.
Reardon, T., Stamoulis, K., Balisacan, A., Cruz, M., Berdegué, J., & Banks, B. (1998).
Rural non-farm income in developing countries. The state of food and agriculture,
1998, 283-356.
Rizov, M., Gavrilescu, D., Gow, H., Mathijs, E., & Swinnen, J. F. (2001). Transition and
enterprise restructuring: The development of individual farming in Romania. World
Development, 29(7), 1257-1274.

51
Schmidt, P., & Knox Lovell, C. (1979). Estimating technical and allocative inefficiency
relative to stochastic production and cost frontiers. Journal of Econometrics, 9(3),
343-366.
Singh, K., Dey, M. M., Rabbani, A. G., Sudhakaran, P. O., & Thapa, G. (2009). Technical
efficiency of freshwater aquaculture and its determinants in Tripura, India.
Agricultural Economics Research Review, 22(2), 185-119.
Solis, D., Bravo‐Ureta, B. E., & Quiroga, R. E. (2007). Soil conservation and technical
efficiency among hillside farmers in Central America: A Switching regression
model. Australian Journal of Agricultural and Resource Economics, 51(4), 491-510.
Tan, S., Heerink, N., Kuyvenhoven, A., & Qu, F. (2010). Impact of land fragmentation on
rice producers’ technical efficiency in South-east China. NJAS-Wageningen Journal
of Life Sciences, 57(2), 117-123.
Tirkaso, W. T. (2013). The role of agricultural commercialization for smallholders
productivity and food security.An M.Sc. Thesis presented to Department of
Economics,Swedish University of Agricultural Sciences,Uppsala,Sweden.
Van Keilegom, I., & Wilson, P. W. (2012). Exploring research frontiers in contemporary
Statistics and Econometrics: Springer.
Waldman, D. M. (1982). A stationary point for the stochastic frontier likelihood. Journal of
Econometrics, 18(2), 275-279.
Woldehanna, T. (2000). Economic analysis and policy implications of farm and off-farm
employment: A case study in the Tigray region of Northern Ethiopia.
Landbouwuniversiteit Wageningen (Wageningen Agricultural University),
Wageningen.
Woldehanna, T. (2002). Rural farm/nonfarm income linkages in Northern Ethiopia.
Promoting farm/nonfarm linkages for rural development: Case studies from Africa
and Latin America, Rome: FAO, 121-144.
World Bank (2007). World development report 2008: Agriculture for
development.Washington,DC: World Bank.
Yue, B., & Sonoda, D. (2012). "The effect of off-farm work on farm technical efficiency in
China". Working paper, Nagoya University. Furi-cho,Chikusa-ku,Nagoya,Japan.
Zahonogo, P. (2011). Determinants of non-farm activities participation decisions of farm
households in Burkina Faso. Journal of Development and Agricultural Economics,
3(3), 174-182.

52
Declaration
I hereby affirm that the present thesis with the title “OFF-FARM INCOME AND
TECHNICAL EFFICIENCY OF SMALLHOLDER FARMERS IN ETHIOPIA: A
STOCHASTIC FRONTIER ANALYSIS” was prepared by myself alone and did not involve
the use of any impermissible help or of any other tools than the ones indicated. All parts of
the text – including tables, maps, figures, etc. – which were taken over verbatim or
analogously from other published or unpublished works have been identified accordingly.
The thesis has not yet been submitted in the same or a similar form within the context of
another examination, and has not been published either in part or in its entirety.

Uppsala, Sweden, 19th June 2014

Place, Date of submission Signature

53