Farm/non-farm linkages in smallholder agriculture - Student Positive ...

Katholieke Universiteit Leuven
Faculteit Bio-ingenieurswetenschappen
Farm/non-farm linkages in smallholder agriculture:
Evidence from Tigray, Northern Ethiopia
(De link tussen landbouwbedrijf en niet-landbouwbedrijf gerelateerde activiteiten in
kleinschalige landbouw: bewijs uit Tigray, Noord-Ethiopië)
Promotor: Prof. Erik Mathijs
Copromotor: Prof. Miet Maertens
Departement Aard- en Omgevingswetenschappen
Afdeling Landbouw- en Voedseleconomie
Masterproef voorgedragen
tot het behalen van het diploma van
Master in de bio-ingenieurswetenschappen:
land- en bosbeheer
Vandercasteelen Joachim
Juli 2011

“Dit proefschrift is een examendocument dat na de verdediging niet meer werd
gecorrigeerd voor eventueel vastgestelde fouten. In publicaties mag naar dit proefwerk
verwezen worden mits schriftelijke toelating van de promotor, vermeld op de
titelpagina.”

Acknowledgements
ACKNOWLEDGEMENTS
I would like to put a note of thanks to all the people who helped me with realizing my
master thesis. First of all, I would like to thank all the people involved in my Ethiopian
research. I would like to express my gratitude to the VLIR (Vlaamse Interuniversitaire
Raad) for awarding me a travel grant, the MU-IUC (Mekelle University – Institutional
University Cooperation) for providing me with an office at the University of Economics &
Business and the Afdeling Landbouw- en Voedseleconomie for giving me the
opportunity to do research in a developing country. Special message of thanks to
professors Erik Mathijs and Miet Maertens for guiding, supporting and assisting me
during my research and writing this thesis process.
This study wouldn‟t be possible without the help and supervision of Kidane. He
introduced me to the research area, methodology and the country. Further, I owe many
thanks to all the other supervisors, Alemtsehay, Abebe and Berhanu because they
helped me to make my stay in Mekelle as nice as possible and introduced me to the
incredible Ethiopian food. I would also like to thank Daan, who helped me with
statistical issues and treated me with a lot of coffees. Also a big thank you to the
extension agents Mebratu en Tesfai for assisting me during the field trips. Although we
have spent only a short time together, they have become real friends.
I‟m also very thankful for all the people that lived with us in Mekelle. Thanks to all the
faranji‟s that stayed in our house, Lutgart and Daan and his wife Anita. Next to this, I
want to thank all the Ethiopians I have met at university and beyond: the drivers of the
VLIR, Kidanemariam Hailu, Jemael, Aklilu and Temesgen. Thanks for your help,
kindness and laughter. Special gratitude to Tsegay, my dear friend in Mekelle. He
introduced me to Mekelle and the Ethiopian culture and he taught me so many things.
He made my stay in Mekelle unforgettable and he will always remain a special friend of
mine.
Last but not least, I would like to express my gratitude to all the people around me.
Especially my parents and sister, without them it wouldn‟t be possible for me to come
this far. Thank you for your support and the opportunities you gave me. I also like to
thank my girlfriend Hanne, who has always been there for me. Also thanks to my close
friends who made it possible for me to relax after work. Finally, I want to thank my
roommates Sebastian, Mattias, Jeroen and Jan for creating a nice working atmosphere
at home.
i

Abstract
ABSTRACT
The rural non-farm employment sector has become an important component of the
rural economy in developing countries. The (historical) dominance of the agricultural
sector raises questions how the two sectors are inter-linked in a synergistic rural
development approach. Literature has identified several mechanisms of farm/non-farm
linkages. This dissertation focuses on the complementary investment linkages between
farm and non-farm activities as suggested in the recent literature. Investment linkages
imply that non-farm income is used by farm households as an additional income source
to finance farm activities. They occur when these households face credit constraints,
which hamper their farm investments. The ultimate objective of the study is to examine
whether households that participate in non-farm activities spend more money on farm
investments and expenditures. We differentiated the impact for wage and selfemployment
non-farm income, farm input use and the different types of farm
investments. We used cross-sectional data from a household survey conducted in the
Tigray region of Ethiopia in 2009. An instrumental variable regression was used in order
to overcome the endogeneity problem caused by omitted variable bias.
The instrumental variable estimation results showed that non-farm income is significant
and positively related with farm expenditures and investments and in particular with
livestock and equipment investments. We derived similar findings when we estimated
the relationship separately for wage income; while self-employment activities did not
have a significant impact on agricultural activities. Hence, our results supported the
hypothesis that access to non-farm income has alleviated farmers‟ credit constraints.
However, access to non-farm did not increase the farm input use because of
unfavorable local conditions. Finally, the estimation of the level of non-farm income
showed that the family size (both dependent and labor forces) and low agricultural
wealth (especially of low livestock assets) are major determinants, indicating that
households with sufficient labor supply and poor resources are pushed into non-farm
activities. The results did not provide exclusive evidence that entry barriers are present
in Tigray. This might indicate the low level of skill and capital requirement to participate
in non-farm activities. Hence, a virtuous circle could possibly exist, in which access to
non-farm income increases farm expenditure and hence modernize agricultural
production. This would suggest that increase in availability and access of non-farm
activities will have a positive impact on farm activities. These findings support policies
in rural parts of Ethiopia that effectively target less endowed and marginal farm
households.
ii

Nederlandstalige Samenvatting
NEDERLANDSTALIGE SAMENVATTING
De landelijke niet-agrarische sector is uitgegroeid tot een belangrijk onderdeel van de
plattelandseconomie in ontwikkelingslanden. De dominantie van landbouw doet vragen
rijzen hoe de twee sectoren verbonden zijn. De literatuur identificeert een aantal
mechanismen van agriculturele/niet-agriculturele verbanden. Deze studie richt zich op
complementaire investeringsverbanden, die impliceren dat niet-agrarisch inkomen
gebruikt wordt als extra bron van inkomsten om agrarische activiteiten te financieren.
Ze komen voor wanneer huishoudens tegen kredietbeperkingen aanlopen, die hen
belemmeren om te investeren in landbouw. Het uiteindelijke doel van deze studie is om
te onderzoeken of huishoudens die actief zijn in niet-agrarische activiteiten meer geld
investeren in hun boerderij. We onderscheidden de effecten voor het loon en
zelfstandige niet-agrarische inkomsten, landbouwinput gebruik en de verschillende
soorten investeringen. We maakten gebruik van cross-section gegevens uit een
onderzoek onder huishoudens in 2009 uitgevoerd in de Tigray regio van Ethiopië. Een
instrumentele variabele regressie werd gebruikt om het endogeniteit probleem,
veroorzaakt door ongemerkte heterogeniteit, te overwinnen.
De resultaten van deze schatting toonden aan dat niet-agrarisch inkomen significant en
positief gerelateerd is met landbouw uitgaven, en in het bijzonder met investeringen in
vee en uitrusting. We bekwamen gelijkaardige bevindingen wanneer we deze relatie
afzonderlijk schatten voor inkomen uit loon, terwijl zelfstandige activiteiten geen
significante invloed hebben op agrarische activiteiten. Onze resultaten ondersteunden
de hypothese dat de toegang tot niet-agrarisch inkomen de kredietbeperkingen van
boeren versoepelt. Dit inkomen verhoogde het gebruik van landbouwinput echter niet
omwille van plaatselijke ongunstige omstandigheden. Uit de schatting van het niveau
van niet-agrarisch inkomen bleek ten slotte dat de gezinsgrootte en beperkt agrarische
rijkdom belangrijke determinanten zijn, waaruit volgt dat huishoudens met arbeid
overschot en weinig middelen „geduwd‟ worden in niet-agrarische activiteiten. De
resultaten boden geen exclusief bewijs dat toetredingsdrempels aanwezig waren in
Tigray. Niet agrarische activiteiten vereisen dus slechts weinig kennis of kapitaal.
Bijgevolg is het mogelijk dat een deugdzame cirkel bestaat waarin de toegang tot nietagrarisch
inkomen de investeringen (en dus ook de modernisering) in de
landbouwproductie doet stijgen. Een stijging van de beschikbaarheid van en toegang
tot niet-agrarische activiteiten zal dus een positieve invloed hebben op de landbouw
activiteiten. Deze bevindingen ondersteunen het plattelandsbeleid in Ethiopië om
minderbedeelde en marginale boerderij huishoudens effectief te bereiken.
ii

List of abbreviations and symbols
LIST OF ABBREVIATIONS AND SYMBOLS
2SLS
BoANRD
CSA
ETB
IV
REST
RNFE
TDA
masl
MU-IUC
OLS
VLIR
: Two Stage Least Squares
: Bureau of Agriculture and Natural Resource Development
: Central Statistical Agency of Ethiopia
: Ethiopian birr
: Instrumental Variables
: Relief Society of Tigray
: Rural Non-Farm Employment
: Tigray Development Agency
: meters above sea level
: Mekelle University – Institutional University Cooperation
: Ordinary Least Squares
: Vlaamse Interuniversitaire Raad
iii

List of tables
LIST OF TABLES
Table 4.1: Households‟ total farm expenditures, input use and investments in ETB .. 37
Table 4.2: Households' income composition in ETB .............................................. 40
Table 4.3: The nature of wage and self-employment jobs ..................................... 42
Table 4.4: Wage employment requirements and duration (in percentages) ............. 43
Table 4.5: Attitude towards additional non-farm employment ................................ 44
Table 4.6: Comparison of households with and without access to non-farm income .. 45
Table 4.7: Summary statistics of individual and household characteristics ............... 47
Table 4.8: Summery statistics of the instruments used in the 2SLS ........................ 48
Table 4.9: First stage OLS regression results ....................................................... 51
Table 4.10: OLS and 2SLS estimations results of total farm expenditures ................. 56
Table 4.11: OLS and 2SLS estimations results of farm investments ......................... 58
Table 4.12: IV estimation results of investments in water&land, livestock,
equipment and building ..................................................................... 60
Table 4.13: OLS and 2SLS estimations results of farm input use ............................. 61
Table 4.14: OLS and 2SLS estimations results of farm investments with wage
income ............................................................................................ 63
Table 4.15: OLS and 2SLS estimates results of farm investments with business
income ............................................................................................ 64
Table 4.16: Summary of the impact of non-farm income and control variables .......... 69
iv

List of figures
LIST OF FIGURES
Figure 3.1: Map of Ethiopia (small) and Tigray Regional state ................................. 25
Figure 3.2: Research area .................................................................................. 26
v

Table of Contents
TABLE OF CONTENTS
ACKNOWLEDGEMENTS ........................................................................................ I
ABSTRACT ......................................................................................................... II
NEDERLANDSTALIGE SAMENVATTING ............................................................... II
LIST OF ABBREVIATIONS AND SYMBOLS ......................................................... III
LIST OF TABLES ................................................................................................ IV
LIST OF FIGURES ............................................................................................... V
TABLE OF CONTENTS ......................................................................................... VI
1 INTRODUCTION .......................................................................................... 1
1.1 Background ............................................................................................ 1
1.2 Problem statement ................................................................................. 2
1.3 Research questions and hypothesis ........................................................ 4
1.4 Significance of the study ........................................................................ 4
2 LITERATURE REVIEW ................................................................................. 6
2.1 Farm/non-farm linkages ........................................................................ 6
2.1.1 Definitions ..................................................................................... 6
2.1.2 Evidence ....................................................................................... 7
2.2 The impact of farm activities on the RNFE sector ................................... 8
2.3 The impact of the RNFE sector on farm activities ................................... 9
2.3.1 Underestimation of the RNFE sector ................................................ 10
2.3.2 Importance of the RNFE sector ....................................................... 11
2.3.3 Participation in non-farm activities .................................................. 13
2.3.4 The nature of the impact of non-farm activities ................................ 15
2.3.4.1 Competing linkages ........................................................................... 16
2.3.4.2 Complementary linkages .................................................................... 17
2.4 Investment linkages ............................................................................. 19
2.4.1 Definition..................................................................................... 19
2.4.2 Liquidity and credit constraints ....................................................... 19
2.4.3 Evidence ..................................................................................... 21
vi

Table of Contents
2.5 Virtuous circle ...................................................................................... 23
3 METHODOLOGY ........................................................................................ 25
3.1 Study and survey area .......................................................................... 25
3.2 Survey design ....................................................................................... 28
3.3 Data analysis ........................................................................................ 29
3.3.1 Ordinary Least Squares estimation ................................................. 30
3.3.2 Omitted variable bias .................................................................... 31
3.3.3 Two Stage Least Squares estimation ............................................... 32
3.3.4 Instrumental Variables .................................................................. 34
4 RESULTS AND DISCUSSION ...................................................................... 37
4.1 Descriptive statistics ............................................................................ 37
4.1.1 Farm expenditure, input use and investments .................................. 37
4.1.2 Farm and non-farm activities ......................................................... 39
4.1.3 Households with non-farm income vs. households without non-farm
income ........................................................................................ 44
4.1.4 Control and instrumental variables.................................................. 46
4.2 Multivariate analysis ............................................................................ 49
4.2.1 First Stage Results ........................................................................ 50
4.2.2 Second Stage Results .................................................................... 54
4.2.2.1 The impact of non-farm income on farm expenditures ............................ 55
4.2.2.2 The impact of non-farm income on durables investments ....................... 57
4.2.2.3 The effect on different types of farm investments .................................. 59
4.2.2.4 The impact of non-farm income on farm inputs ..................................... 60
4.2.2.5 The effect of different types of non-farm income ................................... 62
4.2.3 Discussion ................................................................................... 64
4.2.3.1 The impact of non-farm income ........................................................... 64
4.2.3.2 The effect of control variables ............................................................. 68
5 CONCLUSIONS AND RECOMMENDATIONS ................................................ 71
5.1 Conclusion ............................................................................................ 71
5.2 Policy Recommendations ...................................................................... 73
REFERENCES .................................................................................................... 75
APPENDIX 1 ..................................................................................................... 81
APPENDIX 2 ..................................................................................................... 83
vii

Chapter 1: Introduction
1 INTRODUCTION
1.1 Background
Agriculture has historically been dominating the rural economy of Ethiopia. The leading
role of the agricultural sector was reflected in earlier government policies as
development programs and interventions focused mainly on the agricultural sector. As
observed in other developing countries, most efforts of the local government to develop
rural growth, corresponded with policies to enhance farm productivity (Escobal, 2001)
or increase employment in agriculture (Woldenhanna, 2000) and they thereby
neglected the importance of the rural non-farm sector (Woldenhanna, 2002; Adams,
2002). Despite the importance of agriculture in the Ethiopian economy, the country has
been facing structural food insecurity since the early 1970s (Belay and Abebaw, 2004;
Shimelis and Bogale, 2007). This makes it doubtful whether the agricultural sector is
capable to support rural development on its own and questions the leading role of
agriculture in overall economic processes. Two major factors enhanced this recognition,
making the traditional view of agriculture being the sole engine of rural growth in rural
Ethiopia to be superseded.
First, the agricultural productivity in Sub-Saharan Africa is the lowest in de world (Ehui
and Pender, 2005). Agriculture is not able to provide sufficient food for its human
population because of the stagnating, subsistence-oriented, small-scaled agricultural
sector; the low utilization of modern inputs; crucial dependence on rainfall, poor
resources endowment; poor policy environment and poor public investments
(Woldenhanna, 2002; Belay and Abebaw, 2004; Ehui and Pender, 2005; Pender et al.,
2005). Despite the importance of agriculture, Ethiopia has been facing structural food
insecurity since the early 1970s (Belay & Abebaw 2004, Shimelis & Bogale 2007).
Moreover, the OECD (2010) believes that Ethiopia will continue to face food insecurity.
Population pressure has resulted in land fragmentation, decreased farm size and
environmental degradation. It is assumed that further labor absorption in agriculture
will be difficult due to these natural and human induced problems (Woldenhanna, 2002;
Jayne et al., 2003). This will only be possible through the intensification of agricultural
production and irrigation use, which is unlikely in the short term (Woldenhanna, 2002).
The initial policy of the Ethiopian government to promote access to credit, high yielding
crop varieties and fertilizer to achieve productivity increase has not been successful in
the majority of Ethiopia (Holden et al., 2004; Ashworth, 2005).
1

Chapter 1: Introduction
Second, the rural development literature has pointed out that rural households make
up their livelihood based on complex strategies and not just on agricultural production
(Ellis, 1998; Woldenhanna, 2000; Barrett et al., 2001; Adams, 2002; Reardon et al.,
2006; Shimelis and Bogale, 2007; Kilic et al., 2009). It is recognized that non-farm
income sources are becoming increasingly important for rural households in developing
countries (Islam, 1997; Reardon et al., 1998; Escobal, 2001; Lanjouw et al., 2001).
The livelihood of rural households is the result of the interaction between complex
strategies and multiple income generating activities (Kilic et al., 2009). The livelihood
concept should therefore be adopted in development programs (Islam, 1997;
Farrington et al., 2000). Although agriculture remains the most important sector, the
study of diversification behavior offers insights necessary for broader development
strategies (Barrett et al., 2001). The recognition of the importance of the rural nonfarm
sector in the rural development process has promoted Rural Non-Farm
Employment (RNFE) as a policy and gained the support from a broad range of
development agencies. This trend occurs especially in countries like Ethiopia, which
suffer from consumption and income shocks (Block and Webb, 2001).
These considerations make it clear that policy makers are obligated to look for wider,
alternative development strategies, in order to successfully tackle poverty and food
insecurity in the East African highlands (Block and Webb, 2001; Jayne et al., 2003;
Holden et al., 2004). Employing rural people in non-farm activities has several
advantages: it offers an alternative remunerative allocation of resources, provides a
socially cost-effective expansion of the economy, diminishes land pressure and
degradation and reduces rural outmigration. Moreover, non-farm income provides
farmers additional income and is less risky and fluctuating (Islam, 1997; Lanjouw and
Lanjouw, 2001; Ruben and van den Berg, 2001; Woldenhanna, 2002). Development
policies should therefore focus on an alternative synergistic approach in which
agriculture is combined with other sources of employment and the different sectors
receive equal emphasis. The role of RNFE in overall rural development gave researchers
incentives to study the non-farm sector and its components (Islam, 1997).
1.2 Problem statement
The awareness of the importance of the RNFE sector and the need for a synergistic
development approach produces an economy where both the agricultural and nonagricultural
sector are emphasized. In such economies, the linkages between these two
sectors become important. While the impact of agriculture on RNFE has been a major
2

Chapter 1: Introduction
object of study, the study of the impact of non-farm activities on agriculture is only
recent. The impact of RNFE on farm activities is, however, not clear a priori: RNFE
could be complementary or competing with agricultural production. The RNFE literature
provides evidence of both effects of non-farm income on agriculture, making the overall
impact of RNFE not ambiguously defined. Complementarity implies that RNFE provides
non-labor variable inputs, credit or capital to farmers which can be used to increase
agricultural productivity, consumption or intensification. Competing implies that RNFE
withdrawals resources from the farm and thus decreases farm agricultural productivity.
We therefore want to examine the direction of the impact of RNFE on agricultural
activities. One particularly important type of complementary linkage are investment
linkages. These linkages imply that income gathered from RNFE is used to finance
investments in farm durables or input use. By this, non-farm income is spent on their
farm. Investment linkages are likely to occur in the presence of liquidity constraints.
Households lack credit or liquidity to invest in their farming activities, and therefore
look for other sources of credit. Households are willing to participate in non-farm
activities, as the latter result in additional cash which can be reinvested in the farm.
The objective of this study is to investigate whether positive investment effects
outweigh the loss in family farm labor availability. This is done by regressing farm
investments on non-farm income and other household and individual characteristics.
Finding a substantial and positive impact of the RNFE on the purchased farm inputs and
capital investments will have important implications. RNFE could be a potentially
important determinant of farm investment when farm households face credit market
failures. Davies et al. (2009) hypothesize the existence of a virtuous circle. Promoting
the RNFE, through access to microcredit or training, enhances the access to the RNFE
and facilities the vanquish of the associated entry restrictions. Households could
provide themselves with non-farm income and increase their total budget. Farmers are
able to modernize, commercialize or intensify agricultural production via investment
linkages. It is assumed that this will either increase the demand for agricultural wage
labor or create a surplus of capital to invest in education, migration and higher skilled
RNFE. This will in turn promote the RNFE sector. By this, a circle of self-reinforcing
mechanisms is set up. If the existence of such a virtuous circle is likely to occur,
potential opportunities for policymakers and development agents will be created.
3

Chapter 1: Introduction
1.3 Research questions and hypothesis
The possible existence of such a virtuous circle poses some interesting research
questions. It hypothesizes that households who participate in non-farm activities will
use the non-farm income to finance investments in their farm. As the farm becomes
modernized, non-farm wage labor demand will be increased. In this way, a selfreinforcing
effect will be launched. To indicate whether such virtuous circle is realistic to
occur, this dissertation will investigate the linkages between the RNFE and agricultural
sector. More specific, we investigate whether non-farm and farm activities are
complementary or not. We therefore look at the impact of non-farm income on farm
investments and expenditures.
We hypothesize that households use non-farm income as an additional capital source to
finance farm activities. We thus assume a positive relation between farm investments
and non-farm income. This favorable relation is suggested to be driven by credit
restraints. We therefore suppose that households are mainly pushed into RNFE because
they face credit constraints, and use non-farm income to overcome these restrictions.
We will also distinguish different types of non-farm income and farm expenditures. By
analyzing the impact of non-farm income on farm expenditures, not only the impact of
non-farm activities on agricultural production will be evaluated, but also the influence of
human capital, household characteristics and regional factors will be examined.
However, if such virtuous circle exists, the existence of factors that constrain the
participation in RNFE will hinder the modernization or diversification of farm
households. Several studies have investigated the existence of such „entry barriers‟ by
studying the concentration in RNFE over households (Dercon and Krishnan, 1996;
Dercon, 1998; Barrett et al., 2001; Davies et al. 2002). These studies indicate the
existence of entry barriers to participate in the more lucrative RNFE. For example,
Woldenhanna and Oskam (2001) find entry barriers in the Tigray region of Ethiopia.
This study will therefore also pay attention to the determinants of non-farm income to
examine exactly which factors hinder or enhance participation in non-farm activities.
1.4 Significance of the study
Our research focuses on Ethiopia, a poor country in the Horn of Africa. The linkages
between the agricultural and RNFE sector are believed to be very important in
developing countries. In the African context, Davies et al. (2002) conclude that forward
4

Chapter 1: Introduction
and backward production linkages are generally limited because of low agricultural
inputs. Instead, they hypothesize that agricultural growth affects the non-farm sector
mainly through expenditure linkages in the form of both consumption and investment
linkages. This finding is consistent with the results of Woldenhanna (2002) in Ethiopia
and Hazell and Hojjati (1995) in Zambia. Hence, this study will add to the empirical
strand of literature on complementary farm/non-farm linkages in developing countries.
Our results will provide evidence of the existence of investment linkages between
agricultural activities and RNFE. By this, the study will contribute to a better
understanding of the impact of non-farm income on agricultural production and the role
of the RNFE in the rural development process. If our results indicate that the RNFE is
complementary to farm activities and is used to overcome credit constraints, promoting
the RNFE will be important, if not necessary, to modernize farm activities in rural areas
of developing countries. As a result, some policy recommendations to promote the
access to non-farm activities and hence farm investments can be formulated.
Recent literature has focused on the RNFE sector and studied its role in rural
development. We are therefore interested in the importance of non-farm income (and
its different sources) in rural parts of Ethiopia. We will study the nature of the RNFE
and which factors may be important in determining the level of non-farm income.
Simple descriptive statistics will indicate the significance of non-farm activities in rural
areas and will contribute to literature focusing on the importance the non-farm sector.
Next to this, we will try to distinguish between the different types of non-farm
activities. The access, nature and requirements of the various types of non-farm
activities are different and they hence have a different effect on farm agricultural
activities. Most studies of farm/non-farm linkages do not differentiate between wage
and self-employment activities (except for, among others, Woldenhanna and Oskam,
2001; Maertens, 2009).
Finally, our study is well-founded by a robust empirical model. This is done for two
reasons. First, the deviation of the normality assumption of the error term must be
dealt with. Second, it is recognized that, for several reasons, non-farm activities are
endogenously related with agricultural production and simple regressions will be
therefore biased. An advanced empirical model will be used to solve the possible
endogenous effects of non-farm income. This methodology might serve as an example
for other empirical research that deals with the endogeneity problem.
5

Chapter 2: Literature review
2 LITERATURE REVIEW
2.1 Farm/non-farm linkages
2.1.1 Definitions
The relationship between the farm and non-farm sector is mostly described by using
the concept of farm/non-farm linkages (Reardon et al., 1998). Linkages are financial
transactions between the two sectors over time. A distinction between production and
expenditure linkages is made. Production linkages can be further divided into upwards
and downwards production linkages. Upward production linkages are found in the nonfarm
sector when agricultural output is used as input. Raw agricultural outputs are
processed and distributed by non-farm enterprises (Woldenhanna, 2000). Growth in
farming stimulates agricultural productivity and hence the capacity to supply inputs and
services to the non-farm sector. Downward production linkages refer to non-farm
activities that provide inputs to agricultural production, such as agrochemicals, water
pumps or fertilizers. The non-farm sector is encouraged to invest in supply capacity of
agroprocessing and distribution services (Reardon et al., 1998; Davies et al., 2002).
The characteristics of the local agriculture will determine whether the production
linkages will be upward or downward. For example, farm size determines the
profitability of markets for tractors (Reardon et al., 1998).
Expenditure linkages occur when households finance spendings in one sector by the
money earned from another sector. Farmers can for example purchase non-farm
products with income generated from farm activities. On the contrary, people that have
access to RNFE buy food and other agricultural output with the income derived from
that non-farm activity. When these expenditures are related to household consumption,
consumption linkages establish. Farm income increases the demand of basic goods and
services and results in diversification of consumption (Woldenhanna, 2000). Investment
linkages include expenditures used to finance non-farm or farm activities, which are
mainly important within households. Returns from non-farm activities can be used to
make investments in farm activities and thereby enhance agricultural productivity
(Davies et al., 2002). The profitability of these expenditure linkages depends on the
level and distribution of the income. Poor households will spend more on local goods
and services in the RNFE, while richer households are more likely to invest in goods
from the modern and urban manufacturing sectors or in imports.
6

Chapter 2: Literature review
The structure of the agricultural sector and the type of growth determines the type of
linkage that will occur. Davies et al. (2002) illustrates this with several examples. If
significant external inputs are needed for agricultural production, it is expected that
backward production linkages will occur. Agricultural output that requires processing
before selling induces forward production linkages. If growth in the agricultural sector is
capable of inducing rural income growth, consumption and potential investments will be
enhanced by expenditure linkages.
2.1.2 Evidence
There exists numerous literature that reviews the inter-sectoral linkages in rural areas.
Refer to Islam (1997) and Lanjouw and Lanjouw (2001), who reviewes the earlier work
done on inter-sectoral linkages. Cross-sectional linkages seem to be crucial for the
initial development of the RNFE. Studies proof that non-farm activities positively
influence the farm activities and vice versa mainly through production and consumption
linkages. Islam (1997) states that growth in farm income stimulates the consumption
of goods and inputs, while agricultural raw material is processed in the rural non-farm
sector. The growth pattern in agricultural income and the technology used in production
determines the strength of the production and consumption linkages. Andreosso-
O‟Callaghan and Yue (2004) did a comparative analysis of the traditional and modern
methods in linkage analysis in China between 1987 and 1997. They conclude that
forward and backward linkages have generally increased, indicating an increased intersectoral
dependency. Blunch and Verner (2006) observe two-way spillover effects
between industrial and agricultural growth in three African countries, indicating
interdependence in sectoral growth.
Davies et al. (2002) discuss several case studies of farm/non-farm linkages in Africa
and Latin America where spin-off activities already exist. In low–income countries such
as Ghana, Kenya or Ethiopia, limited local demand and limited investments oblige for
spin-off activities. In Kenya, contract farming acts as the primary means of interaction
between farmers and the agro-industry. This creates an income surplus which can be
spent in local markets. In Ghana, substantial forward linkages are noticed in the lowinput
cassava sector because processing is necessary and transportation is required.
Other research in Africa has been conducted by Hazell and Hojjati (1995), who
examined different types of inter-sectoral linkages in Zambia. First, they observe that
credit linkages might occur since agricultural income could be used to finance and
invest in easy-entry non-farm activities. Second, it is concluded that seasonal labor
7

Chapter 2: Literature review
shortage affects both farm and non-farm activities. The authors suggest the existence
of a shift of labor depending on the seasonal requirement of the sectors. This effect is
uncertain since the opposite is also observed. Third, it is noted that production linkages
are most pronounced in the agricultural growth of small farms. Long-term investments
in the farm are assumed to create demand linkages to both the farm and non-farm
sector. Finally, consumption linkages are considered to be the most dominant linkages
in Zambia.
In Ethiopia, research on farm/non-farm linkages in the Tigray region has been done by
Woldenhanna (2002). First, it is observed that production linkages, both backward and
forward, are limited because the households‟ purchase of fertilizer and pesticides was
very low. The agricultural sector is unable to support the processing industry due to the
fact that households consume most of their own farm products. Farm households only
sell 15% or less of their production. These limited production linkages imply that the
correlation between RNFE and farm activities is rather weak. Second, the significant use
of wholesale and retail trades by farmers indicates that consumption linkages are more
important than production linkages. These consumption linkages are the strongest
expressed in locally produced food. The author also concludes a positive relation
between the demand for consumption goods and agricultural income. Third, it is
suggested that agricultural production has the potential to enhance the demand for
non-food goods. The elasticities of local non-food expenditures are high, implying that
when households‟ income increases, their importance in the household budget also
rises. However, this effect is only valid in the short term, as the elasticities of imported
non-food expenditures are even higher. Finally, non-farm activities absorb labor that
cannot be allocated on agricultural activities. This is suggested by the negative relation
between agricultural output and small enterprises.
2.2 The impact of farm activities on the RNFE sector
The previous section shows that the rural development literature attributed attention to
inter-sectoral linkages. Furthermore, the existence of farm/non-farm linkages in rural
areas is suggested. Given the historical dominant role of agriculture and the fact that
the majority of the rural population is engaged in it, the rural development literature
initially emphasized the fostering role of agriculture in the overall economy. It was
believed that the rural sector in developing countries was entirely driven by agriculture
and that agricultural growth would stimulate overall economic growth by promoting
other sectors via production and consumption linkages (Islam, 1997; Reardon et al.,
8

Chapter 2: Literature review
1998; Woldenhanna, 2000; Escobal, 2001; Adams, 2002). Rural development policies
and programs in developing countries prioritized the income derived from agricultural
production, while the activities related to the non-farm sector received little attention to
date (Reardon et al., 1998; Woldenhanna, 2000; Escobal, 2001). As a consequence, it
was believed that agricultural growth would be the main stimulus for overall economic
growth by promoting other sectors (Islam, 1996; Lanjouw et al., 2001; Ashworth,
2005). Hill (1982), as cited by Reardon (1997), mentions that until the early 1980s the
widespread view persisted that rural African farmers mainly undertook farm activities
and less non-farm activities, except when they migrated out of the rural areas.
According to Mills, this resulted in an initial neglect and underestimation of the rural
non-farm sector.
In developing countries, the impact of agriculture on RNFE occurs mainly through both
production and consumption linkages (Haggblade et al., 1989; Islam, 1997; Lanjouw
and Lanjouw, 2001; Blunch and Verner, 2006). Haggblade et al. (1989) provide
evidence suggesting that consumption and production linkages explain the positive
effect of agricultural growth on RNFE and non-farm income. They indicate that higher
agricultural income increases households‟ demand multiplier effect across the rest of
the rural economy, especially in closed economies. Many developing countries face
closed economies due to high transaction costs, but this result was more pronounced in
Asia compared to Africa. Ellis (1998) reviews earlier work done on rural growth linkages
in developing countries and finds that agricultural growth provided the stimulus for the
growth in non-farm activities mainly due to consumption linkages. Reardon et al.
(2001) discuss the growth engines for the development of RNFE and notice the positive
relation between the rise in farm output (and income) and demand for non-farm
products through consumption linkages. Ashworth (2005) states that the initial
development programs in Ethiopia expected agricultural growth to simulate non-farm
sector development and RNFE creation as well as overall economic growth. The number
of people directly dependent on agriculture for their income and subsistence reduces as
employment in the non-farm sector is created.
2.3 The impact of the RNFE sector on farm activities
The impact of non-farm activities on agricultural production has not been a major topic
in literature of developing economies. However, the RNFE literature has increasingly
paid attention to the impact of non-farm activities on agricultural production during the
last decades. Moreover, this switch in focus of literature is only recent, both because of
9

Chapter 2: Literature review
the mastery of the agricultural sector and the historical disregarding and
underestimation of the RNFE sector. Drivers behind the change in focus are the
increasing acknowledgement of the significant role and importance of the non-farm
sector in rural areas and the stagnation of the agricultural sector (Islam, 1997;
Woldenhanna, 2002). Therefore, the rural non-farm sector has recently received
increasing attention from both analysts and policymakers.
2.3.1 Underestimation of the RNFE sector
Previously, the common view persisted that rural growth in developing countries is
entirely driven by agriculture and that non-agricultural activities are not significant.
Davies et al. (2009) and Pfeifer et al. (2009) point out that previous literature assumed
that farm households mainly rely on farm activities to attain their food security and that
the contribution of the RNFE in rural development was mostly ignored. Only if the farm
activities were not able to fulfill the household‟s needs, households‟ members would
work on other farmers‟ land for wage or would be sent out as migrants for remittances.
The bias against the rural non-farm sector has been further enhanced by the fact that
the sector was initially a poorly understood component of the rural economy of
developing countries and little was known about its role in the broader development
process. Further, it was wrongly assumed that the non-farm sector was characterized
by low productivity and production of low quality goods; and that it will vanish as a
country develops (Lanjouw and Lanjouw, 2001).
As a consequence, most government policies were did not support the non-farm sector,
nor did the governments devote resources to upgrade the non-farm sector (Islam,
1997; Lanjouw and Lanjouw, 2001). In Ethiopia, Woldenhanna (2000) confirms that
the non-farm sector has been ignored because of the power given to the agricultural
sector. He states that “while substantial resources have been spent on agricultural
research and extension to alleviate food shortage in the nation, no research and
extension have been done on the issue of non-farm employment versus farm
employment. Despite this fact, farmers are engaged in a variety of non-farm activities
to diversify their income and to feed themselves during crop failures” (Woldenhanna
2000, p. 94).
As a result, the impact of the RNFE on agricultural production and activities has been
mostly ignored. Pfeiffer et al. (2009, p. 139) state that “as rural households become
increasingly involved in non-farm activities, their on-farm production changes. The
10

Chapter 2: Literature review
ways in which non-farm activities transform agricultural production have not been a
major subject of inquiry in the development literature”. Also Davies et al. (2009, p.
120) state “that the RNFE impact on agriculture, especially on micro level, has received
relative little attention, although the importance of the RNFE has been proven and
attention has been given to agriculture as a determinant of RNFE”. Stampini and Davis
(2009, p. 177) review the previous literature and find that “the existing empirical
literature on household-level linkages between rural non-farm activities and farming is
limited and inconclusive”.
2.3.2 Importance of the RNFE sector
The literature and empirical research of the significance of the RNFE sector has
increased, since the late 1980s (Davies et al., 2009). This literature provides evidence
of the importance of the RNFE sector and thereby removes the bias against it. In this
way, the significance of non-farm sector has been revealed and, to some extent, the
attitude towards the RNFE has changed. A review of this literature can be found in Ellis
(1998), Escobal (2001) and Lanjouw and Lanjouw (2001). In developed countries like
the U.S., Fernandez-Cornejo et al. (2007) state that non-farm work has risen steadily
and has become the most important component of farm household income, far more
important than farm income. However, the impact of non-farm income is not that
straightforward in developing countries (Reardon, 1997). It is assumed that farm
activities remain important in rural households as they provide the main source of
employment and income in rural areas of developing countries (van den Berg and
Kumbi, 2006). For example, Reardon (1997) finds that there was little known about
why or how much farm households engage in RNFE, especially in Africa. Carswell
(2002) concludes that it is not the importance of the RNFE that is increasing, but the
interest and recognition given to it.
Nevertheless, it is a widely observed phenomenon that the rural non-farm sector in
African developing countries is growing rapidly (Islam, 1997; Lanjouw et al., 2001;
Reardon et al., 2001; Escobal, 2001). Reardon et al. (1994) are among the firsts to
provide evidence from Burkina Faso that show that income from RNFE can be an
important source of household cash. Islam (1997) reviews earlier studies and published
data on the RNFE in developing countries and states that globally 23% to 50% of total
households‟ income is made up by non-farm earnings. In Latin America, Asia and Africa
respectively 26%, 28% and 14% of the rural labor force is employed in the non-farm
sector. Reardon (1997) reviews earlier studies on income diversification, small
11

Chapter 2: Literature review
enterprises and inter-sectoral linkages in sub-Saharan Africa. He notices that local
wage labor earnings in the non-farm sector were dominant in the majority of the case
studies. The average share of non-farm income in total household income for the
different case studies was 45%, ranging from 22% (Burkina Faso) to 93% (Namibia).
Reardon et al. (1998) study the nature, importance and determinants of the RNFE in
developing countries based on a review of about 100 household surveys. They show
that the non-farm sector accounts for 25% of employment and that it generates 40%
of household income in rural Latin America. The income shares for the RNFE accounts
for 32% in Asia and 42% in Africa. Escobal (2001) concludes that in rural Peru almost
35% of the labor is allocated to non-farm activities and 51% of the income comes from
the RNFE. He therefore suggests that the RNFE no longer has a marginal contribution.
Lanjouw and Lanjouw (2001) review conceptual and empirical literature concerning the
RNFE and focus on the experience in developing countries. They made a review of
aggregate data of the RNFE based on than 60 surveys within the RNFE literature on
developing countries. They conclude that the RNFE sector is substantial in many
countries and has been growing over time. They stress that the percentage of RNFE
may be underestimated because of temporary non-farm jobs during slack periods in
agriculture or unremunerated work (especially done by women).
De Janvry et al. (2005) use a detailed household survey dataset from Hubei province in
China to analyze the impact of economic reforms. They state that farming remains the
main income source for rural households. However, non-farm income is becoming more
important. The study shows that 72% of the rural households have a non-farm income
that accounts on average for 36% of total household income. The average income of
households that only participate in farm activities is lower than the average income of
households that have access to non-farm activities. Next to this, the authors find that
non-farm income decreased income inequality and reduced rural poverty. Ruben and
van den Berg (2001) use the national income expenditure survey from 1993 to 1994 in
Honduras to study the role of non-farm income of rural farm households. They estimate
that 68% of the rural adults are somewhat involved in non-farm activities.
Furthermore, the income from non-farm wage en self-employment accounts for 16-
25% of the household‟s income.
Kilic et al. (2009) review empirical evidence on the RNFE in a number of developing
countries. They came across literature stating that RNFE income accounts for 50% and
35% of the total income in respectively Latin America and Africa. Comparing this with
other literature, RNFE income accounts for 58% of the total income on a global scale.
12

Chapter 2: Literature review
Refer to Jayne et al. (2003), Otsuka and Yamano (2006) and Reardon et al. (2006) for
more recent data on the increased importance of non-farm income and patterns of
RNFE in case studies from Asia, Latin America and Africa. Carswell (2002) finds
evidence that non-farm and off-farm activities are carried out by a significant
proportion of adults in Wolayat, southern Ethiopia. Trading, a form of self-employment,
seems to be the most important activity, carried out by 14% of the adults. 84% of the
participants in non-farm activities does this at least once a week, while for more than
50% the job is active throughout the year.
2.3.3 Participation in non-farm activities
The actual participation of households in non-farm activities depends on the incentive
and capacity to participate (Reardon, 1997) and the occurrence of entry barriers
(Dercon and Krishnan, 1996). Two opposite forces will determine the households‟ motif
to diversify its income sources. Push factors (or necessity) are the involuntary and
sometimes desperate reasons to diversify; they include income risk management,
coping mechanisms, diminishing or time-varying returns to productive assets, longterm
constraints or smoothing household consumption (Ellis, 1998; Ellis, 2000b;
Barrett et al., 2001; Reardon et al., 2001). Voluntary diversification is opted for
accumulation objectives (Reardon et al., 2006) or with the goal to maximize profits
(Kilic et al., 2009). Reardon et al. (2000) suggest that poor households will be attracted
to low-risk RNFE in order to decrease income variability, even though they might have
low returns (Reardon et al., 2006). Wealthier households will be less diversified in their
income sources because risk aversion motivation declines as wealth increases under
perfect market conditions. Pull factors will attract households to the non-farm sector
when the non-farm activities offer higher returns compared to farming. For the poor,
their relative return is higher because they have a lower reservation wage
(Woldenhanna and Oskam, 2001; Reardon, 1997). However, rich households can be
pulled into the RNFE to maximize their profits (Kilic et al., 2009).
According to Davies et al. (2009) the ability of households to act on these incentives
depends on a set of capacity variables. On the micro level, these variables include the
vector of assets such as physical, social, human, and organizational capital, and
liquidity from sources such as cash cropping. On the meso level, these variables contain
access to local assets such as hard infrastructure (roads) and soft infrastructure
(financial services). The access to credit and financial markets proofs to play a crucial
13

Chapter 2: Literature review
role (Dercon and Krisnan, 1996; Barrett et al., 2001b; Reardon et al., 2006; Davies et
al., 2009).
However, high return non-farm activities have certain requisites to enter these
activities. These requirements include among others education, skills and investments.
Activities with no or low entry barriers generally offer low returns, while activities with
significant entry restrictions have higher returns. Participation in these lucrative nonfarm
activities is thus conditioned by the possibility to overcome the required entry
barriers. When entering a non-farm activity requires substantial investments, liquidity
constraints will hamper households with restricted assets to enter these activities
(Dercon and Krishnan, 1996; Woldenhanna, 2000; Barrett et al., 2001; van den Berg
and Kumbi, 2006). The ability of households to overcome these entry barriers depends
on their capacity variables (Davies et al., 2009). Collateral requirements, market
imperfections and differences in repayment capacity make credit constraints more
severe for poor households than for rich (Woldenhanna, 2000). As a consequence, poor
households will be unable to overcome these entry barriers while wealthier households
will have less problems in diversifying their income.
Not everything is decided based on motivation only. Participation is also influenced by
capacity and entry barriers. As a result, the opposite outcome of what would seem
logical could be observed. Poorer households might be motivated to diversify their
income for risk aversion, but entry barriers prevent this (Dercon and Krishnan, 1996;
Dercon, 1998; Barrett et al., 2001; Escobal, 2001; Davies et al., 2002) resulting in
participation in non-farm activities with lower returns and higher risks (Sumberg et al.,
2004) and thus less diversified income sources. On the contrary, rich households may
have less incentive to diversify, but they find themselves in a superior position to make
investments to overcome entry barriers (Reardon et al., 2000). Even though the
marginal value of non-farm activities exceeds the reservation wage, entry barriers will
prevent poor households to enter. They are forced to participate in low-return non-farm
activities for which the entry barriers are low, most probably wage employment.
Wealthier households will face less binding credit constraints and will engage in highreturn
activities with higher entry barriers, mostly self-employment (Woldenhanna and
Oskam, 2001). Rich households have a great freedom to choose among a wider range
of non-farm activities, while poor households are relying on unskilled labor and
activities with low barriers and therefore low returns (Sumberg et al., 2004).
14

Chapter 2: Literature review
2.3.4 The nature of the impact of non-farm activities
As it is suggested that the RNFE has a substantial impact on agriculture, the question
emerges how participation in non-farm activities influences the households‟ farm
decisions. Ellis (1998) finds that possible adverse (competition for labor and credit) and
beneficial (reinvestments and insurance) impacts of non-farm income on household
level are suggested in the literature. The net impact of non-farm activities is however
highly specific in time and space. Reardon et al. (1994) did research in Burkina Faso
and suggest that non-farm activities influence farm activities indirectly through capital
investment and input acquisition. They hypothesize that non-farm activities can either
draw resources away from agricultural production or stimulate reinvestments in farm
activity. Whether non-farm activities are complementary or competing to farm activities
depends on both physical, economic and institutional factors and determinants of
households‟ allocation choice of resources over farm and non-farm activities.
On the one hand, Woldenhanna (2000) states that productivity can be decreased due to
lack of specialization, management inefficiency and competition for inputs. On the other
hand, productivity can be increased because non-farm activities increase manager
skills, reduce land pressure and provide credit for farm investments in case of credit or
capital constraints. It is concluded that income diversification can have both a positive
and a negative impact on farm income and its net impact cannot be determined a
priori. Pfeiffer et al. (2009) suggest that as market imperfections occur, agricultural
income can be affected by non-farm income in different ways. A first direct impact is
through the loss of labor, because agricultural inputs must be sacrificed when
household members are participating in non-farm activities. Under perfect labor market
conditions, households are able to hire perfect substitutes for missing labor on the farm
and agricultural production can be maintained. However, developing countries face
labor market constraints and non-farm activities will reduce the labor input in
agricultural activities. The non-farm income affects agricultural production in a direct
and indirect way when credit and liquidity markets imperfections occur. The direct
effect of non-farm income is the relaxation of household budget constraints and the
increase of the purchase of normal goods. The indirect effects of non-farm are more
complex: non-farm activities provide households capital, security and liquidity to invest
in technology or farm inputs.
15

Chapter 2: Literature review
2.3.4.1 Competing linkages
Many authors recognize that the impact of the RNFE on the agricultural sector can be
either complementary or competing. Competing linkages occur if households‟ decisions
about non-farm and farm activities are made jointly and households face limited
resources and inputs such as capital and labor (Reardon et al., 1994). Participation in
non-farm activities requires reallocation of those limited resources and results in an
inevitable withdrawal from the farm. For example, RNFE consumes agricultural labor so
that the labor availability on the farm decreases. Another consequence of the
competing nature is that the factor bias of farm technology can be affected (Reardon et
al., 1998). If non-farm activities have a higher return and agricultural investments are
risky, investments in land conservation and technology could be impeded (Reardon et
al., 2000). According to Reardon et al. (2001), the high productivity and higher returns
of non-farm activities lead to concentration of resources in the RNFE. As a
consequence, participation by households in non-farm activities can hamper their own
farm productivity (Ellis and Freeman, 2004; Phimister and Roberts, 2006). Agricultural
production and farm income will decrease and thereby hampers agricultural
commercialization or modernization (Ruben and van den Berg, 2001). The amount of
non-farm activities performed and the importance of each activity depends on the
relative returns to farm versus non-farm activities and their input requirements.
There is empirical evidence of competing linkages between farm and non-farm
activities. Goodwin and Mishra (2004) studiy the relationship between farm efficiency
and non-farm labor supply in the U.S. They report that high involvement in non-farm
activities decreases farm efficiency. They suggest that this inverse relationship can be
explained by the hypothesis of Smith (2002, as cited in Goodwin and Mishra, 2004)
“Does off-farm work hinder smart farming”. Smith suggests that non-farm activities
have strong implications for the efficiency of farming, because less attention can be
devoted to issues important for farm productivity such as adoption of best management
practices. As Goodwin and Mishra (2004) find evidence that non-farm labor allocation
and farming efficiency are jointly determined, their results show that farmers who are
more efficient on the farm (this is a higher implicit farm wage) tend to allocate less
labor to non-farm activities. This is in line with the results of Chang and Wen (2010) in
Taiwan. They suggest that farmers without RNFE are likely to have better knowledge of
and pay more attention to farm management. Therefore, their use of inputs is more
productive than farmers with access to RNFE. Huang et al. (2009) study the linkages
between non-farm labor markets and the on-farm labor allocation to production of fruit
crops in the Shandong Province of China. Among other things, the authors conclude
16

Chapter 2: Literature review
that the intensity of fruit production is reduced with increased RNFE. The fruit
production sector is associated with lower entry barriers than the RNFE, attracting
individuals that are unable to engage in the RNFE because of high age or low education.
Kilic et al. (2009) use data from the 2005 Albania Living Standards Measurement
Survey to explore the overall impact of the RNFE on agricultural expenditures and
technical efficiency of rural households. They find that Albanian households do not use
non-farm earnings to invest in time-saving, efficiency increasing agricultural
technologies. Less expenditures are made in productivity enhancing crop input
investments. The authors suggest that the existence of sectoral problems is driving
income diversification out of crop production but towards livestock production. In the
Ethiopian context, Holden et al. (2004) find that non-farm activities negatively
influence farm production in two ways. First, better access to the RNFE reduces total
agricultural production and farm inputs expenditure. Second, rural non-farm income
reduces households‟ incentive to invest in conservation measures, increasing land
degradation and soil erosion. Not only agricultural intensity is decreased, a drop in farm
productivity is suggested. This has important consequences for rural development, as
the need to import food in the area increases.
2.3.4.2 Complementary linkages
Complementary linkages indicate that non-farm income acts as an important source of
capital and cash in the households‟ total budget. This extra budget could be used by the
household to invest, hire labor, purchase inputs or finance consumption. Reardon et al.
(1994) suggest that consumption and investments compete for the use of household
income and therefore household decisions affect the nature of the consumption and
investment linkages. It is not clear how the households allocate their resources over
consumption and investments. According to Woldenhanna (2000), Ruben and van den
Berg (2001) and Pfeifer et al. (2009), non-farm income is important to satisfy
consumption requirements when agricultural production cannot provide food security. It
can, however, also be used to finance farm activities. Non-farm income might have a
positive impact on agriculture as it can be used to buy food which frees up other
resources that can be invested in farm activities (Kilic et al., 2009; Woldenhanna,
2000) or to buy farm inputs (Ruben and van den Berg, 2001). However, non-farm
income might also be used outside the farm because of complex agricultural sectoral
problems that cannot easily be surmounted. In such case, finance investments in the
RNFE, education and training or to migrate out of the rural sector (Kilic et al., 2009).
17

Chapter 2: Literature review
Complementary linkages have been suggested earlier. Reardon et al. (1994) show that
RNFE can provide an important source of cash for households in Burkina Faso. Reardon
et al. (1998) find that income from agroindustrial activities influences farm households‟
capacity to invest in farm input, capital and appropriate technology. It is suggested that
non-farm activities are potentially important for long-term food security because it
stimulates spending on farm inputs and thereby farm productivity. The authors provide
evidence from Burkina Faso, the Niger and Senegal, indicating that non-farm income is
usually the main source of cash for the purchase farm inputs. The authors suggest a
dynamic effect as non-farm income surplus is invested in farm inputs, creating capital
that substitutes for labor and thus reduces farm labor demand. Lanjouw and Lanjouw
(2001) suggest that non-farm income leads to higher average income from agriculture
in two different ways. First, technologies or crop varieties with higher productivity are
often related to higher variability. Access to an alternative stable income source will
facilitate the adoption of these technologies or varieties. Second, if access to low cost
credit is absent, additional income sources can stimulate the investments in farm
inputs. Davies et al. (2002) suggest that the growth linkages that arise from a first
round of agricultural boom could be reinvested in capitalizing agriculture. Moreover, the
RNFE sector can provide opportunities for households to reduce agricultural risks and
stabilize their income (Phimister and Roberts, 2006).
Chang and Wen (2010) study the impact of non-farm wages on agricultural efficiency
and production risks in Taiwan. They use a nationwide survey of rice farmers to
investigate the efficiency and yield difference between households that participate in
RNFE and households that do not. It is noticed that the resource use for both
households is different and that RNFE does not necessarily induce technical inefficiency.
In the lower percentiles of the efficiency distribution, farmers with RNFE are more
efficient than farmers without. Lien et al. (2010) analyze the determinants of the nonfarm
work decision and its influence on farm performance, based on panel data from
Norwegian grain farms from 1991 to 2005. They did not observe a negative impact of
non-farm work on farm efficiency or production. They show evidence that households
that participate in RNFE could increase production to some extent. This positive effect
first increases but then decreases with increasing hours spend on non-farm work.
18

Chapter 2: Literature review
2.4 Investment linkages
2.4.1 Definition
While the previous section suggests competing and complementary linkages possible to
occur, we now turn our focus to investment linkages. In the context of investment
linkages, farm households selectively engage in the rural non-farm market to earn an
additional income. This supplementary income is a source of liquidity and credit for the
household. Moreover, Reardon et al. (1994, p. 1175) state that “non-farm income can
also serve as collateral and thus facilitate access to credit”. The additional source of
credit and liquidity can be used by the household to finance risky investments in
agriculture (Barrett et al., 2001; Kilic et al., 2009; Pfeifer et al., 2009). Households use
their non-farm income sources to finance farm investments, to self insure or purchase
cash inputs for agricultural production (Reardon et al., 1994; Ellis, 1998; Barrett et al.,
2001; Davies et al., 2002; Reardon et al., 2006). Oseni and Winters (2009) and Pfeifer
et al. (2009) expect the surplus of cash generated by non-farm income to directly
influence the purchase of agricultural inputs.
Under the condition that the additional source of credit is invested in durable
investments and farm input purchase, households are in the position to increase their
farm expenditures and expand investments (Reardon et al., 1994; Oseni and Winters,
2009). The non-farm income thereby improves farm productivity and boosts the
agricultural production and farm income in the long or short term (Pfeiffer et al., 2009).
The former refers to the purchase of inputs like fertilizer, while the latter refers to the
adoption of productivity increasing technologies like improved seeds. The ultimate
result is that RNFE could be an important driver to foster farm commercialization,
modernization or agricultural diversification into higher value activities or agricultural
production intensification (Davies et al., 2009; Oseni and Winters, 2009).
2.4.2 Liquidity and credit constraints
Investment linkages are likely to occur when households are pushed into non-farm
activities because they face credit and liquidity constraints or access to other sources of
cash is not available (Oseni and Winters, 2009; Pfeifer et al., 2009). These credit
constraints hamper households‟ investment in production and productivity increasing
technologies and inputs. As a result, credit restrictions drive households to self-insure
or finance input expenditure with their own credit (Reardon et al., 2006; Oseni and
Winters, 2009). Participation in non-farm activities could be an important way to
19

Chapter 2: Literature review
generates cash and substitute for the absence of credit or the high cost linked with
borrowing credit. It implies that households diversify their livelihood into non-farm
activities to overcome some of the credit and insurance constraints.
How reasonable is it to expect that credit constraints determine the linkage between
farm and non-farm activities? Households‟ engagement in RNFE is conditioned by their
motives (push or pull), capacities and the existence of entry barriers, as explained in
section 2.2.3. While it is viable that a combination of motives drive participation in the
RNFE, some authors suggest that the most important reason for households in rural
areas of developing countries to participate in non-farm activities is the relaxation of
constraints (Oseni and Winters, 2009; Pfeifer et al., 2009; Stampini and Davis, 2009).
In Tigray, Woldenhanna (2000) suggests that if a binding liquidity constraint is present,
these constraints will induce farm households to participate in non-farm activities to
help them finance farm input and hire farm labor.
The constrained accessibility of credit is driven by market imperfections (Oseni and
Winters, 2009; Pfeifer et al., 2009; Stampini and Davis, 2009). Credit or insurance
market failures imply low availability and access to credit which hampers households‟
investment in production and productivity. Reardon et al. (2004) already suggested
that the RNFE is likely to have a positive impact on farm activities in cases where the
rural markets do not function properly. This is in line with the arguments given by Kilic
et al. (2009), who suggest that “given the imperfect covariance between farm and nonfarm
income, non-farm earnings may help households overcome credit and insurance
market constraints by providing liquidity that can be utilized for productivity enhancing
input purchase and long-term investments in agriculture (Kilic et al., 2009, p. 140)”.
Oseni and Winters (2009) and Pfeifer et al. (2009) state that market imperfections are
likely to occur. Moreover, several studies suggest the existence of imperfect markets in
developing countries (Ellis, 2000b; Barrett et al., 2001). De Janvry and Sadoulet
(2003) review literature dealing with market imperfections and study households‟
behavior under market imperfections. They conclude that market imperfections are
present. Pfeifer et al. (2009) find evidence that labor and credit markets are imperfect
in rural Mexico. They suggest that either liquidity constraints or the presence of
transaction costs in markets impede access for small farmers to hired labor. Oseni and
Winters (2009) state that in rural areas of developing countries, credit and insurance
markets do not function properly or even non-existent in some cases. The lack of
access to credit hinders households‟ investing possibilities and consumption smoothing.
In conclusion, market imperfections exist, and merely occur in credit markets.
20

Chapter 2: Literature review
Woldenhanna (2000) assumes that farm households in Tigray are not fully integrated
into the market. The survey respondents stated that their demand for credit is not fully
satisfied. Moreover, private supply of credit and consumption credit is almost absent.
The available credit is supplied by public organizations and linked with participation in
extension activities. Pender et al. (2006) state that credit is practically absent for
smallholders in the East African highlands. However, there is limited credit available
through cooperatives, private firms and government support programs, but there are
almost no opportunities to borrow from formal financial institutions such as commercial,
insurance or construction banks. Moreover, these organizations require collateral and
involve time consuming screening processes (Woldenhanna, 2000). Households are
therefore left with uncertain loans provided by credit schemes or small traders.
2.4.3 Evidence
In response to the awareness that the impact of the RNFE on agricultural production
could be mainly through investment linkages, the evidence proving the existence of
investment linkages is growing. Phimister and Roberts (2006) investigate the extent to
which the RNFE changes the intensity of agricultural input for 2,419 farmers in England.
Their results suggest that fertility intensity declines while conservation inputs increase
with RNFE. Ellis and Freeman (2004) make a comparison of the rural livelihoods in
Uganda, Kenya, Tanzania and Malawi. One of their findings is the positive relation
between non-farm employment and agricultural production, which can be explained by
the availability of cash to invest in farm inputs or conservation practices and to hire
labor. De Janvry et al. (2005) find that participation in RNFE has a positive spillover
effect on household agricultural production. They use a household survey dataset from
the Hubei province in China and study the influence of the RNFE on households‟
income, poverty and inequality. Deficient rural credit markets force households to
engage in RNFE, enhancing on-farm investment capacity, mitigate income fluctuations
and function as insurance system.
Ruben and van den Berg (2001) analyze the role of the RNFE of farm households, using
a national expenditure and income survey from 1993 to 1994 in Honduras. Among
other things, they conclude that rural non-farm income can be used as a capital source
to finance investments used to optimize yield and labor productivity. This linkage is
most pronounced if credit constraints occur as non-farm activities can be considered
the collateral for borrowing. Anriquez and Daidone (2009) explore the effect of the
21

Chapter 2: Literature review
growing RNFE on farm diversification, household input demands and production
efficiency in Ghana. The linkages between agricultural and RNFE are measured by a
household level input distance function. They conclude that the expansion of the RNFE
increases investments in most agricultural inputs. Stampini and Davis (2009) examine
the use of agricultural inputs as a result of the relationship between participation in
non-agricultural labor activities and farming production decisions. The authors use a
longitudinal survey from 1993 to 1998 in Vietnam. They state that households that
engage in non-agricultural labor activities spend significantly more on agricultural
inputs. In case of incomplete credit markets, participation in RNFE relaxes credit
constraints through credit provision for purchases of agricultural input. Also, the share
of the RNFE in total rural income increases over the same period.
Maertens (2009) mentions that access to low-skilled RNFE has alleviated farmers‟
liquidity constraints, resulting in the increase in smallholder agricultural production. The
author uses a household survey dataset in the main horticulture region in Senegal. The
results prove that the income from employment in the horticulture export industry is
used to invest in the farm, resulting in higher expenditures, higher farm incomes and
larger farm sizes. Entry barriers in credit and input markets can be tackled by the
RNFE. Hertz (2009) documents the relationship between non-farm income and
agricultural investments in Bulgaria. More specifically, the influence of non-farm wage
employment and pensions on expenditures, working capital and investments in
livestock is studied. Farm expenditure is regressed against RNFE and other variables in
a two part model. This research points out that farmers fund farm expenses from nonfarm
income and use credit for consumption. The latter implies that if credit is available
or the access to credit would be enhanced, the borrowed funds are not used to finance
agricultural investments but used for consumption purposes.
Pfeiffer et al. (2009) use the 2003 National Rural Household Survey Dataset from
Mexico to explore the effect of the RNFE on agricultural production activities.
Instrumented variable models were used to study whether households with and without
RNFE differ in farm decisions. The authors state that RNFE negatively influences the
family labor in crop production and farm output, but increases the use of purchased
inputs. As access to RNFE is increased, households will earn more and this induces a
shift out of cropping. This effect is caused by both the relaxation of the credit constraint
but and the effort by households to seek for family labor substitutes when the returns
of non-farm work increase. Finally, Woldenhanna (2000) investigates the impact of
non-farm employment and income on farm households and agricultural production. His
analysis uses a farm household model with liquidity constraints build upon a farm
22

Chapter 2: Literature review
household survey from Tigray, northern Ethiopia. He finds that farm households with
more diversified sources of income have a higher agricultural productivity. Non-farm
income surplus helps households to finance farming activities such as the investment in
labor and inputs (seeds, fertilizer and pesticides). When non-farm income increases by
1% expenditures on variable input will increase by 0.43%. Hence, the author concludes
that non-farm income can be complementary to farm income if households face
borrowing constraints.
2.5 Virtuous circle
The existence of a virtuous circle is important for rural development. When households
face liquidity constraints, they will be pushed into non-farm activities. Promoting the
latter makes access to non-farm employment easier for households. In this way,
households increase their income and diversify their income sources. This income
surplus will then be reinvested in the farm, facilitating farm modernization and
commercialization. It is assumed that this will increase agricultural productivity and
demand for agricultural labor which ends the virtuous circle (Davies et al., 2009).
Investment linkages should enhance rural modernization, commercialization and
specialization, because the increase in agricultural productivity enables households to
participate in the RNFE without lowering agricultural production (Kilic et al., 2009).
The hypothesis of such a virtuous circle is however not new. John Mellor, (undated)
cited by Lanjouw and Lanjouw (2001), observes a virtuous cycle emerging whereby
increasing agricultural productivity and farm income would be magnified by multiple
linkages with the RNFE. Increased agricultural income would enlarge demand for goods
and services, but also potential linkages through the supply of capital and labor were
assumed to occur. The increased agricultural productivity will reallocate labor or
increase income so that the new agricultural surplus could be used for investment in
the RNFE. The assumed growth in the RNFE would in turn stimulate further growth in
agricultural productivity via lower input costs and the fact that profits are invested back
in agriculture and technological improvements. Mellor believes the income and
employment in the agricultural and non-farm sector are mutually reinforced when the
both sectors grow.
If the virtuous circle occurs, it is important to promote the access to non-farm activities
and to make sure households can enter the RNFE sector. The existence of entry
barriers hinder poor households to participate in certain non-farm activities. These
23

Chapter 2: Literature review
entry barriers can be present in the form of abilities or credit (Dercon and Krishnan,
1996). Some non-farm activities require better education and specialized skills. Credit
constraints hinder poor households to make the required investments in capital and
equipment necessary for the participation in the RNFE. As a result, income
diversification in high return niches within the RNFE will be more difficult to obtain for
poor households than for rich households. More educated, skilled and better endowed
households will have better access to high-return non-farm activities (Dercon and
Krishnan, 1996; Barrett et al., 2001; Woldenhanna and Oskam, 2001).
The existence of entry barriers hinders households to participate in non-farm activities,
and as a result the gained non-farm income cannot be used as a liquidity source for
farm investments. The occurrence of entry barriers determines whether investment
linkages are likely to occur. Barrett et al. (2001, p. 324) suggest that “those with the
least agricultural assets and income are typically also least able to make up this
deficiency through non-farm earnings because they cannot meet the investment
requirements for entry into remunerative non-farm activities”. Davies et al. (2002)
point out that the lack of access to credit can hamper the linkage between RNFE and
agriculture in several ways. First, households‟ capacities to expend their current
activities may be limited, undermining the households‟ ability to exploit the
opportunities for selling to agribusiness. Credit is necessary to enter in these activities,
which will foster the development and expansion of the RNFE. Access to credit will not
be a guarantee for this, but it will facilitate the process. Second, credit is necessary to
adopt new crops and technology in order to produce the required processing quality.
An interesting study in the Tigray region of Ethiopia has been conducted by
Woldenhanna and Oskam (2001). They prove that increasing access to RNFE can
expand the economic activity in Tigray, but due to entry barriers, relative wealthy
households will dominate the most profitable RNFE such as petty trade, masonry and
carpentry. They conclude that constrained access to credit and liquidity is the most
important underlying factor that hinders participation in RNFE. However, van den Berg
and Kumbi (2006) present evidence that entry barriers for the poor to participate in
non-farm sector are low and they therefore participate actively in non-farm activities in
the Oromia state. The non-farm sector provides an alternative to use labor excess from
agricultural production. They conclude that further growth in the non-farm sector will
not increase income inequality, as is sometimes assumed.
24

Chapter 3: Methodology
3 METHODOLOGY
3.1 Study and survey area
The survey area is located in Tigray, the most northern Federal State of Ethiopia
(Figure 3.1). Tigray is divided into six zones, which are further subdivided into 34
woredas. Each woreda contains a number of tabias which are regarded as the lowest
administrative hierarchy (Negash, 2008). According to the Central Statistical Agency of
Ethiopia (CSA), Tigray covers an area of more than 50 thousand km² and has an
estimated population of 4.6 million with an average annual population growth rate of
above 2.6% (CSA, 2008; cited by Negash, 2008). The altitude in the region ranges
between 300 meters above sea level (masl) in the east and more than 3,000 masl in
the north and central part, covering three agro-climatic zones: lowland (kolla: below
1,500 masl), medium highland (woina dega: between 1,500 and 2,3000 masl) and
upper highland (douga: between 2,300 and 3,200 masl) (Kidane, undated).
Figure 3.1: Map of Ethiopia (small) and Tigray Regional state
source: Figure created by the local organizing committee of the International Congress Water
2011 [online], available at http://ees.kuleuven.be/water2011/excursions/index.html [date of
search: 13/04/2011]
25

Chapter 3: Methodology
The analysis of the farm/non-farm linkage is based on the survey of rural households
conduced in 2009 at a sample of 734 households in the Geba catchment. The research
site covers 4600 km² and contains 8 woreda and 168 tabias in the Tigray region. Using
a structured questionnaire, a survey was conducted in Tanqua, Samre, Atsbi and Wukro
woreda (Figure 3.2). The woredas were not chosen at random as it would be impossible
to survey the extensive area and the available time and budget were limited. Moreover,
it is in line with the research areas of the MU-IUC Collaboration Program. The woredas
in the Geba Catchment were stratified into three agro-climatic zones. Based on
population, four woredas were picked randomly: one from the highland, two from the
middle-highland and one from the lowland. By this, the survey was designed to be
representative for the whole region, as the contrasting socio-economic (access to
market in woreda and Mekelle) and agro-climatic (altitude, temperature and rainfall)
zones were reflected. From each woreda, samples were drawn randomly to constitute
sample research sites. As a result, from each Woreda two Tabias were selected
randomly (so 8 Tabias in total): Rubafeleg, Barka, Negash, Adiqsanded, Addisalem,
Andewoyane, Lemlem and Hadnet. Households were drawn from each tabia based on
their respective population size and participation in agricultural extension service, with
densely populated tabias getting a relative higher number of households quota.
Figure 3.2: Research area
Source: BoANRD 2004 (Bureau of Agriculture and Natural Resource Development)
26

Chapter 3: Methodology
Next to the variability of rainfall and susceptibility to droughts, the fragile Ethiopian
highlands still suffer from natural soil erosion depletion and other land degrading
processes, low soil fertility and aridification (Pender, 2000; Block and Webb, 2001;
Holden et al., 2004; Ehui and Pender, 2005). These natural problems are enhanced by
population pressure, intensive land (over)use by livestock grazing and cultivation,
crop/animal pests, cultivation of marginal lands on steep slopes and deforestation
(Woldenhanna, 2000; Negash, 2008; Tesfay, 2009). Furthermore, the region has
suffered from droughts, famines, food insecurities, civil wars, political conflicts, border
conflicts, poor governance, underdeveloped infrastructure, restricted access to formal
financial institutions and underdeveloped education or training (Pender, 2000;
Woldenhanna, 2002; Negash, 2008; Tesfay, 2009). This all resulted in environmental
and ecological problems, degraded and fragmented land and poor resource
management and hence even poorer performance of agriculture (Woldenhanna, 2000).
Farm households participate in a wide variety of both farm and non-farm activities.
Subsistence agriculture 1 remains the main occupation of the rural people in Tigray
which is highly dependable from variability in rainfall and recurrent droughts (Tesfay,
2009). Teff is the main staple food and occupies the largest share of cultivate land, but
maize production is increasing in importance (Pender et al., 2006). Woldenhanna
(2002) finds that in 1996, 78% of the households was engaged in mixed farming, while
only 19% was engaged in cropping only and 3% in livestock husbandry only. Next to
this, 81% of the rural households participated in various non-farm activities. He also
shows that, on average, farm production accounted for 57% of the total income and
hence non-farm activities 43%.
In 1996, the average farm size was 0.97 ha and 70% of the households owned less
than one ha (Woldenhanna, 2002). The average farm size now ranges from 0.6 ha in
the eastern part to 1.2 ha in the western part and average landholdings generally
decrease with altitude. Agricultural production is below the natural average, on average
below one ton per hectare, even in good years (Woldenhanna, 2002). Yields have
increased only marginally because of increased maize yields and expansion of the
fertilized area (Dercon and Christiaensen, 2010). As a result, agricultural production is
not able to support farm households of five to six members for longer than six months
(TBORAD, 2008; as cited in Kidane, undated). Deforestation, continuous cultivation,
1 mixed farming, subsistence oxen plough single cropping cereal crop dominated combined with livestock
rearing production is the typical farming system (Kidane, undated)
27

Chapter 3: Methodology
soil nutrient depletion, using dung and crop residues as fuel and severe soil erosion
have resulted in low fertilizer application use (below 15kg/ha). Farming systems in
Tigray are characterized by traditional technology use, mainly rain-fed land and animal
traction. During drought years, people become fully dependent on food aid (Kidane,
undated). It is assumed that further increase in agricultural employment is difficult
because of the above mentioned constraints (Woldenhanna, 2002).
In 2003, the Agricultural Sample Enumeration (CSA, 2003; as sited in Kidane, undated)
reported the total livestock in Tigray to be 10.8 million animals of which 2.8 million
cattle, 1.8 million goats and 0.7 million sheep. Households on average own 3.2 units of
cattle. This makes the Ethiopian herd to be the largest in Sub-Saharan Africa (Pender et
al., 2006). The average milk production is low, about 1.5 liters per day per cow.
Woldenhanna (2002) states that livestock plays a secondary role, but oxen plough
stays very important. Also, the lack of pasture, fodder and scarcity of veterinary clinic
constrain livestock development. Moreover, after a drought, the revival of livestock
farming is difficult because a significant number of the livestock dies. There are almost
230,000 beehives in Tigray. The annual honey production is estimate at 19,000 ton of
which 98% comes from traditional beehives.
3.2 Survey design
The survey was conducted under the supervision of Kidane M.G. Egziabher, as part of
his Ph.D. to study the impact of agricultural extension on household income and income
diversification in 2009. Refer to Kidane (undated) for a description of the methodology
used in this research. Both qualitative and quantitative methods were used to collect
data. The qualitative method included in-depth semi-structured interviews and focus
group discussions with policy makers, agricultural researchers, development
practitioners, farmers and development agents. The quantitative method included 500
rural household surveys and 200 household choice experimentation surveys using
questionnaires.
The dataset contains detailed information collected using questionnaires about the
households‟ head individual characteristics (age, gender, education, …), households‟
characteristics (family size, access to irrigation, …), households‟ access to capital
(finance, landholdings, livestock, fixed, …), types of households‟ occupation (migration,
transfer, non-farm and farm), local conditions (regional variables, access to markets
and capital, …), investments (durables and inputs) and consumption (food and non-
28

Chapter 3: Methodology
food). The questionnaire designed for the household survey was planned to be collected
in two rounds and was administered for each of the selected head of household to
collect the relevant information by deploying trained enumerators. However, this
research only uses first round data (cross-sectional data set). Separate checklists were
developed to gather secondary information. Focus group approach (administered by the
researcher) was used to collect in-depth information through discussion on a wide
range of issues related to households‟ recruitment criteria to extension package
programs, technology selection, farms and policy makers perception regarding
programs performance and possible perspectives as how to improve the lives of
farming communities.
Qualitative data included information about extension (institutional linkages,
coordination, research outputs and priority setting processes), farmers‟ satisfaction
level (extension, marketing and credit systems), development agents‟ skills and
farmers‟ attitude. This qualitative information was gathered through desk review of
government policy documents, previous studies and related information available in the
study area. Relevant information for the desk review was acquired from the Bureau of
Rural Development and Agriculture, Regional Agricultural Research Institute, Regional
Marketing Agencies, Food security Office, Micro-Finance Institutions, Local Government
administrators and Cooperative Development Offices and ngo‟s active in the area.
The validity of the research questionnaires and checklists was verified through pilot
tests. Based on the results of pilot test, to revise and modify the questionnaires were
revised and modified. To improve the quality of the survey, qualified enumerators were
recruited and trained. Despite this careful data collection it is reasonable to expect
some underestimation of the actual income, expenditures and assets. This is due to the
sensitive nature of the information on income and expenditure, the limited capacity to
recall on all the transactions by the respondents and unwillingness of the respondents
to disclose to outsiders. These are however natural shortcomings and the data will give
a reasonable picture of the research sites.
3.3 Data analysis
Our primary interest lied on the analysis of the impact of non-farm income (x) on farm
investments (y). We First defined these investments as the total household expenditure
on farm activities. These expenditures included both investments in durables and farm
input use. Afterwards, we narrowed our analysis to both types of investments.
29

Chapter 3: Methodology
Households‟ durables included land and water conservation measures, animals,
buildings and equipment. We expanded our analysis to these different types of farm
investments. Farm input use included local and improved seeds, fertilizer and labor. In
the empirical specification of the econometric model used, we used the general term
„investments‟ and the latter might refer to either of both investments. Our most
important independent variable was non-farm income. This was defined as the income
earned by farm households by participating in the RNFE sector. Next to this, we
included a large set of observable variable in the model. These control variables include
(1) household head age, sex and schooling; (2) households landholding, animal assets
and fixed assets; (3) distance to Mekelle and (4) edir membership, access to irrigation
and number of adult working labor forces.
Non-farm activities were defined as rural non-farm labor activities and therefore
consisted only of wage income and self-employment. We thus not included migration
and transfer income in the non-farm income. The rationale behind this definition is that
transfer income is not a real non-farm labor activity as it is support received from the
government or ngo‟s. As migration income is not influencing households‟ decisions
during the time of production season, we prefered not to include it in our analysis.
Migratory activities contrast with rural activities because households‟ allocation of labor
resources during the production season does not affect migratory activities. It is
assumed that wage and self-employment activities are important for farm households.
In order to avoid sample selection bias we added a positive constant to the non-farm
income variable and take logarithms: ln(x+c). The same is done for farm investments,
because the logarithmic form is also necessary to correct for the clear positive
skewness at the right side of the income and investment variable. This prevented
outliers to influence our estimates, especially in small samples.
3.3.1 Ordinary Least Squares estimation
The effect of non-farm income on farm investments was analyzed by comparing
households that have access to non-farm income with those households that do not,
while controlling for a set of observable factors. The systematic differences between
these two groups of households were captured through the inclusion of these
observable individual and household characteristics. The following regression equation
estimates the effect of non-farm income on farm investments, and is referred as the
structural equation:
y=β 0
+β 1
x+β i
n i + u (eq. 1)
30

Chapter 3: Methodology
where β 0 is the intercept, β 1 is the parameter associated with non-farm income, β i the
parameter of the i th explanatory variable n i and u the error term. Assume that the
vector N i of explanatory variables is exogenous, but x is correlated with u and is
therefore an endogenous explanatory variable, this is Cov(x,u)≠0. The latter condition
causes the Ordinary Least Squares (OLS) estimation of eq. 1 to be inconsistent
(Wooldridge, 2002b).
3.3.2 Omitted variable bias
Cov(x,u)≠0 is caused by the fact that u contains omitted variables that are correlated
with x, but not with N i . This means that unobserved variables are present that influence
both our dependent variable farm investments and our most important explanatory
variable non-farm income. Excluding them from the regression analysis makes x
correlated with u and thus the zero conditional mean assumption to fail. As a result, the
coefficients in a simple OLS regression will be biased. This is termed omitted variable
bias or unobserved heterogeneity. The most optimal solution to this problem is the use
of panel data to control for time-invariant unobserved households-level fixed effects,
find a suitable proxy variable for these unobserved variables or use the instrumental
variable (IV) method. In the absence of the first two possibilities, we will use the IV
regression model. As many researchers dealing with this problem have pointed out: the
solution is to cure the disease rather than prevent it (Wooldridge 2002b).
It is most likely that non-farm income is correlated with unobserved variables that also
influence farm decisions. Several different possible sources of omitted variable bias are
found in the literature, such as Hertz (2009), Kilic et al. (2009), Oseni and Winters
(2009) and Maertens (2009). This literature has pointed out that it is impossible to
determine a-priori the direction and magnitude of the omitted variable bias. Farm and
non-farm income activities are complex related and our data set includes many sources
of non-farm income and farm activities. Many potential effects could therefore be
present and it is impossible to assume one mechanism to be dominant. It is more likely
that several factors will have a plausible effect.
Entrepreneurship and general ability are assumed to have a positive effect on most
economic activities, including both farm investments and non-farm activities. Economic
motivation might drive household members to seek for non-farm employment as well
as to invest in their farming activities. Both unobserved variables would lead to an
upward bias in the estimated effect of non-farm income on farm investments. However,
31

Chapter 3: Methodology
economic motivation can also force households to gradually leave farm activities as
non-farm activities have a better pay-off and therefore spend less money on farm
activities. Also risk aversion might have such ambiguous effect. On the one hand, it
may stimulate households to participate in non-farm activities in order to finance farm
investments. By this, they reduce the risks associated with farm activities. On the other
hand, risk aversion might lead to a downward effect: households are avoiding costly
investments in farm activities and focus on non-farm activities to diversify their income
sources. Households that face more constraints to agricultural production might allocate
more labor to non-farm activities.
3.3.3 Two Stage Least Squares estimation
The bias in the estimation can be reduced by including as much as theoretically
relevant control variables as possible in our analysis. These control variables should be
correlated with the unobservable variables. It is however impossible to exclude all
variation in the error term and a more specialized method is necessary. To overcome
the problem of omitted variables here, we use the IV approach in order to isolate
exogenous variation in non-farm income. This approach tries to solve the omitted
variable bias problem by leaving the unobserved variables in the error term and using
estimation methods that recognize the presence of omitted variables (Wooldridge 2002,
Wooldridge 2002b). By finding appropriate IV we isolate the movements in x that are
uncorrelated with y.
We will describe the Two Stage Least Squares (2SLS) estimation model following
Wooldridge (2002, 2002b). Recall the structural equation eq. 1 where farm investments
is the endogenous dependent variable, non-farm income x is the main explanatory
variable of interest, that is likely endogenous, and N i is a set of exogenous variables.
Assume that z j is a set of exogenous variables that are not included in eq. 1 and which
are assumed to be a set of valid instruments. An appropriate IV candidate for x must
satisfy two econometric assumptions. First, the instrument z j should be uncorrelated
with the error term u: Cov(z j ,u)=0. This is often referred as the orthogonality
assumption and indicates that the instruments z j should have no partial effect on
investments y and z j is uncorrelated with other factors that affect y. Second, the
instrument must be correlated with non-farm income, this is Cov(z j ,x)≠0, meaning that
z j must be related to the endogenous non-farm income (Wooldridge, 2002b).
32

Chapter 3: Methodology
The regression equation (eq. 1) is solved using a two stage least squares (2SLS)
model. In the first stage of the 2SLS model, x is written as a linear function of the
exogenous variables (instruments and control variables) and an error term v. This
equation is the reduced form equation for x:
x=η 0
+η j
z j +η i
n i +v (eq. 2)
Out of all possible linear combinations of the exogenous variables that can be used as
instruments for x, 2SLS will choose that which is the most correlated with x. The key
identification assumption is that there is at least one of the π j or ≠ 0. The best IV for x
is the linear combination of z j and n i which is called x * ,
x * =η 0
+η j
z j +η i
n i (eq. 3)
and is uncorrelated with u. By this, we can see that eq. 2 consist of two parts: x * and
v. x * can be interpreted as the part of x that is uncorrelated with u while v is possible
related to u. To obtain the fitted value x (this is the estimated version of x * ), a
regression on eq. 2 is run:
x=η 0
+η j
z j +η i
n i (eq. 4)
Therefore, the first stage of 2SLS purifies x of its correlation with u before doing an OLS
regression. It estimates the relationship between on one side the control variables and
the IV and on the other side non-farm income. By this, the component of non-farm
income that is uncorrelated with the error term is isolated. The second stage uses this
component to estimate the coefficient of non-farm income in the farm investment
regression. The second stage is the OLS regression, but with x instead of x, which gives
us the reduced form for y:
y=β 0
+β 1
x * +β i
n i + u +β 1
v (eq. 5)
The composite error u + β 1 v has zero mean and is uncorrelated with all explanatory
variables x * and n i . the OLS will give unbiased results for the reduced form parameters
because of the latter feature.
Another advantage of IV methods is that it is a general solution to isolate exogenous
variation in non-farm income (Wooldridge, 2002b). Simultaneity bias could occur when
the decision to allocate labor (to farming, but also to non-farm activities or leisure) are
made simultaneously with decisions related to expenditures on farm investments. The
non-farm income is therefore related with the error term. If large investments in farm
activities free up labor time devoted to farming, this labor time could be allocated to
non-farm activities. This would create a positive relation between non-farm income and
33

Chapter 3: Methodology
the error term, leading to an upward bias in our estimates on its effect. It is however
also possible that farm investments intensify farm activities and thereby drawing labor
away from farm activities. This would reduce non-farm income and create a downward
bias (Hertz, 2009). Also Stampini and Davies (2009) suggest the OLS regression to be
downward biased because farming and non-farm labor decisions are made jointly. They
suggest that the more households farm, the less likely they are to participate in nonfarm
activities.
The use of 2SLS methods brings however also some limitations. First, the IV technique
produces less precise estimates as it exploits only a part of the correlation between the
dependent variable and the exogenous regressors. This correlation is determined by the
correlation between the exogenous regressors and the instruments (Stampini and
Davis, 2009; Oseni and Winters, 2009). Moreover, the calculated causal effect places
greater weight on those households most influenced by the instruments and thereby
capturing the effects of a subpopulation (Oseni and Winters, 2009). Next to this,
Wooldridge (2002) states that an important cost related to the IV estimation is the
larger asymptotic variance of the IV estimator which is sometimes much larger than the
asymptotic variance of the OLS estimator. This implies that 2SLS are inherently biased
as their standard errors are larger than those from OLS. The magnitude of the standard
errors of 2SLS depends on the quality of the IV used in estimation. Therefore, solving
the omitted variable bias always comes with the cost of efficiency.
3.3.4 Instrumental Variables
As stated before, appropriate instruments must satisfy two assumptions. Good
instruments should be both valid and relevant, this is orthogonal to the errors and
correlated with the endogenous covariate non-farm income (Baum et al., 2003). The
orthogonality assumption can never be checked as the error term u is unobservable.
However, if an equation is overidentified, we can test whether the instruments are
uncorrelated with the error process. Roughly speaking, we test whether the
instruments are correlated with farm investments or not. This is what is called
instrumental exogeneity and can be tested through the overidentifying restriction test.
As we only can suggest instrumental exogeneity instead of proving it, we must be sure
that the instruments have strong theoretical grounds. The second assumption is often
referred as the relevance of instruments and can be tested by the significance of the
instruments in the first stage IV regression, the underidentification test and weak
instrument identification test. Variation in relevant instruments causes variation in non-
34

Chapter 3: Methodology
farm income which is uncorrelated with the error term. Underidentification test is a test
whether the regression is identified, meaning correlated with the non-farm income, and
is essentially a test of the full column rank of a matrix. Weak identification refers to
instruments that are weakly correlated to non-farm income, causing estimators to
perform poor (Wooldridge, 2002b; Baum, 2006). These tests will be discussed together
with the results.
Finding appropriate instruments is not obvious. Ideally, a set of instruments or
alternative sets of instruments should be used to avoid focusing on a subpopulation
(Oseni and Winters, 2009). The best instruments are those that reflect push and pull
factors driving households into non-farm activities but have no direct impact on farm
activities. Such factors could include distance to nearby a factory, literacy in English
and so on. During the time of survey, not much attention was given to collect
information related to push and pull variables. We therefore lack information about
distinct push and pull factors, so alternative sets of instruments were not found. We
were however able to construct one push and one pull variables that might serve as
strong instruments.
Our first instrument is the number of dependents in the household (dependents). This
variable includes all children younger than 14 years and all elderly older than 65 years.
It is assumed that these members of the household are no working forces at the time
of survey. They must therefore be supported by the labor forces within the household.
Households are pushed into non-farm activities because they must find means to
maintain all their dependents. It is hypothesized that the number of dependent
stimulate households to allocate labor forces to non-farm activities for additional
income and that it does not affect farm investments directly. Its effect is indirect,
through its effect on non-farm income. However, it might be unrealistic to think that
the number of dependents has no influence on farming decisions in any way. The more
dependent people in the family, the more effort that needs to be devoted to farming
and food production. The tests of instruments validity and weakness will decide
whether the number of dependents is an appropriate instrument.
The second instrument is the non-farm income share in the total household income at
tabia level (non-farmshare_tabia). This variable is calculated as the mean of the
households‟ non-farm income share for each tabia. The latter is constructed by dividing
the non-farm income by the total household income for each household in the tabia.
This non-farm income share is a proxy for the non-farm employment rate at the tabia
level. It indicates the households‟ potential to diversify their income sources and to
35

Chapter 3: Methodology
increase their non-farm income if they are willing to do so. This variable can thus be
used as an indicator of non-farm employment opportunities and a predictor for nonfarm
income. The existence of non-farm activities on its turn is a proxy for the demand
for non-farm labor. We hypothesize that the non-farm income share positively
influences non-farm income and has no direct effect on farm investments. The fact that
the instrument is defined at tabia level and we control for regional effects (through the
distance to Mekelle) makes it unlikely that the instrument is correlated with other local
factors that affect farm activities (Kilic et al., 2009).
36

Chapter 4: Results and discussion
4 RESULTS AND DISCUSSION
4.1 Descriptive statistics
4.1.1 Farm expenditure, input use and investments
Our data contain information about the value of households‟ investments in farming
activities. As described in section 3.3, the latter consist of investments in durables as
well as expenditures on farm input. All investments are expressed in Ethiopian Birr 2
(ETB). The total value of all investments made in farm activities is defined as the total
farm expenditure and hence includes all money spent by households on durable
investments and farm input use. Table 4.1 shows that the value of farm expenditures is
on average 3,247 ETB. 96% of the households in the Geba catchment spends at least
some money on farm expenditure and this value ranges from zero to 44,000 ETB. More
importantly, Table 4.1 shows that investments in farm durables make up the major
part (79%) and farm input use only a minor part (21%) in the total farm expenditures.
Table 4.1: Households’ total farm expenditures, input use and investments in ETB
Variable Mean Std. Dev. Min Max Share Access
Farm expenditures 3,246.78 4,099.92 0 44,000 1.00 0.96
Farm input use 693.56 580.47 0 3,684 0.21 0.90
Farm investments 2,553.22 3,956.55 0 43,250 0.79 0.86
Input use types
Fertilizer 209.96 245.05 0 1,620 0.26 0.62
Improved seeds 109.27 206.87 0 1,600 0.14 0.37
Local seeds 361.29 346.78 0 2,634 0.57 0.88
Other inputs 10.42 33.41 0 368 0.02 0.13
Labor 2.62 7.95 0 87 0.01 0.18
Investments types
Water and Land 557.77 1,567.56 0 22,750 0.29 0.50
Livestock 1,039.22 1,750.35 0 23,400 0.39 0.53
Buildings 718.94 2,470.19 0 31,000 0.15 0.22
Equipment 237.29 873.47 0 15,950 0.17 0.59
Notes: observations=733, shares calculated if value > 0 (input use= 661 obs; investments= 663 obs)
2 One ETB = 0,06397 € (National Bank of Ethiopia, http://www.nbe.gov.et, 08/06/2009)
37

Chapter 4: Results and discussion
To get a view of the farming system, households were asked about their usage of
fertilizer, improved seeds, local seeds, other inputs and labor usage. 86% of the
households has used any of these inputs during the last production season. The total
value of the agricultural input use is on average 694 ETB, with a maximum amount of
3,684 ETB. Almost nine out of ten households has access to local seeds which makes
up the biggest part of their spending (361 ETB). Out of the total sample households,
62% of them has used some fertilizers on their fields, which on average has a value of
210 ETB. One on three households has used improved seeds during the last production
season, with an estimated value of only 110 ETB on average. 13% of the households
has used other inputs, but the value of these inputs is low (10 ETB). Also the use of
farm labor seems to be rather low, only 18% of the households has actually spent
some money on labor during the last production season. The local seeds have the
highest share in total input use (57%), followed by fertilizer(26%) and improved seeds
(14%). The contribution of both labor and other inputs is very small, respectively 1%
and 2%. As we expected, the values of the fertilizer and improved seeds used by
households are low and indicate that farm input use is restricted. Households may
hence face liquidity problems to buy farm inputs and rely on local seeds.
Other studies have found that household farm input use is rather low in Ethiopia.
Pender et al. (2006) find that households‟ in Tigray on average cover only 27% and 2%
of their plots with respectively fertilizer and improved seeds. The total use of seeds
amounts to 118 kg/ha. Dercon and Christiaensen (2010) report that on average 22% of
the households use fertilizer in Ethiopia. They note that fertilizer use in Ethiopia has
remained limited because of limited knowledge and education, risk preferences, credit
constraints, low profitability of fertilizer use, lack of market access as well as limited or
untimely availability of inputs. The low use of farm input is in contradiction with the
efforts done by the government in Ethiopia to promote the adoption of modern varieties
and accompanying inputs in order to boost productivity. The Ethiopian government
supports two holding companies who control the market for fertilizer. In addition, spot
markets are very thin. Fertilizer use is heavy promoted by the government extension
and credit program (Pender et al., 2006; Dercon and Christiaensen, 2010).
Herweg (1993), as cited by Pender et al. (2006), notes that fertilizer use is more risky
and less profitable in low-rainfall areas due to the fact that the uptake of nutrients may
be limited by inadequate soil moisture. Contrary, adoption of soil and water
conservation measures might be less risky and more profitable here because they have
a larger impact on yields in the short term. Also Ehui and Pender (2005) and Pender
(2004) state that in areas with lower agricultural potential, intensification of agriculture
38

Chapter 4: Results and discussion
through fertilizer use and improved seed use is more limited. Investments in irrigation,
water harvesting or water and soil conservations to overcome soil moisture constraints
have more potential.
Table 4.1 describes the value of expenditures for households‟ investment on durables
during the last production season. We see that 86% of households in the Geba
catchment makes farm investments of any kind, and spends on average 2,553 ETB on
farm investments. Following Table 4.1, the households‟ investments can be divided in
four big groups. First, households make investments in water and land which include
investments in terracing, water pond, water well and water diversion projects. At least
half of the households spends some money on these investments, with an average
amount spent representing 558 ETB. Second, investments are made in the households‟
livestock resources. More than half of the households has made at least some
investments in cows, oxen, sheep and goats, poultry and/or equine (donkey, horse and
mule). Households spend on average 1,039 ETB on animal related investment. Third,
investments in buildings include investments or renovation in respectively new and old
houses for both human and animal purpose. Only one out of five of the households
spends money on their housing and invests on average 719 ETB. Finally, investments in
treadle pump, generator, plow, axe and spade and beehive and bees are collected in
investments in equipments. 59% of the households spends money on equipment, which
represents an average value of 237 ETB. Table 4.1 concludes that the share of livestock
investments in the total households‟ investments (39%) is the highest, followed by the
share of investments in water and land (29%). The shares of investments in buildings
and equipment are lower, both close to 15%.
4.1.2 Farm and non-farm activities
Table 4.2 summarizes the different farm and non-farm activities undertaken by
households in the Geba catchment. Of the total 734 households, one household has
reported to have no income of any kind, and is therefore excluded from our analysis.
The total household income is on average 8,301 ETB ranges from 315 ETB up to
213,425 ETB. 97% of the households in the survey has access to farm income, which is
the major share of the households‟ income (65%). Farm activities refer to the
production or gathering of unprocessed crops, livestock, forest or fish products from
natural resources (Barrett et al., 2001). Farm income can be broken down into crop
income and livestock income. Crop income is derived from selling crops during the
three different seasons (Kiremt, Belg and Irrigation). Households cultivate different
39

Chapter 4: Results and discussion
crops, mainly teff, barley, wheat, maize and sorghum. Livestock income is generated
from selling livestock such as oxen, cows, heifers, calves, sheep, goats, horses, camels,
mules, donkeys, beehives and poultry.
Table 4.2: Households' income composition in ETB
Income Obs Mean Std. Dev. Min Max Share Access
Total 733 8,302.58 10,841.39 315 213,425 1.00 1.00
Farm 733 5,639.93 6,273.28 0 67,410 0.65 0.97
Crop 733 4,883.96 6,025.16 0 67,200 0.55 0.94
Livestock 733 755.97 1,181.34 0 13,877 0.10 0.66
Non-Farm 733 1,873.39 4,358.04 0 97,000 0.27 0.80
Wage 733 1,439.41 3,316.94 0 80,000 0.22 0.74
Business 733 433.98 2,070.61 0 44,000 0.05 0.22
Transfer 731 596.41 7,427.93 0 200,000 0.06 0.29
Migration 730 190.84 821.21 0 10,500 0.02 0.13
Note: missing values are interpreted as zeros in the calculation of income shares
Table 4.2 describes the different types of non-farm activities available in the dataset.
Non-farm activities are defined as the opposite of farm activities: all activities
undertaken by rural households that are not related with agricultural production on the
farm (Barrett et al., 2001). Distinction is made based on the nature of the non-farm
activities. First, distinction is made between labor and non-labor non-farm activities.
This distinction is crucial because non-labor activities do not have an impact on the
labor availability of the farm household, and thus require no farm labor input. The nonlabor
activities are present in the form of transfers from governments, ngo‟s (food aid,
cash or in kind), friends or relatives. 29% of the households has access to transfer
income which is on average 596 ETB. It only has a minor share in the total household
income (6%).
Labor non-farm activities can be further divided into self-employment, wage activities
and migration. Self-employment (or business) activities include any activities run and
owned by the household located in their home or nearby towns. Self-employment
activities include weaving, milling, handicraft, trade in grain, trade in livestock,
traditional healer (or religious teacher), transport by pack animal, selling cactus, selling
wood and charcoal, selling food and/or drinks and others. Wage activities include all
types of agricultural and non-agricultural wage employment such as farm worker,
traditional labor sharing, professional (teacher, government worker, administration,
health worker, clerical), skilled laborer (builder, thatcher, barber), trader, soldier,
40

Chapter 4: Results and discussion
driver/mechanic, unskilled worker, domestic servant, food for work and others.
Migration remittance is the income sent back or brought back by a migrant labor
members living in other parts of Ethiopia or in foreign countries. Migratory refers to
domestic urban and foreign activities and therefore does not come to terms with rural
activities.
Several authors (Dercon, 1998; Barrett et al., 2001; Woldenhanna and Oskam, 2001;
Maertens, 2009) stress the importance to make a distinction between the different nonfarm
activities. Dercon (1998) states that it is incorrect to treat all non-farm activities
as the same, because possible entry and returns are likely to be different. The nature,
structure and input requirements for non-farm activities are different and pose different
demands on the household asset base. Non-farm self employment requires more
capital and skills than wage employment. If credit constraints occur, the participation in
wage employment will be easier than in self employment. Wage employment is more
likely to be available in areas near to towns and commercialized agriculture. Farm
households can participate in petty trade activities in areas far from urban centers.
Contrary, near to urban areas, non-farm employment may face serious competition
(Dercon, 1998; Woldenhanna, 2000).
Table 4.2 shows that the majority (80%) of the sample households earns non-farm
income. Hence, four out of five households has a member who is active in a non-farm
activity of any kind. Although non-farm activities have a lower share in the total
household income than farm activities, their contribution is far from marginal (27%).
Non-farm income is mainly gained from participation in wage activities (72%) while
only one on five households are engaged in self-employment activities. The mean
income generated from wage activities is also higher than the income derived from
business activities, respectively 1,439 ETB and 434 ETB. It seems as if wage income
activities are easy accessible for the households and also have a higher payoff in the
Geba catchment. This is in contrast with the conventional idea that self-employment
activities offer a higher income than wage activities. Only 13% of the households has
access to migration income which has only a share of 2% in the total household
income. The average amount of money sent back to the household by migrant
members is on average 191 ETB. These numbers are in line with the findings of
Woldenhanna and Oskam (2001) and Woldenhanna (2002) in Tigray.
The survey contains specific information about wage and self-employment activities.
The household head and his or her spouse were inquired after the nature of their nonfarm
jobs. We constructed the participation of households in the different types of self-
41

Chapter 4: Results and discussion
employment, based on the mean total days worked by the household head and his/her
spouse on the specific type of business activity, because there was no other information
available. Table 4.3 shows that households worked most of the days on trade (general
and livestock), handicraft and others. As Woldenhanna (2000) notes, farm households
need to invest some level of working capital to get started in self-employment activities
such as petty trade, handicraft, etc.
Table 4.3: The nature of wage and self-employment jobs
Percentage of total amount wage
employment jobs
Farm worker 7.22
Trad. labor sharing
1.71
Professional
0.01
Laborer
5.12
Trader
2.62
Soldier
0
Driver/mechanic
0
Unskilled worker
18.18
Domestic servant
60.09
Food for work
3.41
Percentage of total amount of selfemployment
jobs
Weaving 2.90
Milling
6.47
Handicraft
10.42
General trade
21.86
Livestock trade
11.78
Trad. healer/rel. teacher
2.36
Transport
0.87
Selling cactus
4.05
Selling wood
1.40
Selling food
10.57
others
27.32
In total, 762 wage jobs are performed by the household head an his or her spouse.
Table 4.3 describes the contribution of each type of wage employment in the total
amount of wage jobs. „Food for work‟ is the most important wage employment, which is
clearly the dominant wage job (60%). Food for work is followed by unskilled work
(18%) and farm work (7%). Household members participating in wage employment
were also asked if they needed qualification, experience or educational training to get
the wage job. The responses are reported in Table 4.4. It is clear that more than half of
the wage employments did not need any qualification or training. If such thing was
required, experiences seemed the most appropriate. Next to this, Table 4.4 shows that
most wage employment are of temporary nature. Baumeister (1996) cited by Ruben &
van den Berg (2001) notes that permanent wage labor is mainly engaged by landless
farmers, while small farmers will seek temporary employment during the off-season on
other farms in their neighborhood.
42

Chapter 4: Results and discussion
Table 4.4: Wage employment requirements and duration (in percentages)
Does the job require qualification, experience or educational training?
Experience only 24.15
Training only 10.37
Qualification/training only 0.78
Nothing 64.70
Temporary or permanent work?*
Permanent 9.82
Temporary 90.18
Note*: information on the duration of the wage employment was missing for 39 of the in total 762 wage-jobs
Woldenhanna (2000) finds very similar results in Tigray. Food for work is the dominant
wage type because it does not require experience, skill and initial capital investment. It
wage rate is however the lowest of all types of wage employment. Poorer households
are given priority when there are not enough jobs in the program. Unskilled wage
employment requires some purchase of equipment and although experience and skill
are not required, households spend a lot of time in finding such a job. Skilled wage jobs
require the most investments, skills and experience, but the wage rate is the highest.
Ehui and Pender (2005) find that during 1998-1998 food for work projects accounted
for 40% of the average non-farm income and more than 10% of the average total
household income. Households that participate in such projects also earn higher
incomes than other households (Pender et al., 2006).
Next to this, Woldenhanna (2002) states that most non-farm work in Tigray is of
temporary nature and does not require skilled labor. As Woldenhanna and Oskam
(2001) point out, most farmers in Tigray have difficulties with acquiring skills, startup
capital and jobs in non-farm activities. Until 1998, non-farm workers had to obtain their
skills and experience without training and could only acquire them during employment.
There are no schools for building, carpentry or other technical training, except some
limited training given by the Tigray Development Agency (TDA). The only organization
which provides credit to small farmers is the Relief Society of Tigray (REST). Access to
jobs in urban areas is only possible after a long process of searching, via relatives and
family.
In summary the descriptive statistics show that non-farm income is an important part
of the total household income and is accessible for most households in the survey. In
addition, some interesting information about the attitude of the households towards the
RNFE is available. Table 4.5 describes the attitude towards additional non-farm work.
43

Chapter 4: Results and discussion
Three quarter of the households has a member willing to work for additional wage if a
non-farm job would be offered to that member. Even 40% of the households has
reported to have a household member that is willing to work during planting, weeding,
harvesting or threshing season. Table 4.5 shows that agriculture is unable to absorb all
the available household labor and could indicate that households have a positive
attitude towards non-farm activities and are willing to work more non-farm if they
would have the opportunity. Similar, Woldenhanna and Oskam (2001) find that 66% of
the farmers in their sample did not work more hours in non-farm activities because
they could not get non-farm employment. They suggest that this response indicates the
fact that not all available labor could be allocated to agriculture.
Table 4.5: Attitude towards additional non-farm employment
Attitude towards Mean Std. Dev. Min Max
additional non-farm work 0.75 0.43 0 1
work for wages during planting weeding harvesting
and threshing
0.40 0.49 0 1
Note: observations=733
4.1.3 Households with non-farm income vs. households without non-farm
income
Table 4.6 describes the differences in characteristics between households that have
access to non-farm activities and those that do not. The mean difference between these
two groups is calculated and we will check if this difference is significant using a t-test 3 .
As 80% of the households has access to the RNFE (Table 4.2), we can see that 150
households do not participate in non-farm activities, while 583 households do.
Households with access to non-farm activities have on average a significant lower total
household income (8,035 ETB vs. 9,344 ETB), farm income (5,170 ETB vs. 7,470 ETB)
and households‟ resources. Households with non-farm income have lower landholdings,
number of animals and fixed assets than households that do not participate in non-farm
activities. It seems as if the participation in non-farm activities is not conditioned by the
households‟ resource base. This could indicate that non-farm activities are especially
important for less endowed households.
3 Although the normality assumption for most variables is violated, the t-statistic has approximately a t
distributions, at least in large sample sizes (Wooldridge 2000b)
44

Chapter 4: Results and discussion
Table 4.6: Comparison of households with and without access to non-farm income
Variable Mean (150 obs) Mean (583 obs) Mean difference
Income
Only farm
income=0
Non-farm
income > 0
Total household 9,344.16 8,034.59 1,309.57 *
Farm 7,469.99 5,169.08 2,300.91 ***
Fm1age 52.34 41.77 10.57 ***
Fm1sex 0.77 0.73 0.04
Fm1schooled 0.27 0.39 - 0.12 ***
hhlandholdsize 5.23 4.17 1.06 ***
animalasset 8,060.20 5,921.62 2,138.58 ***
fixedasset 21,695.56 11,086.02 10,609.55 **
tabiadismak 66.58 71.13 -4.55**
edirm 0.24 0.23 0.01
havei 0.26 0.21 0.05 *
adults 3.03 2.91 0.12
Farm expenditures 3,278.70 3,238.57 40.13
Investments
Total 2,391.77 2,594.76 -202.99
Water and Land 829.55 487.84 341.72
Livestock 911.31 1,072.13 -160.83
Buildings 360.83 811.08 -450.25**
Equipment 290.08 223.71 66.37
Farm input use 886.93 643.81 243.12
Fertilizer 285.27 190.58 94.68 ***
Improved seeds 165.56 94.78 70.78 ***
Local seeds 423.34 345.33 78.01 **
Other inputs 8.07 11.03 -2.96
Labor 4.70 2.09 2.67 ***
Notes: *,**,*** respectively significant at 10%, 5% and 1%
Households with access to non-farm income tend to have a family head that is almost
10 years younger than those in households not having non-farm earnings. This
difference is significant at the level of one percent. This indicates that households with
younger heads are more likely to participate in non-farm activities. The contrary is true
for education, indicating that non-farm activities require more education or skills.
Remarkable is that the gender of the household head, the households‟ edir membership
and the number of working forces do not vary much between the two groups and these
differences are insignificant. Households with access to non-farm income live significant
further away from the regional capital city, which is however doubtful as integration in
45

Chapter 4: Results and discussion
markets is lower. Households without non-farm income are more likely to have access
to irrigation.
More interesting is, as shown in Table 4.6, that farm expenditures differ insignificantly
between the two groups. Households that have access to non-farm income spend on
average more on total investments. However, this difference is also insignificant. The
picture is somewhat different if we divide the investments according to their nature. On
the one hand, households with non-farm income invest more in livestock and buildings,
of which only the latter is significant. On the other hand, households without access to
non-farm income invest more in water and land and equipment of which only the
former is significant. Farm input use is higher for households that do not have access to
non-farm income, then dose with access. The same can be said for the value of
fertilizer, improved seeds, local seeds and labor use. All these difference are significant.
Only the value of other inputs is higher for households who participate in RNFE, but this
difference is insignificant. So based on the descriptive statistics, some indications about
the impact of non-farm income on farm investments can be made. We will however
analyze this impact more detailed through empirical analysis.
4.1.4 Control and instrumental variables
In addition to income and investments, several individual and household characteristics
are used in the empirical estimation. These variables include a set of individual,
household and regional characteristics, as summarized in Table 4.7. The individual
variables refer to the characteristics of the household head (fm1) and include gender
(fm1sex), age (fm1age), schooling (fm1schooled) and educational level (fm1leveled).
Table 4.7 shows that on average 73% of the household heads is male and they have
an average age of almost 44 years. Only one on three household heads has had some
form of education. In case of some education, they finished on average at least primary
school. So, most households in the Geba catchment are headed by a male of middle
age who acquired limited education.
Among the household characteristics we distinguish households‟ wealth, membership in
social capital, access to irrigation and the number of adult labor forces. The households‟
wealth is measured by the households‟ resources of land (hhlandholdsize), livestock
(animalasset) and fixed assets (fixedasset). The households‟ resource base influences
the ability of the households to finance investments and to obtain credit. The land
46

Chapter 4: Results and discussion
accessible to households in the Geba catchment is on average 4.4 Tsimidi 4 (1.1 ha) and
is measured as the total plot size of the plots owned by the households. This illustrates
that most farmers in the Geba catchment are smallholders.
Table 4.7: Summary statistics of individual and household characteristics
Variable Obs Mean Std. Dev. Min Max
fm1sex 733 0.73 0.44 0 1
fm1age 733 43.93 14.61 16 87
fm1schooled 733 0.37 0.48 0 1
fm1leveled 268 2.42 1.37 1 12
hhlandholdsize 733 4.38 3.37 0 23
animalasset 733 6,359.25 8,131.34 0 111,900
fixed 733 13,257.14 50,790.70 22 1,201,296
adults 733 2.93 1.55 0 10
tabiadismak 733 70.20 29.68 42 120
edirm 733 0.23 0.42 0 1
haves 733 0.11 0.32 0 1
takenloan1000 733 0.50 0.50 0 1
havei 733 0.22 0.41 0 1
The average value of the livestock owned by the households is almost 6,359 ETB and is
measured as the sum of the number of each animal multiplied by the value of that type
of animal. The animal types taken into consideration are oxen, cows, heifers, calves,
sheep, goats, horses, camels, mules, donkeys, beehives and poultry. The households‟
fixed asset is on average nearly 11,612 ETB and includes furniture and households‟
durable, agricultural equipment, electronics, valuables and housing equipment. To
avoid simultaneity bias between households‟ wealth indicators and farm investments in
the empirical estimation, we define new variables for animal („livestock‟) and fixed
assets („fixed‟) in which all types of respectively animals and fixed assets on which
investment information is available are excluded.
Households have on average almost three adult labor forces available and it can take
up to maximum 10 persons. Almost one out of four households is member of edir,
resembling the social capital of the household. Edir, in the context of rural and
specifically the research site, is defined by Kidane (undated) as a community–based
4 1 Tsimid = 1/4 of an hectare
47

Chapter 4: Results and discussion
institution established on mutual interest of members. Its primary objective is to
support members during the time of crises, such as death of family members. Social
capital is an indicator for farmer networks and access to (better) information. One out
of five households in our sample has access to irrigation, which is highly important in
smallscale agriculture. This number is higher than the 5.7% reported by Woldenhanna
(2002). The distance to the regional capital city Mekelle (tabiadismak) is included as
regional variable, and is on average 70 km. Distance is an important factor determining
the access to the RNFE sector and markets. This variable captures many area specific
characteristics like population density, land size and various agro-climatic features.
Next to this, households were asked if they had access to savings and if they have
taken up loan in the last five years. On average only 11% of the households had
savings during the last production season, while 50% has ever taken out a loan of at
least 100 ETB. These rather low responses could indicate the existence and importance
of liquidity constraints. Information about credit and savings is endogenous to both
farm investments and non-farm income and these variables are therefore excluded
from the empirical analysis. Finally, Table 4.8 reports the descriptive statistics of the
two variables used to instrument non-farm income. It can be seen that households on
average have more than two dependent members in the household, and this number
can range from zero to seven people. The non-farm share in the tabia is on average
26%, which indicates that on average and on the tabia level, the non-farm employment
rate is slightly higher than 25%.
Table 4.8: Summery statistics of the instruments used in the 2SLS
Instruments Mean Std. Dev. Min Max
dependent 2.35 1.49 0 7
non-farmshare_tabia 26.40 10.21 15.04 42.03
Note: observations=733
48

Chapter 4: Results and discussion
4.2 Multivariate analysis
The empirical analysis was conducted using the statistical software package STATA11.
We ran a two-stage least squares regresses on equation eq. 1 using an instrumental
variable estimator, as outlined above. In stata11, the IVREG2 command will be used to
obtain an 2SLS estimation. We will first discuss the first stage and afterwards the
second stage. The former is necessary to check the relevance, significance and strength
of our instruments. Our main interest lied in the second stage, and more specifically in
the sign and significance of the impact of non-farm income on farm investments. We
compared the 2SLS estimation with the OLS estimation to check the direction of the
omitted variable bias.
After running the regression in STATA11, the homoskedasticity 5 assumption of our
regression was checked. Homoskedasticity means that the variance of the error term is
constant, conditional on the explanatory variables. It assumes that the error terms are
independently and identically distributed. If the homoskedasticity assumption is not
valid (this is non-constant variance), we speak of heteroskedasticity (Wooldridge
2002b). It is important to assume homoskedasticity, as it is needed to justify the usual
statistical test (Baum et al., 2003). Moreover, Wooldridge (2000b) states that if
heteroskedasticity is suspected, the reported standard errors in the estimation are
invalid and corrective action should be taken. We tested for homoskedasticity using the
Pagan-Hall general test statistic. This statistic involves regression of the squared OLS
residuals on the independent variables (Wooldridge, 2002b). The null hypothesis, that
the disturbance is homoskedastic, was strongly rejected, revealing the presence of
(arbitrary) heteroskedasticity.
The problem of heteroskedasticity can partially be addressed through the use of
heteroskedasticity-consistent (or robust) standard errors and statistics. The
conventional IV estimator is consistent but not efficient in the presence of
heteroskedasticity (Baum et al., 2003). As a consequence, we used the two-step
feasible generalized method of moments (2-Step GMM) estimator to get efficient
estimates of the coefficients and consistent estimates of the standard errors (Baum,
2006; Baum et al., 2003). The generalized method of moments (GMM) estimator of the
coefficients in equation eq. 1 is based on the moment or orthogonality conditions. The
5 Homoskedasticity assumes that the error terms are independently and identically distributed
49

Chapter 4: Results and discussion
advantage of the GMM is that it does not require the distribution of the data to be fully
known, unlike the maximum likelihood estimation. The disadvantage of the use of GMM
is the poor small sample properties (Baum et al., 2003).
4.2.1 First Stage Results
The first stage of the 2SLS model is a simple OLS regression of equation (2), whereby
non-farm income is regressed on the control variables and the IV. The OLS estimates
with robust standard errors are reported in Table 4.9. The interpretation of these OLS
results must be done with much caution, because most literature analyzing the
determinants of non-farm income use more specific models like Heckman or Cragg
double hurdle model. These models try to deal with incidental truncation of potential
outcomes: non-farm income might not be observed because of the outcome of another
latent variable. Households only have non-farm income if they participate in the RNFE
sector. These models therefore consist of two parts: first the participation in non-farm
activities is analyzed, and second for those who participate in non-farm activities, the
level of non-farm income is analyzed. However, these models have been criticized
because non-farm income is an observed outcome: it is zero for households that do not
have access to non-farm activities (corner solution). Moreover, it is very difficult to
meet the exclusion restriction (Hertz, 2009). Our main interest lies in the second stage
of the 2SLS, and the first stage will only be used to indicate which factors may
influence non-farm income (and hence determine the access to non-farm activities)
and, more importantly, to evaluate the relevance of the instrumental variables.
As can be seen from Table 4.9, most control variables do not seem to significantly
influence non-farm income. However, the age of the household head seems to have a
significant and negative impact on non-farm income. This seems to be odd, as it is
expected that older people earn more income, but this might illustrate the
shortcomings of the OLS regression to determine non-farm income determinants.
Because we were not able to separate the effect of the control variables on the
participation and level of non-farm income, it is impossible to distinguish the effect of
the age of the household head on either participation in or the level of non-farm
income. It can however be assumed that younger household heads are more likely to
participate in non-farm activities, and therefore age has a negative impact on non-farm
income. Younger household heads might be less bounded to traditional agricultural
activities. This is in agreement with the results found by Woldenhanna and Oskam
(2001). The authors suggest population pressure to hinder younger farm households to
50

Chapter 4: Results and discussion
get enough land to support their livelihood and they are therefore obliged to participate
in non-farm activities. Non-farm activities also require more physical strength and
effort, which is more likely to be done by younger individuals. Besides this, older
household heads were historically prohibited to participate in non-farm activities,
making them inexperienced with non-farm income and more productive in farm
activities.
Table 4.9: First stage OLS regression results
OLS
Coef.
Std. Err.
fm1sex 0.3283 0.2759
fm1age - 0.0673 *** 0.0088
fm1schooled 0.0645 0.2328
hhlandholdsize - 0.0667 0.0487
livestock - 0.0667 * 0.0339
fixed - 0.0094 0.0844
tabiadismak 0.0073 0.0051
edirm 0.2662 0.2709
havei - 0.2565 0.2845
adults 0.3218 *** 0.0815
dependents 0.2201 *** 0.0751
non-farmshare_tabia 0.0674 *** 0.0121
constant 5.2844 0.8467
Joint significance of IV
F test of excluded instruments 20.11***
Angrist-Pischke multivariate F test of IV 20.11***
Underidentification test
Kleibergen-Paap rk LM statistic 35.41***
Weak identification test
Cragg-Donald Wald F statistic 19.39
Kleibergen-Paap Wald rk F statistic 20.11
Weak-instrument-robust inference
Anderson-Rubin Wald test 16.70***
Stock-Wright LM S statistic 15.70***
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
The effect of both gender and education is positive, but not significant however. It is
often hypothesized that an increase in educated level lowers the incentive for farm
activities and encourages participation in (more remunerative) non-farm activities
51

Chapter 4: Results and discussion
(Corral and Reardon, 2001; De Janvry and Sadoulet, 2003). Van den Berg and Kumbi
(2006) find that in Ethiopia, only formal and primary education significantly increase
non-farm income, while higher education seems to be irrelevant. We have tried to
incorporate different dummies for education levels, but none of them where statistically
significant. This could be explained by the fact that non-farm activities are dominated
by food for work and unskilled wage employment, which does not require much
education or skills. Our results do not strongly suggest a gender bias, which is in line
with the observations of van den Berg and Kumbi (2006). It seems as if the non-farm
sector does not significantly consist of jobs that have higher returns to education nor
does men have better access to non-farm activities.
Surprisingly, regional effects do not have a significant negative effect. We expected
households that live further from the capital city to be less engaged in non-farm
activities, because of fewer opportunities and more costs related with transport. Our
results indicate a positive effect, but the coefficient is statistically not different from
zero. Membership in social capital does increase non-farm income, but its impact is not
significant. Finally, the number of adult labor forces has a positive and very significant
impact on the level of non-farm income. As we expected, non-farm income is positively
related with the number of adults that are potential working forces in household. The
latter could indicate the availability of excess family labor that cannot be allocated on
the farm and households are therefore obligated to look for non-farm activities.
Moreover, Woldenhanna (2000) notes that larger family size decreases the marginal
value of households‟ consumption of leisure. Our finding is in accordance with the
results of Ruben and van den Berg (2001), Woldenhanna and Oskam (2001), Matsche
and Young (2004) and van den Berg and Kumbi (2006).
The animal assets of the households also have a negative and statistically significant
impact on non-farm income. Again, it looks like households with a higher value of their
livestock participate less in non-farm activities and hence have lower non-farm income.
Having more animals will probably indicate that households are highly engaged in farm
activities and livestock require intensive care (Ruben and van den Berg, 2001). The
impact of both the total plot size (as indicator of farm size) and the total value of the
fixed assets on the level of non-farm income is negative, but statistically not
significantly different from zero. Overall, shortage of households‟ wealth indicators can
push farm households into non-farm activities: farm households without land or with
small farms; and households with less fixed or animal assets seem to seek for
additional income by participating in non-farm income. Having more assets could
52

Chapter 4: Results and discussion
increase the returns to labor on the farm and work less off the farm. However, only the
number of livestock has a significant impact.
Another push factor could be the lack of access to irrigation, although not statistically
significant. Non-farm income might be important to set up an irrigation system.
Another explanation could be that, because access to irrigation decreases non-farm
income, households with irrigation systems spend more time on the farm. In
conclusion, our results do not provide an indication of the existence of entry barriers:
better educated or endowed households do not earn more non-farm income. This is
consistent with the findings of Ruben and van den Berg (2006) in Oromia, but in
contradiction with Block and Webb (2001) in Ethiopia and Woldenhanna and Oskam
(2002) in Tigray.
The requirement that instruments are correlated with the endogenous non-farm income
can easily be tested by examining the fit of the first stage regression. Both instruments
are positive and highly significant related with non-farm income, which suggests the
relevance of the instruments. The positive impact of non-farm work opportunities is
straightforward: having better access to non-farm activities makes it more likely to
participate in non-farm activities and hence increases non-farm income. The more
dependents in a household (hence a larger household) increases the non-farm income.
This confirms our hypothesis that households with more members that need to be
supported are more likely to participate in non-farm activities. As they are not potential
working forces and therefore cannot significantly increase farm productivity, the
household will be „pushed‟ to look for additional income sources. This finding is in line
with Block and Webb (2001) who find a positive relation between participation in nonfarm
activities and the dependency ratio. Woldenhanna (2000) finds that the number of
dependents affects non-farm work decisions because larger family size increases the
availability of labor, reduces the marginal utility of consumption and leisure.
The instruments passed the test for joint significance of the instruments, as both the F-
test for excluded instruments and the Angrist-Pischke multivariate F-test of excluded
instruments were highly significant. They were both higher than the critical values 6
developed by Stock-Yogo (2005) for single endogenous regressor. This indicates that
these test values are high enough to reject the hypothesis that the weak instrument
bias is greater than 10% of the magnitude of the endogeneity bias under OLS, at the 5
6 Stock-Yogo weak ID test critical values for single endogenous regressor at 10% maximal IV size is 19.93
53

Chapter 4: Results and discussion
% level of significance. The Kleibergen-Paap rk LM statistic was used as
underidentification test and is highly significant, indicating the relevance of our
instruments. Also the weak identification tests (Cragg-Donald Wald F-statistic and
Kleibergen-Paap Wald rk F-statistic) were conducted to test whether the estimator is
weakly identified: non-zero but small correlation between non-farm income and the two
instruments. These statistics were higher than the critical values given by the Stock-
Yogo weak ID test, revealing the strength of our instruments. Finally, the Weakinstrument-robust
inference test was used to test for the significance of the
endogenous regressors in the structural equation being estimated. The test was also
highly significant, rejecting the null hypothesis that the coefficients of the regressors in
the structural equation are jointly equal zero.
4.2.2 Second Stage Results
We began our empirical analysis with the impact of non-farm income on total farm
expenditures. Farm expenditures included all expenses on farm related activities. In the
last two sections, we expanded our analysis to the impact of non-farm income on two
types of investments, and the results will be discussed separately. For each regression
estimation, both OLS and 2SLS estimates are reported. Before we interpret the
regression estimates, an endogeneity test is performed to indicate whether it is
necessary to use an IV method or not. Using IV methods always comes with the price
of efficiency loss. We are however willing to pay this price if the OLS estimates are
inconsistent and biased. It is therefore useful to test the necessary of IV method and
the appropriateness of the OLS estimation (Baum et al., 2003). As the conditional
homoskedasticity is violated, we use the heteroskedasticity robust GMM distance
endogeneity test 7 . This test is a variant of the Durbin-Wu-Hausman test of the
endogeneity of regressors. Under the null hypothesis, non-farm income can actually be
treated as exogenous.
7 The C statistic is computed as the difference between two J statistics: one using the unrestricted regression
with the regressor being tested as endogenous (less efficient but consistent estimator) and one using the
restricted regression with the regressor treated as exogenous (more efficient estimator) (Baum et al 2003)
54

Chapter 4: Results and discussion
4.2.2.1 The impact of non-farm income on farm expenditures
Both OLS and 2SLS estimations of farm expenditures are reported in Table 4.10. Our
results show that we fail to reject the null hypothesis that non-farm income is
exogenous in the farm expenditure regression. As a consequence, it is not proven that
the OLS estimates will yield inconsistent estimates. Moreover, it implies that we do not
have a statistical-based argument to prefer the IV method over the OLS regression.
However, such endogeneity tests are restricted as they depend on the instruments
used for non-farm income. It is therefore difficult to conclude that non-farm income is
actually endogenous based on these kinds of endogeneity tests, and the test might only
serve as an indication of the presence of endogenous effects of non-farm income. We
prefer to work with the IV estimation as the endogenous effects of non-farm activities
have a strong theoretical background.
With the gmm option, the overidentifying condition can be tested by the Hansen J
statistic. The null hypothesis is that the instruments are jointly valid, conditional on at
least one being valid and that the excluded instruments are correctly excluded from the
estimated equation. Rejection implies that the instruments do not satisfy the
orthogonality assumption. This is plausible because either they are incorrectly excluded
from the regression or because they are not truly exogenous. We fail to reject the null
hypothesis (p=0.15) which implies that the validity of the used instrument is not
disproven. Interpretation of this statistic is however not straightforward. As Hertz
(2009) has pointed out, rejection of the null hypothesis could indicate that none, some
or either all of the instruments are valid.
Table 4.10 shows that non-farm activities have a positive and significant impact on
farm expenditures. Non-farm income increases the total expenditures households made
on farm related activities in both the OLS and IV estimation. After comparison of the
OLS and 2SLS models, we see that the coefficient of non-farm income in the OLS
estimation is five times lower than the coefficient of non-farm income in the IV
estimation (respectively 0.04 and 0.18). the IV method hence predicts that a
percentage increase in non-farm income increases farm expenditures with 0.18% and
this effect is significant at the one percent level. In the OLS estimation, the effect is
only 0.04%. We see that the OLS underestimates the effect of non-farm income on
farm expenditures. When potential factors that make non-farm income endogenous are
removed, the effect of non-farm income on farm expenditure is even greater. As a
result, failing to correct for the endogeneity of non-farm income underestimates the
impact of non-farm income on total farm investments indicating that the coefficient of
55

Chapter 4: Results and discussion
non-farm income in the OLS regression is downward biased. We assume that the
downward bias is caused by omitted variables.
Table 4.10: OLS and 2SLS estimations results of total farm expenditures
OLS
IVREG2
Coef. Std. Err. Coef. Std. Err.
non-farmincome 0.0365 * 0.0206 0.1792 * 0.0924
fm1sex 1.1756 *** 0.1817 1.1586 *** 0.1831
fm1age - 0.0043 0.0055 0.0049 0.0082
fm1schooled 0.1685 0.1212 0.1639 0.1260
hhlandholdsize 0.0546 *** 0.0197 0.0678 *** 0.0223
livestock 0.0541 *** 0.0161 0.0655 *** 0.0170
fixed 0.2103 *** 0.0552 0.1881 *** 0.0598
tabiadismak - 0.0046 * 0.0025 -0.0055** 0.0025
edirm 0.5918 *** 0.1227 0.5678 *** 0.1269
havei 0.2737 ** 0.1350 0.3082 ** 0.1478
adults 0.2621 *** 0.0437 0.2180 *** 0.0546
constant 3.8269 *** 0.5361 2.8614 *** 0.8395
Endogeneity test
Endogeneity test of endogenous regressors 2.53
Overidentification test
Hansen J statistic 2.09
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
We describe the effect of individual and household characteristics on the amount spend
on farm expenditures more detailed for investments in durables and farm input use
separately. Some first indications are given here. Of the household head characteristics
it is surprising that only gender has a significant impact on farm expenditures. Male
headed households are more likely to farm on a bigger scale and thus spend more
money on farm expenditures. Both the age and schooling of household does not seem
to have a significant impact on farm expenditures. All types of household wealth
indicators are positive and significant (at the one percent level) related with farm
expenditures. Having more agricultural and fixed assets increases the expenses on total
farm investments.
The distance to Mekelle is negative related with farm expenditures, indicating the
importance of being close located to markets and the capital city. However, this effect
56

Chapter 4: Results and discussion
is not so pronounced: every addition km that the household is located further from
Mekelle decreases farm expenditure with 0.6%. As we expected, households with
access to irrigation significant spend more money on farm expenditures than
households that do not have access to irrigation. This result shows that more inputs are
used and investments are required if households have an irrigation system on their
farm. Also membership in edir enhances the farm expenditures, which underlines the
importance of social capital. This effect is the strongest (except for gender): being
member in edir increases farm investments by 57% and this effect is highly significant
at the one percent level. Finally, the number of adult labor forces in the household has
a positive impact on farm expenditures and this effect is highly significant. When
households have more members that could participate in labor markets, they are able
to spend more on farm expenditures.
4.2.2.2 The impact of non-farm income on durables investments
We turn now to the effect of non-farm income on the money households spend on
durable investments. Table 4.11 reports the coefficients of the OLS and IV estimation of
non-farm income, individual and household characteristics. Our results show that the
null hypothesis of the endogeneity test is rejected at one percent significance level. As
a consequence, the OLS estimates will be incoherent. The endogenous effect of nonfarm
income on farm investment is hence meaningful and an IV method is required.
The overidentifying restriction is tested with the Hansen J statistic. We significantly fail
to reject the null hypothesis, and this effect is much stronger as before (p=0.991),
which proves the validity of the instruments. We thus have strong statistical evidence
that our instruments are meaningful and the 2SLS method will give the most
appropriate results.
The results indicate that non-farm income has a statistically significant, positive and
strong impact on farm investments in both regressions. The elasticity of non-farm
income in the OLS regression is 0.087 and 0.58 in the IV regression. The results of the
IV regression indicate that an increase in non-farm income by one percent increases
farm investments with 0.58%. Comparing both estimations, the statistical significance
remains the same but the coefficient of the IV estimation is more than six times the
OLS coefficient. When potential factors that make non-farm income endogenous are
removed, the effect of non-farm income on farm investments is even greater
57

Chapter 4: Results and discussion
Table 4.11: OLS and 2SLS estimations results of farm investments
OLS
IVREG2
Coef. Std. Err. Coef. Std. Err.
non-farmincome 0.0869 *** 0.0320 0.5819 *** 0.1686
fm1sex 1.0199 *** 0.2651 0.9109 *** 0.3069
fm1age 0.0068 0.0083 0.0372 *** 0.0141
fm1schooled 0.3621 * 0.2115 0.2833 0.2417
hhlandholdsize 0.0160 0.0355 0.0876 * 0.0469
livestock 0.0484 0.0296 0.0805 ** 0.0358
fixed 0.3146 *** 0.0860 0.2368 ** 0.1040
tabiadismak - 0.0160 *** 0.0047 - 0.0199 *** 0.0052
edirm 0.7910 *** 0.1904 0.7700 *** 0.2350
havei 0.2866 0.2089 0.4680 * 0.2701
adults 0.3330 *** 0.0704 0.1717 * 0.0990
constant 2.0796 *** 0.7573 - 1.2001 1.3541
Endogeneity test
Endogeneity test of endogenous regressors 11.52***
Overidentification test
Hansen J statistic 0.00
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
As expected, the age and gender of the household head have a positive impact on farm
investments and are significant at the one percent level. The gender effect is the
strongest of all effects: male headed households have larger, almost double, farm
investment than female headed households. It indicates that there is still a clear gender
bias. The effect of age is less strong: every additional year of life experience increases
farm investments with almost four percent. Hence, male and older household heads
positively affect the money spend on durable investments. More surprising is that
schooling does not have a significant impact on farm investments. We expected more
educated household heads to have higher expenditures on farm investments because
they are more literate. The coefficient of education in the OLS estimation might suggest
this, however the coefficient in the IV method is not statistically significant.
Households‟ wealth on the contrary has a clear positive impact on the total investments
households made on water and land, livestock, equipment and buildings. Every
additional Tsimidi of land a household owns, increases farm investments with 9 %.
Almost the same effect is reported for livestock and fixed assets, but their coefficients
are lower. A percentage increase in the total value of heifer, calves or camels owned
58

Chapter 4: Results and discussion
increases farm investments with 0.08%. For fixed assets, the effect is slightly stronger:
a percentage increase of the value of fixed assets increases farm investments with
0.24%. The household wealth indicators seem to have a positive effect on farm
investments and the effect of landholdings is the most pronounced. Also access of the
household to social capital has a positive impact on durable investments: being a
member of edir increases the expenditure on durables with 77%.
Similarly, the number of adult labor forces in the household increases farm
investments, suggesting that households with more members of working forces are
able to invest more in durables. Durable investments are increased by 17% for every
additional adult labor force. Access to irrigation has a comparable effect: having an
irrigation system increases the expenditure on durable investments with 47%. Finally,
the only variable that seems to have a negative impact on farm investments is the
distance to Mekelle. Every additional km increase in the distance from the households‟
residence to the capital city, decreases farm investments with two percent. Although
this effect is small, it is statistically highly significant. We expected that households that
live further away from Mekelle will have fewer opportunities to invest in durables.
4.2.2.3 The effect on different types of farm investments
Next to this, we broke farm investments up into its different parts: water and land,
livestock, equipment and buildings. This is done to see whether non-farm income has
more influence on a particular type of farm investment. We regressed non-farm income
and the control variables on each of the types of investments. The results of the
different IV estimations are reported in Table 4.12 and the results of the OLS
estimation are given in appendix 2. For both investments in livestock and equipment,
the results are similar with the results on total investments: non-farm income is
positively related with investments and this relation is significant at the one percent
level. The coefficient of non-farm income in the livestock and equipment regression is
however higher than the coefficient of non-farm income in the total farm expenditures
regression: 0.63 and 0.89 respectively. The effect of non-farm income on farm
investments is thus most pronounced in investments made in equipment and livestock.
Again, the OLS regression underestimates the effect of non-farm income on the two
types of investments, and this downward bias in OLS estimation is most probably
caused by the omitted variables bias.
59

Chapter 4: Results and discussion
Table 4.12: IV estimation results of investments in water&land, livestock, equipment and building
2SLS estimation coefficients
Water&Land Livestock Equipment Buildings
non-farmincome 0.1557 0.6326 *** 0.8921 *** 0.3539 **
fm1sex 0.5844 * - 0.0031 1.1762 *** 0.4714 *
fm1age 0.0098 0.0604 *** 0.0667 *** 0.0110
fm1schooled 0.1395 0.9964 *** - 0.0409 0.1217
hhlandholdsize 0.1630 *** 0.0002 0.0870 0.0330
Livestock 0.0086 0.1340 *** 0.0757 * 0.0201
Fixed - 0.1556 0.1911 0.1730 0.4631 ***
tabiadismak - 0.0241 *** - 0.0003 - 0.0197 *** - 0.0114 **
Edirm 0.3675 0.8191 ** - 0.1134 0.4794
Havei 0.2727 0.7220 * 0.5108 0.1489
Adults 0.3467 *** - 0.0089 - 0.0564 - 0.1774 *
constant 2.1850 - 4.9654 *** - 6.4511 *** - 3.4033 **
Endogeneity test 1.34 6.51 ** 41.02 *** 0.12
Overidentification 10.82 *** 0.21 0.41 2.17
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
Non-farm income has also a positive and significant impact (at the five percent level)
on investments in buildings. The endogeneity test is however unable to reject the null
hypothesis that non-farm income is exogenous and the IV method is hence not
preferred or even necessary. We discussed the weakness of such endogeneity tests
before, and assume that the model is valid. Finally, non-farm income has a positive, yet
insignificant effect on investments made in water and land. The model validity is
questioned as the endogeneity test cast doubts on non-farm income‟s endogeneity and
the overidentification test fails to prove the instruments validity. It seems as if the
number of dependents and the share of non-farm income are related with the
investments in water and land. Indeed, when both of them are separately included as
control variables, they significantly influence water and land investments.
4.2.2.4 The impact of non-farm income on farm inputs
Finally we estimated the impact of non-farm income on farm input use. We restricted
the farm input use to the value of fertilizer and improved seeds. This means that we
excluded local seeds (which however make up the biggest part) and labor use (which is
the smallest part). This distinction was made to estimate the impact of non-farm
income on modern farm inputs. The results of the estimation are reported in Table
60

Chapter 4: Results and discussion
4.13. The null hypothesis of the endogeneity test is strongly rejected, indicating the
necessity of use of the IV estimation. The test of the validity of the instruments
however performs dramatically. The overidentification test is strongly rejected which
cast doubts on the legitimacy of the instruments used. Indeed, when we regress both
instruments as part of the control variables they are highly significant in explaining the
value of farm inputs. As a result, we will focus on the OLS estimation.
Table 4.13: OLS and 2SLS estimations results of farm input use
OLS
IVREG2
Coef. Std. Err. Coef. Std. Err.
non-farmincome - 0.0535 ** 0.0269 -0.8939*** 0.1905
fm1sex 1.3215 *** 0.2424 1.4288 *** 0.3257
fm1age - 0.0092 0.0069 -0.0615*** 0.0151
fm1schooled 0.0602 0.1912 0.2297 0.2641
hhlandholdsize 0.2570 *** 0.0349 0.1375 ** 0.0597
livestock 0.1107 *** 0.0273 0.0514 0.0406
fixed - 0.0630 0.0657 0.0385 0.0999
tabiadismak - 0.0161 *** 0.0038 -0.0096* 0.0057
edirm 0.7076 *** 0.1946 0.7339 ** 0.2921
havei 0.8808 *** 0.1943 0.5569 * 0.3087
adults 0.2136 *** 0.0605 0.4915 *** 0.1091
constant 2.8783 *** 0.6175 8.7494 1.5262
Endogeneity test
Endogeneity test of endogenous regressors 34.06***
Overidentification test
Hansen J statistic 13.05***
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
Participation in non-farm activities seems to have a negative impact on the
expenditures made on fertilizer and improved seeds. This inverse effect is significant at
the five percentage level, but the estimated coefficient is only small. A percentage
increase in non-farm income decreases households‟ expenditures on farm inputs by
0.05%. Most of the control variables have the same effect on modern inputs
expenditure as they had on total expenditure and durable investments. The estimation
of effect of the sex and educational level of the household head, the landholding size,
the value of the livestock assets, the distance to Mekelle, the membership in edir, the
access to irrigation and the number of adult working forces in the household is similar
in comparison with the estimations in the farm expenditure and durable investment
61

Chapter 4: Results and discussion
regressions. Male headed households with bigger landholdings seem to spend more
money on farm inputs. The effect of the gender of the household is again the strongest:
being a male household head increases farm investments with 132%.
Moreover, if the household has access to irrigation or more labor forces, the level of
expenditures on fertilizer and improved seeds is increased. This effect is the strongest
for farm modern inputs: having an irrigation installation increases the expenditure on
farm input with 88%. Next to this, if the household is member of edir they will spend
more money on modern inputs. The latter effect has the same magnitude as before:
being a member of edir increases the purchase of modern inputs with 71%. Households
that live further away from Mekelle spend significantly less on modern farm inputs.
Finally, both the education of the household head and the value of fixed assets do not
have a significant impact on the expenditure on fertilizers and improved seeds. Their
estimated coefficient is not statistically significant different from zero.
4.2.2.5 The effect of different types of non-farm income
Finally, we broke non-farm income up into its components and study their effect
separately on farm investments. The rationale behind this distinction was that the two
types of non-farm activities differ in nature. Hence, we regressed equation eq. 1 but
with respectively wage and self-employment (business) income instead of total nonfarm
income. We used durable investments rather than total farm expenditure because
the model specification of the former are superior. Both OLS and IV estimation results
of these regressions with wage and self-employment are respectively given in Table
4.14 and Table 4.15. As we were unable to construct separate instruments for both
wage and self-employment income, we run the IV regression with the same
instruments used for non-farm income.
Appendix 1 reports the first stage results of the regression of the control variables and
instruments on both wage and self-employment income. It seems as if the number of
dependents and the share of non-farm income are good instruments for wage income,
as all the previous discussed tests perform even better. Table 4.14 shows that the
endogeneity test is rejected, indicating the requirement of an IV method. Wage income
positively influences farm investments and has an elasticity of 0.42. Hence, a
percentage increase in wage income increases farm investments with 0.42%. this effect
is significant at the one percent level. Similarly, the OLS regression underestimates the
62

Chapter 4: Results and discussion
impact of the wage income on farm investments: the coefficient of wage income is only
0.07.
Table 4.14: OLS and 2SLS estimations results of farm investments with wage income
OLS
IVREG2
Coef. Std. Err. Coef. Std. Err.
wageincome 0.0748 ** 0.0283 0.4233 *** 0.1167
fm1sex 1.0132 *** 0.2646 0.8958 *** 0.2906
fm1age 0.0066 0.0083 0.0304 ** 0.0121
fm1schooled 0.3720 * 0.2116 0.3557 0.2265
hhlandholdsize 0.0158 0.0356 0.0731 * 0.0413
livestock 0.0485 0.0296 0.0755 ** 0.0336
fixed 0.3230 *** 0.0857 0.2977 *** 0.0943
tabiadismak - 0.0161 *** 0.0048 - 0.0198 *** 0.0050
edirm 0.8214 *** 0.1928 0.9445 *** 0.2257
havei 0.2863 0.2097 0.4321 * 0.2501
adults 0.3332 *** 0.0707 0.2027 ** 0.0880
constant 2.1436 *** 0.7527 - 0.2387 1.0650
Endogeneity test
Endogeneity test of endogenous regressors 10.92***
Overidentification test
Hansen J statistic 0.03
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
Appendix 1 shows that for self-employment, the two instruments perform problematic,
as they fail for the overidentification, underidentification and weak instrument test. The
instruments used are not strong related with self-employment income. Also
Woldenhanna and Oskam (2001) find that the number of dependents affects wage
activities but not self employment activities. Moreover, according to Table 4.15, the
endogenous nature of self-employment income is questioned. Although the model is not
valid, it suggest that self-employment is negatively, but insignificant, related to farm
investments.
63

Chapter 4: Results and discussion
Table 4.15: OLS and 2SLS estimates results of farm investments with business income
OLS
IVREG2
Coef. Std. Err. Coef. Std. Err.
businessincome 0.0188 0.0329 - 1.2242 1.1183
fm1sex 1.0495 *** 0.2657 0.2941 0.7315
fm1age 0.0018 0.0082 - 0.0157 0.0221
fm1schooled 0.3790 * 0.2129 0.2262 0.4099
hhlandholdsize 0.0036 0.0353 - 0.0001 0.0624
livestock 0.0429 0.0298 0.0423 0.0534
fixed 0.3241 *** 0.0852 0.6044 ** 0.2791
tabiadismak - 0.0151 *** 0.0048 - 0.0232 ** 0.0109
edirm 0.7873 *** 0.1920 1.2957 ** 0.5994
havei 0.2580 0.2081 0.0365 0.4567
adults 0.3615 *** 0.0692 0.3088 ** 0.1220
constant 2.6227 *** 0.7389 4.7342 2.2847
Endogeneity test
Endogeneity test of endogenous regressors 2.19
Overidentification test
Hansen J statistic 4.42**
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
4.2.3 Discussion
4.2.3.1 The impact of non-farm income
Our results indicate a positive relation between non-farm activities and both farm
expenditures and investments. This means that non-farm income increases the amount
of money spent on farm expenditures and in particular on investments. Hence, our
results provide evidence that participating in non-farm activities has a positive impact
on agricultural activities: households are able to increase their expenditures and
investments in farm activities. Our results are in line with the empirical evidence of the
literature on investment linkages (Ruben and van den Berg, 2001; Anriquez and
Daidone, 2009; Maertens, 2009; Stampini and Davis, 2009).
The effect of non-farm income is the strongest on durable investments. We have
distinguished the impact of non-farm income on the different types of investments and
our results indicated that the impact of RNFE is particularly important for livestock and
equipment. The former result might be explained by the findings of Pender and
64

Chapter 4: Results and discussion
Gebremedhin (2004) in Tigray. The authors find that investments in livestock increase
crop productivity and households‟ income. Hence, it can be assumed that non-farm
income would be invested in farm activities with the highest payoff such as livestock.
The strong impact of non-farm income on equipment indicates the RNFE can support
the purchase of new or improved tools, replacement of old materials and the adoption
of modern techniques. This makes sense, as Woldenhanna (2002) finds that the
dominant type of farm input in Tigray is traditional farming technology: simple hand
tools and oxen-driven implements. Finally, the RNFE also has a strong impact on
investments made in buildings. This result might suggest that non-farm income is seen
as an income surplus which can be invested in important, though not vital, investments
in the farm. Other sources of cash might be used to finance daily consumption needs.
Non-farm income did not have a significant effect on investments made in water and
land. This is a somewhat surprising result. Pender et al. (2006) note that the most
common investments in Tigray are stone terraces and soil bunds. These investments
have been widely promoted through food for work programs and community labor or
private incentives for soil and water conservation. Pender et al. (2006) and Pender
(2004) find that investments in stone terraces increases crop production, because they
help households to overcome soil moisture deficiency. The average rate of return to
stone terraces ranges between 34% (Pender and Gebremedhin, 2004) and 46%
(Pender et al., 2006). Hence, it is not clear why non-farm income does not increase
these investments, as it is found that these investments have a high payoff.
Although the RNFE had a stimulating effect on investments in durables, the impact of
non-farm income on farm input use was inverse. Households with more non-farm
income spend significantly less money on modern farm input use. Our results are in line
with other literature in Ethiopia that suggest or find the same relation. Both Pender
(2004) and Holden et al. (2004) find a negative relation between non-farm activities
and fertilizer use in Ethiopia. This outcome is however in contradiction with the results
found by Woldenhanna (2000). He finds the use of variable farm inputs to be highly
and statistically significant influenced by non-farm income. His results show that
increase in non-farm income by 1% increases expenditure on input use by 0.43%.
How can it be explained that non-farm income decreases expenditures on farm input?
The profitability of modern inputs in low agricultural areas is limited (Pender, 2004;
Ehui and Pender, 2005). Pender et al. (2006) find that on average, fertilizer use is
unprofitable in Tigray because the use of fertilizer has only a marginal impact on crop
production. This yield increase is insufficient to cover the average costs of fertilizer.
65

Chapter 4: Results and discussion
This explains the abhorrent attitude of farmers towards the adoption of modern inputs
despite the governments‟ efforts to promote its use. Our descriptive statistics show that
the use of fertilizer and improved seeds was restricted, and its value was minor in
comparison with investments in durables (Table 4.1). Moreover, the purchase of
modern inputs is closely bound with credit obtained from extension programs
(Woldenhanna, 2002). Fertilizer is hence supplied on credit basis, and households
participating in extension programs are less likely to spend their non-farm income on
modern farm inputs (Kidane, undated).
Not only are modern inputs unprofitable, Kidane (undated) suggests the net benefit
cost ratio for farm input use to be low in comparison to alternative investment lines. It
is hence more profitable to invest in livestock for example. Also Dercon and
Christiaensen (2010) find that the main raison for not using modern inputs is the high
price. Input use is expensive and real fertilizer prices have increased. Moreover, the
availability of credit for inputs is widespread because of parastatal structures of
fertilizer provision, but the high interest rates are problematic. This has decreased the
relative output-fertilizer price ratio. This trend is especially problematic when harvests
are poor, for example because of poor weather conditions. The sunk cost of fertilizer
will hence decrease the returns, which will be lower than the returns if fertilizer was not
used. The use of fertilizer is thus a risky activity with moderately higher returns
compared to not using fertilizer (Dercon and Christiaensen, 2010).
In the majority of regressions, we were unable to reject the endogenous nature of nonfarm
income. Hence, the use of an IV method was necessary and superior to the OLS
estimation. Overall, the two variables used to instrument non-farm income have proved
to be appropriate instruments. In most regressions, the under-, over- and weakidentification
tests have proven the validity and relevance of the instruments. However,
both instruments were positively related with farm input use and water and land
investments. We were aware of the possible relation between the number of
dependents and farm decisions. However, the instrument pass all the tests in other
models, so we can assume that our instruments are not too weak in general.
In the expenditure and investment regressions, the OLS estimates underestimate the
coefficient of non-farm income in comparison with the IV estimation. Hence, the impact
of non-farm income is stronger if we eliminate the endogenous effects of non-farm
income, causing the OLS estimation to be biased downward. Such a bias was expected
and can be explained by several possible omitted variables. Various factors might be at
stake, and because of the latter, the direction of the bias was not clear a priori.
66

Chapter 4: Results and discussion
However, as our hypothesis that households use income from RNFE to invest in farm
activities is likely to be confirmed, this could indicate that the omitted variable bias is
highly related with credit constraints. As we were not able to gather information and
hence control for the latter, it is possible that they have a strong correlation with the
omitted variables.
Our results hence suggest that households participate in non-farm activities in order to
earn additional cash to overcome credit or liquidity constraints. This additional income
gives farm households the opportunity to invest in their farm activities. Similarly,
Stampini and Davies (2009) and Lee and Sadawa (2007) find that in the context of
liquidity constraints, omitting variables associated with these credit constraints causes
a downward bias in the regression result. The direction of the bias is the result of the
product of the signs of, on the one hand the covariance between non-farm income and
the omitted variable and on the other hand the covariance between the omitted
variable and farm investments. As Stampini and Davies (2009) point out, the former is
positive (households react to credit constraints by engaging more in non-farm
activities) and the latter is negative (credit constraints hinder households to spend
money on farm investments). Lee and Sadawa (2007) find that the estimation of
precautionary savings suffers from omitted variable bias when a large proportion of the
households is credit constrained. This results in a downward bias: correcting for the
omitted variable bias increases the regression coefficient of the estimated prudence.
Finally, we have also distinguished the impact of the two types of non-farm activities on
farm investment and found a distinct effect. The difference in impact of wage and selfemployment
income is however not surprising. Wage employment is the dominant type
of non-farm activity and therefore has the biggest contribution to the non-farm income
and hence total household income. Hence, it dominates the impact of non-farm income
on farm investments. Wage jobs are easily accessible and important for households
during and outside farm-intense periods. Self-employment income is only a minor
component of the households‟ total income, as can be seen in Table 4.2. This explains
why the impact of non-farm income and wage income on farm investments is highly
similar, whereas self-employment has a small (and even insignificant) impact.
The different effect of wage and self-employment income on investments can be
explained by the inherent differences between the two types of non-farm activities.
Woldenhanna (2000) observes that households in Tigray participate in off-farm wage
employment because of push factors and self-employment because of pull factors. The
latter implies that non-farm wage activities can be considered as residual employment
67

Chapter 4: Results and discussion
that absorbs family labor which cannot be fully employed on the farm. Self employment
activities are undertaken by households to reap the attractive returns. Wage and selfemployment
activities also require different input requirements. Self employment
generally requires more capital and managerial skills than wage employment so
undertaking self-employment might be more difficult than wage employment. If the
access to credit is constrained and own capital is restricted, farmers will prefer nonfarm
activities that require less initial capital (Woldenhanna, 2000).
Wage employment might also be a safer and more direct source of liquidity and plays a
stronger role in inducing spendings (Oseni and Winters, 2009). Moreover, Hertz (2009)
notes that wage income is an important cash-on-hand, especially if there are no
alternative opportunities for saving, or if there is a big difference between the interest
rates on savings versus loans. It will therefore have a stronger role in determining
households‟ investment expenditures. It is more likely that households will participate
in wage activities to overcome liquidity constraints. According to Hertz (2009), it makes
sense that cheaper forms of finance would be used for farming if the coverage of
current bills is mandatory and agricultural expenditure are optional and risky. Selfemployment
jobs require investments, which compete with farm investments. Liquidity
constraints hamper the entry to these jobs, hence these activities suffer from liquidity
constraints rather than solving it. Only households with a good asset position may face
relatively less credit constraints and hence prefer to work in (more remunerative) selfemployment
activities (Woldenhanna, 2000). We therefore observe a negative,
although not significant impact of self-employment income on farm investments.
4.2.3.2 The effect of control variables
The impact of the control variables on the amount of money spent by the households
on expenditures, seems to be similar for total expenditures, durable investments and
farm input use. Table 4.16 gives an overview of the effect of the different individual and
household characteristics that have a statistically significant effect on total farm
expenditures, durable investments and input use. The sex of the household head is an
important factor that determines all tree expenditures. Male household heads invest
substantially and significantly more than their female counterparts. Oseni and Winter
(2009) find the same relationship and suggest that females are more likely to farm on a
smaller scale and hence spend less on agricultural inputs. However, as the regression
did not incorporate the amount of male and female household members, our results
does not say anything about the gender bias within households. Other studies, like
68

Chapter 4: Results and discussion
Maertens (2009) find evidence of a clear gender bias within the household: female
household members work on family plots while male household members control the
land and make farm decisions.
Table 4.16: Summary of the impact of non-farm income and control variables
Total expenditures Durable investments Input use
non-farmincome + + -
fm1sex + + +
fm1age 0 + 0
fm1schooled 0 0 0
hhlandholdsize + + +
livestock + + +
fixed + + 0
tabiadismak - - -
edirm + + +
havei + + +
adults + + +
Notes: +: positive impact, -: negative impact and 0 : no statistical significant impact
The age of the household head increases the expenditures on durable investments.
Older individuals tend to have more agricultural experience and means to make
investments than then their younger counterparts. It is however more likely that there
is a convex relationship at work: investments increase with age until a certain threshold
is reached and decrease afterwards (the life cycle effect, as noted in Woldenhanna,
2000). Kidane (undated) suggests that older people spend less of the non-farm income.
Due to the possible risk of high multicollinearity we have chosen not to incorporate the
squared term of the age of the household head in our regression. The insignificant
relationship between input use and the age of the household head is rather surprising.
We expected more educated household heads to have higher expenditures on farm
investments because they are more literate. Our results suggest that education has a
positive effect on all types of investments, and this is in line with most empirical
evidence. These studies usually find a positive effect on investments but at a
decreasing rate (Kilic et al., 2009). However, the effect of education is not statistical
significant for any type of investment. This is consistent with the effect of education on
fertilizer expenditure found by Kidane (undated). He suggests that education does not
have the power to expand fertilizer use because the average education attainment in
the Geba catchment is low (below first grade level).
69

Chapter 4: Results and discussion
The total size of landholdings and the value of the livestock owned by the household
both increase farm investments of any kind. The effect of landholdings is much stronger
than that of livestock. Our results suggest that landless farmers spend less on
agricultural inputs than farmers who own their cultivated land. Oseni and Winters
(2009) note that this is not surprising because landless households usually rent land or
sharecrop, while landlords are responsible for the expenditures of investments. Larger
farms may have the advantage of economies of scale and their use of modern inputs is
more likely to be profitable. Households‟ value of fixed assets on the one hand
increases investments in durables, while on the other hand it decreases input use. The
latter is however statistically not significant. This confirms that households with more
land and assets spend more money on farm related activities. This could indicate that
more endowed farmers are more active in farm activities and therefore make more
investments in their farm. Also Kidane (Undatedb) suggests that farmers with more
assets have more incentives to use modern inputs or adopt new technologies.
Households‟ wealth can help farmers to afford, bear or spread risks associated with
adoption and larger farms have more opportunities to adopt new farming practices.
Moreover, the larger the land size, the more beneficial fertilizer and input use becomes.
Besides these household wealth indicators, social capital has a strong and highly
significant impact on all types of farm investments. Social networks such as edir often
offer relevant and qualitative information to its members, making them more likely to
invest in agricultural activities. Kilic et al. (2009) suggest social capital enhances farm
investments because it reduces risk and increases access to capital when reciprocal
relations are important. The distance to Mekelle has the most pronounced negative
impact on the amount of money spent on all types of investments. It is believed that
rural areas close to urban centres have greater farm/non-farm linkages (Davis et al.,
2002). Our results confirm that farmers that live near Mekelle have superior access to
markets and are therefore in a better position to invest in their farm. Farmers they live
further away from the regional capital are discouraged to use modern agricultural
inputs because of the high transport cost (Kidane, undated). Having access to an
irrigation system also enhances the expenditures on investments of any kind, indicating
that irrigated land requires more inputs and is cultivated more intensively. This result
shows that irrigation systems promote investments and the use of farm inputs. Finally,
the number of adult labor forces has a positive impact on farm expenditure. This
suggests that households with more contributing members of working forces are able to
invest more in their farm activities.
70

Chapter 5: Conclusions and Recommendations
5 CONCLUSIONS AND RECOMMENDATIONS
5.1 Conclusion
In Ethiopia, rural development drives the agricultural sector in stimulating the rural
economy. However, environmental problems, low agricultural productivity, inadequacy
of previous policies and the emergence of the livelihood concept necessitate a wider
and more synergistic development path. In this context, the development of the RNFE
sector was inevitable. Nowadays the non-farm sector is considered to be a significant
component of the rural economy. As the RNFE and agricultural sector become equally
important in a synergistic development approach, it is essential to study how these two
sectors are inter-linked. The aim of this study was therefore to explore the impact of
RNFE on agricultural production, focusing on a possible complementary relation
between non-farm income and farm investments.
Recent development literature has hypothesized the existence of investment linkages.
Investment linkages imply that farmers use non-farm income to finance farm activities
which will result in an increase of the expenditures on farm investments and input use
and enable farm modernization. These investment linkages are especially important for
credit constrained households, since these are restricted from spend credit on farm
investments. Non-farm income might be used to overcome these credit constraints.
However, empirical evidence does not unambiguously confirm that non-farm income
complements farm activities. It is studied that non-farm activities sometimes withdraw
resources from the farm and hence compete with agricultural production. The primary
objective of this study was therefore to determine whether non-farm income enhances
farm expenditures through investment linkages.
In addition, the importance of the RNFE sector for rural households in a developing
country such as Ethiopia was of interest. Simple descriptive statistics confirm that nonfarm
income constitutes a substantial part of the total household income and access
seems not to be restricted. 80% of the households had at least some non-farm income
and non-farm income on average accounts for 27% of the total household income. We
found no indication for the existence of entry barriers which hamper the entrance into
non-farm activities. Participation in non-farm activities is mainly determined by push
factors: dependent household members that need to be supported, excess labor forces
and low involvement in livestock activities. However, credit constraints are assumed to
71

Chapter 5: Conclusions and Recommendations
be one of the most (or even the most) important push factor. We were unfortunately
not able to incorporate this in our regression model.
The lack of information about credit constraints is suggested to be the major source of
unobserved heterogeneity (omitted variable bias), although several other factors might
be at stake. Omitted variable bias causes non-farm income to be endogenously related
with farm investments. As a consequence a bias in the OLS regression occurs.
Moreover, decisions about RNFE and farm activities might be taken jointly, creating a
simultaneity bias between farm investments and non-farm activities. Using IV methods,
we were able to deal with both biases. Comparison of the IV and OLS estimations
indicated that the OLS estimates of the coefficient of non-farm income are biased
downwards, and this due to omission of liquidity constraints variables We suggested
that unobserved credit constraints hamper households to invest on their farm, but push
households to seek for additional, non-farm income sources.
Using household survey data from Tigray province, we provided evidence for the
investment linkage hypothesis. The regression results proved that RNFE has a
substantial and positive impact on farm expenditures and investments. A percentage
increase in non-farm income increases farm expenditures and investments with
respectively 0.18% and 0.58%. Moreover, these investments seem to concentrate
mainly on livestock and equipment. These results indicate that non-farm income is an
important source to finance investments that have the highest payoff (and hence
impact) on farm activities. It is assumed that this relationship is mainly driven by farm
liquidity constraints, forcing households to look for additional credit sources.
Households participate in the RNFE to gain a surplus income that can be used to invest
in farm activities, in which they would be unable to engage otherwise.
One seemingly surprising result was that non-farm income does not increase farm input
use and hence does not fit in with the evidence of investment linkages. We suggested
that this apparent contradiction is the result of local conditions. While extension and
credit programs have increased the access of farmers to inputs (especially fertilizers),
the efficiency (and profitability) of modern inputs is low because of a decreased
output/fertilizer price ratio, higher rate of return to other investments or vulnerability to
inadequate soil moisture. Hence, households prefer to spend the credit that is obtained
by participating in non-farm activities on more profitable investments such as livestock
or equipment. The latter are the two most important types of traditional technology
used by farmers in Tigray.
72

Chapter 5: Conclusions and Recommendations
The regression analysis also indicated the importance of dividing non-farm income into
its distinct sources. In our analysis, non-farm activities include both wage and selfemployment
activities. Wage income is the most important non-farm activity in Tigray.
Wage jobs are mainly temporary and do not require specific education or training. They
are therefore accessible to many rural farm households and form a direct source of
cash. On the contrary, self-employment activities only have a minor share in the total
household income and it is assumed that some initial investments are required. As a
result, we noticed that wage income increases farm investments and expenditures,
while self-employment did not have a significant impact on farm activities.
Overall, the credit constraints farmers face can be superseded by participation in the
RNFE sector. This implies that farm/non-farm linkages are present in the development
of the RNFE, and should therefore play an important role in the rural development
approach. The suggested virtuous circle is therefore possible: promoting access to
RNFE increases farm modernization through investment linkages. Better access and
availability of farm households to non-farm activities increases farm investments and
expenditures. This in turn increases demand for non-farm activities, especially nonfarm
wage labor. The next section summarizes policy recommendations, suggested
from our analysis, that should contribute to a more synergistic rural development path.
5.2 Policy Recommendations
The analysis emphasizes the importance of non-farm income in rural areas of
Ethiopia. Not only is it a substantial share of the total household income, it is
used to overcome liquidity constraints. It seems therefore highly recommended
to incorporate RNFE in rural development, as it can be assumed as an
alternative path to overcome poverty.
Non-farm activities should be wider available and wage rates should improve in
order to increase the incentive and capacity of households to participate in nonfarm
activities. Especially wage employment has a strong impact on farm
investments. Policies should encourage the growth of the RNFE and strengthen
the link between the different activities and sectors.
As no indication of entry barriers to participate in the RNFE exist, it is not
necessary to promote policies to overcome entry restrictions. Rather, the
profitability of RNFE activities should be improved. Providing better training and
73

Chapter 5: Conclusions and Recommendations
education, encouraging small-scale enterprises, improving access to markets
and improving public infrastructure will stimulate growth in the RNFE sector and
enable households to take advantage of the opportunities that already exist.
In addition, our results show that farm and non-farm activities are inter-linked:
non-farm income can be used to support agricultural production via investment
linkages. Non-farm and farm activities cannot be seen in isolation. This implies
that both the RNFE and farm sector should be simultaneously supported in rural
policies in order to improve durable rural development and stimulate farm
modernization. The link between both sectors should be reinforced by
complementary programs and policies. Hence, a synergistic development path
seems to be appropriate in Ethiopia, where the agricultural sector is
omnipresent.
Policies should incorporate evidence of investment linkages in order to target
credit constrained households. Liquidity restrictions in the agricultural sector
hamper farm productivity but can be surmounted by additional sources of credit
and liquidity. Systems should be set up to relax the credit constraints of poor
households, besides promotion of non-farm activities. Development of rural
credit markets or microfinance will help households to improve their farm
activities.
RNFE has a positive impact mainly on farm investments, yet the inverse relation
is observed for modern input use. While previous extension systems focused
primarily on the distribution of fertilizers and modern inputs, it did not seem
successful. Therefore, the Ethiopian government should embrace both farm and
non-farm activities and emphasize on the development of non-farm activities
because these foster investments in livestock, buildings and equipment.
Overall, a virtuous circle is possible and should therefore be promoted. However,
more empirical evidence of such self-enforcing effects should be provided by
means of further studies. It is thereby suggested to conduct further research.
74

References
REFERENCES
Andreosso - O‟Callaghan, B. & Yue, G. (2004). Intersectoral linkages and key sectors in
China, 1987 - 1997. Asian Economic Journal, 18: 165-183.
Anriquez, G. & Daidone, S. (2010). Linkages between the farm and nonfarm sectors at
the household level in rural Ghana: a consistent stochastic distance function
approach. Agricultural Economic, 41: 51–66.
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: 315–331.
Baum, C.F. (2006). An introduction to modern econometrics using Stata. College
Station, Stata Press Publication. Texas. 341p.
Baum, C.F., Schaffer, M.E. & Stillman, S. (2003). Instrumental variables and GMM:
estimation and testing. Stata Journal, 3: 1-31.
Belay, K. & Abebaw, D. (2004). Challenges facing agricultural extension agents: a case
study from south-western Ethiopia. African Development Review, 16: 139-168.
Block, S. & Webb, P. (2001). The dynamics of livelihood diversification in post-famine
Ethiopia. Food Policy, 26: 333-350.
Blunch, N.H. & Verner, D. (2006). Shared sectoral growth versus the dual economy
model: Evidence from Cote d'Ivoire, Ghana and Zimbabwe. African Development
Review, 18: 283-308.
Carswell, G. (2002). Livelihood diversification: increasing in importance or increasingly
recognized? Evidence from southern Ethiopia. Journal of International Development,
14: 789-804.
Chang, H.H. & Wen, F.I. (2010). Off-farm work, technical efficiency, and rice production
risk in Taiwan. Agricultural Economics, 42: 269-278.
Corral, L. & Reardon, T. (2001). Rural non-farm incomes in Nicaragua. World
Development, 29: 427-442.
75

References
Davis, B., Reardon, T., Stamoulis, K. & Winters, P. (2002). Promoting farm/nonfarm
linkages for rural development: Case studies from Africa and Latin America. FAO,
Rome. 196p.
Davies, B., Winter, P., Reardon, T. & Stamoulis, K. (2009). Rural nonfarm employment
and farming: household-level linkages. Agricultural Economics, 40: 119-123.
De Janvry, A. & Sadoulet, E. (2003). Income strategies among rural households in
Mexico: the role of off-farm activities. World Development, 29: 467-480.
De Janvry, A., Sadoulet, E. & Zhu, N. (2005). The role of non-farm incomes in reducing
rural poverty and inequality in China. CUDARE working paper 1001. University of
California, Berkeley. 1-29.
Dercon, S. (1998). Wealth, risk and activity choice: cattle in Western Tanzania. Journal
of Development Economics, 55: 1-42.
Dercon, S. & Christiaensen, L. (2010). Consumption risk, technology adoption and
poverty traps: Evidence from Ethiopia. Journal of Development Economics, Article in
press.
Dercon, S. & Krishnan, P. (1996). Income portfolios in rural Ethiopia and Tanzania:
choices and constraints. The Journal of Development Studies, 32: 850-875.
Egziabher, K.G. (undated). Are non-farm income relaxing farm investment liquidity
constraints for marginal farms? A double hurdle approach. Chapter in Ph.D. thesis.
K.U.Leuven, Belgium. Forthcoming
Ehui, S. & Pender, J. (2005). Resource degradation, low agricultural productivity and
poverty in sub-Saharan Africa: pathways out of the spiral. Agricultural Economics,
32: 225-242.
Ellis, F. (1998). Household strategies and rural livelihood diversification. Journal of
Development Studies, 35: 1-38.
Ellis, F. (2000). Rural livelihoods and diversity in developing countries. Oxford
University Press, Oxford. 279p.
Ellis, F. (2000b). The determinants of rural livelihood diversification in developing
countries. Journal of Agricultural Economics, 51: 289-302.
76

References
Ellis, F. & Freeman, H.A. (2004). Rural livelihoods and poverty reduction strategies in
four African countries. Journal of Development Studies, 40: 1-30.
Escobal, J. (2001). The determinants of nonfarm income diversification in rural Peru.
World Development, 29: 497-508.
Fernandez-Cornejo, J., Mishra, A.K., Nehriung, R.F., Hencricks, C., Southern, M.,
Gregory, A. (2007). Off-farm income, technology adoption and farm economic
performance. United States Department of Agriculture Economic Research Report,
36: 1-53.
Goodwin, B.K. & Mishrea, A.K. (2004). Farming efficiency and the determinants of
multiple job holding by farm operators. American Journal of Agricultural Economics,
86: 722-729.
Haggblade, S., Hazell, P. & Brown, J. (1989). Farm non-farm linkages in rural sub-
Saharan Africa. World Development, 17: 1173-1201.
Hazell, P.B. & Hojjati, B. (1995). Farm non-farm growth linkages in Zambia. Journal Of
African Economies, 4: 406-435.
Hertz, T. (2009). The effect of nonfarm income on investment in Bulgarian family
farming. Agricultural Economics, 40: 161-176.
Holden, S., Shiferaw, B. & Pender, J. (2004). Non-farm income, household welfare, and
sustainable land management in a less-favored area in the Ethiopian highlands. Food
Policy, 29: 369-392.
Huang, J., Wu, Y. & Rozelle, S. (2009). Moving off the farm and intensifying agricultural
production in Shandong: A case study of rural labor market Linkages in China.
Agricultural Economics, 40: 203–218.
Islam, N. (1997). The non-farm sector and rural development: review of issues and
evidence. Food, Agriculture and the Environment Discussion Paper, 22. IFPRI,
Washington D.C. 1-48.
Jayne, T.S., Yamano, Y., Weber, M.T., Tschirley, D., Benfica, R., Chapoto, A. & Zulu, B.
(2003). Smallholder income and land distribution in Africa: implications for poverty
reduction strategies. Food Policy, 28: 253-275.
77

References
Kilic, T., Carletto, C., Miluka, J. & Savastano, S. (2009). Rural nonfarm income and its
impact on agriculture: Evidence from Albania. Agricultural Economics, 40: 139-160.
Lanjouw, J.O. & Lanjouw, P. (2001). The rural non‐farm sector: issues and evidence
from developing countries. Agricultural Economics, 26: 1-23.
Lanjouw, P., Quizon, J. & Sparrow, R. (2001). Non-agricultural earnings in peri-urban
areas of Tanzania: evidence from household survey data. Food Policy, 26: 385-403.
Lee, J.J. & Sawada, Y. (2010). Precautionary saving under liquidity constraints:
Evidence from rural Pakistan. Journal of Development Economics, 91: 77–86.
Lien, G., Kumbhakar, S.C. & Hardaker, J.B. (2010). Determinants of off-farm work and
its effects on farm performance: the case of Norwegian grain farmers. Agricultural
Economics, 41: 577-586.
Maertens, M. (2009). Horticulture exports, agro-industrialization and farm-nonfarm
linkages with the smallholder farm sector: evidence from Senegal. Agricultural
Economics, 40: 219-229.
Negash, Z.Z. (2008). Microfinance loan delivery, utilization and impact: with specific
reference to Tigray, northern Ethiopia. Ph.D. thesis. K.U.Leuven, Belgium. 190p.
Matshe, I. & Young, T. (2004). Off-farm labour allocation decisions in small-scale rural
households in Zimbabwe. Agricultural Economics, 30: 175-186.
OECD (2010), "Ethiopia", in OECD/African Development Bank/United Nations Economic
Commission for Africa, African Economic Outlook 2010, OECD Publishing. 281p.
Oseni, G. & Winters, P. (2009). Rural nonfarm activities and agricultural crop
production in Nigeria. Agricultural Economics, 40: 198-201.
Pender, J. (2004). Development pathways for hillsides and highlands: some lessons
from Central America and East Africa. Food Policy, 29: 339-367.
Pender, J., Place, F. & Ehui, S. (2006). Strategies for sustainable land management in
the East African highlands. International Food Policy Research Institute, Washington,
D.C. 483p.
78

References
Pender, J. & Gebremedhin, B. (2004). Impact of policies and technologies in dryland
agriculture: evidence from Northern Ethiopia. In: challenges and stategies for
dryland agriculture. S.C. Rao and J. Ryan (eds.). American Society of Agronomy and
Crop Science of America (CSSA) special Publication, 32. Madison, W.I. 434p.
Pfeiffer, L., Lopez-Feldman, A. & Taylor, J.E. (2009). Is off-farm income reforming the
farm? Evidence of Mexico. Agricultural Economics, 40: 125-138.
Phimister, E. & Roberts, D. (2006). The effect of off-farm work on the intensity of
agricultural production. Environmental & Resource Economics, 34: 493-515.
Reardon, T. (1997). Using evidence of household income diversification to inform study
of the rural nonfarm labor market in Africa. World Development, 25: 735-747.
Reardon, T., Berdegué, J. & Escobal, G. (2001). Rural nonfarm employment and
incomes in Latin America: Overview and policy implications. World Development, 29:
395-409.
Reardon, T., Berdegué, J., Barrett, C.B. & Stamoulis, K. (2006). Household income
diversification into rural nonfarm activities. In: Transforming the Rural Nonfarm
Economy. Haggblade, S., Hazell, P., Reardon, T. (eds.). John Hopkins University
Press, Baltimore, M.D., 115-141.
Reardon, T., Crawford, E. & Kelly, V. (1994). Links between nonfarm income and farm
investments in African households: Adding the capital market perspective. American
Journal of Agricultural Economics, 74: 1172-1176.
Reardon, T., Stamoulis, K., Cruz, M.E., Balisacan, A., Berdegué, T.A. & Banks, D.
(1998). Rural non-farm income in developing countries. Special chapter in The State
of Food and Agriculture 1998. FAO, Rome. 283–356.
Reardon, T., Tayler, E.J., Stamoulis, K., Lanjouw, P. & Balisacan, A. (2000). Effects of
non-farm employment on rural inequality in developing countries: an investment
perspective. Journal of Agricultural Economics, 51: 266-288.
Ruben, R. & van den Berg, M. (2001). Nonfarm employment and poverty alleviation of
rural farm households in Honduras. World Development, 26: 549-560.
Shimelis, A. & Bogale, A. (2007). Rural development policies and programs strategies
among rural households in Dire Dawa, eastern Ethiopia. Tropical Science, 47: 73-80.
79

References
Stampini, M. & Davis, B. (2009). Does nonagricultural labor relax farmers‟ credit
constraints? Evidence from longitudinal data for Vietnam. Agricultural Economics,
40: 177–188.
Sumberg, J., Gilbert, E. & Blackie, M. (2004). Income diversity, technology choice and
agricultural research policy in Sub-Saharan Africa. Development Policy Review, 22:
131-146.
Tesfay, G.B. (2009). Econometric analyses of microfinance credit group formation,
contractual risks and welfare impacts in Northern Ethiopia. Ph.D. thesis. Wageningen
University, Holland. 146p.
van den Berg, M., & Kumbi, G. E. (2006). Poverty and the rural nonfarm economy in
Oromia, Ethiopia. Agricultural Economics, 35: 469-475.
Woldenhanna, T. (2000). Economic analysis and policy implications of farm and offfarm
employment: a case study in the Tigray Region of Northern Ethiopia. Ph.D.
thesis, Agricultural Economics and Rural Policy Group, Wageningen University,
Holland. 277p.
Woldenhanna, T. (2002). Rural farm/non-farm income linkages in northern Ethiopia. In:
Promoting farm/nonfarm linkages for rural development: case studies from Africa
and Latin America. Davis, B., Reardon, T., Stamoulis, K., Winters, P. (eds.). FAO,
Rome, 121-144.
Woldenhanna, T., & Oskam, A. (2001). Income diversification and entry barriers:
evidence of the Tigray region in northern Ethiopia. Food Policy, 26: 351-365.
Wooldbridge, J.M. (2002). Econometric Analysis of Cross Section and Panel Data. MIT
Press. Cambridge, MA. 805p.
Wooldbridge, J. M. (2002b). Introductory Econometrics: A Modern Approach, 2nd
edition. Thomson South-Western. Cincinnati, O.H. 735p.
80

Appendix 1
APPENDIX 1
Table A1.1: First stage OLS regression with wage income
OLS
Coef. Std. Err.
fm1sex 0.5192 * 0.2795
fm1age -0.0765*** 0.0089
fm1schooled -0.0843 0.2477
hhlandholdsize -0.0553 0.0504
livestock -0.0771** 0.0361
fixed -0.1605* 0.0903
tabiadismak 0.0095 * 0.0053
edirm -0.0410 0.2846
havei -0.2657 0.3012
adults 0.3690 *** 0.0828
dependents 0.2796 *** 0.0772
non-farmshare_tabia 0.0945 *** 0.0127
constant 4.9861 0.8612
Joint significance of IV
F test of excluded instruments 35.52 ***
Angrist-Pischke multivariate F test of IV 35.52 ***
Underidentification test
Kleibergen-Paap rk LM statistic 56.93 ***
Weak identification test
Cragg-Donald Wald F statistic 33.86
Kleibergen-Paap Wald rk F statistic 35.52
Weak-instrument-robust inference
Anderson-Rubin Wald test 16.70 ***
Stock-Wright LM S statistic 15.70 ***
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
81

Appendix 1
Table A1.2: First stage OLS regression with business income
OLS
Coef. Std. Err.
fm1sex -0.4740 0.3118
fm1age -0.0162** 0.0079
fm1schooled -0.1764 0.2406
hhlandholdsize -0.0057 0.0412
livestock 0.0001 0.0350
fixed 0.2283 ** 0.0926
tabiadismak -0.0066 0.0049
edirm 0.4053 0.2891
havei -0.1640 0.2793
adults -0.0138 0.0801
dependents -0.0989 0.0734
non-farmshare_tabia 0.0009 0.0130
constant -0.4740 0.3118
Joint significance of IV
F test of excluded instruments 0.91
Angrist-Pischke multivariate F test of IV 0.91
Underidentification test
Kleibergen-Paap rk LM statistic 1.85
Weak identification test
Cragg-Donald Wald F statistic 0.84
Kleibergen-Paap Wald rk F statistic 0.91
Weak-instrument-robust inference
Anderson-Rubin Wald test 16.70 ***
Stock-Wright LM S statistic 15.70 ***
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
82

Appendix 2
APPENDIX 2
Table A2.1:OLS estimations of investments in water & land, livestock, equipment and building
OLS estimation coefficients
Water&Land Livestock Equipment Buildings
non-farmincome -0.0623 0.1360 **
0.0492 0.1160 ***
fm1sex 0.6252 ** 0.1146 * 1.3598 *** 0.5260 **
fm1age -0.0055 0.0301 **
fm1schooled 0.1673 1.0729 ***
0.0154 ** -0.0035
0.0936 0.1724
hhlandholdsize 0.1394 *** -0.0723 * -0.0334 -0.0008
Livestock -0.0090 0.1009 ** 0.0196 0.0033
Fixed -0.1128 0.2686 **
0.3080 *** 0.5012 ***
tabiadismak -0.0229*** 0.0037 * -0.0129*** -0.0095**
Edirm 0.3488 0.8483 **
-0.0667 0.4894 *
Havei 0.1728 0.5356 * 0.1953 0.0638
Adults 0.3969 *** 0.1517 0.2137 *** -0.0987
constant 3.7491 *** -1.6775 * -0.8813 -1.8389 **
Notes: *,**,*** respectively significant at 10%, 5% and 1%; robust standard errors
83