adoption and impact of improved agricultural …
TRANSCRIPT
i
ADOPTION AND IMPACT OF IMPROVED AGRICULTURAL
PRACTICES AND WHEAT PRODUCTION EFFICIENCY OF
SMALLHOLDERS IN ARSI ZONE OF ETHIOPIA
A Dissertation Submitted to College of Agriculture and Environmental
Sciences, School of Graduate Studies
HARAMAYA UNIVERSITY
In Partial Fulfillment of the Requirements for the Degree of DOCTOR OF
PHILOSOPHY IN AGRICULTURAL ECONOMICS
By
Tolesa Alemu
November 2014
Haramaya University
ii
APPROVAL SHEET
SCHOOL OF GRADUATE STUDIES
HARAMAYA UNIVERSITY
As members of the Examining Board of the Final Ph.D. Open Defense, we certify that we
have read and evaluated the dissertation prepared by: Tolesa Alemu, entitled ―Adoption and
Impact of Improved Agricultural Practices and Wheat Production Efficiency of Smallholders
in Arsi Zone of Ethiopia‖, and recommend that it be accepted as fulfilling the dissertation
requirement for the degree of DOCTOR OF PHILOSOPHY IN AGRICULTURAL
ECONOMICS.
Dr. Fekadu Beyene -------------------------- ------------------
Name of Chairman
Dr. Bezabih Emana --------------------- ------------------
Name of Major Advisor
Dr. Jema Haji ---------------------- -----------------
Dr. Belaineh Legesse -------------------------- ------------------
Names of Co-advisors
Dr. Mengistu Ketema -------------------------- ------------------
Name of Internal Examiner
Dr. Legesse Dadi -------------------------- ------------------
Name of External Examiner Signature Date
Final approval and acceptance of the dissertation is contingent upon the submission of the
final copy of the dissertation to the Council of Graduate Studies (CGS) through the
Departmental Graduate Committee (DGC) of the candidate‘s major department.
I hereby certify that I have read this dissertation prepared under my direction and recommend
that it be accepted as fulfilling the dissertation requirement.
Dr. Bezabih Emana ------------------------ ---------------------
Name of Dissertation Advisor Signature Date
iii
DEDICATION
I dedicate this dissertation manuscript to my late grandparents and father, Ayano Akka, Gete
Gebisa and Alemu Ayano, for their affection, love and support at my early ages; and to my
mother, Astedu Tola, to my wife Zewudie and our sons, Robele, Milkias, Kenna, and Moti
for their moral and financial support, affection and patience in the course of my study.
iv
STATEMENT OF THE AUTHOR
First, I declare that this dissertation is my bona fide work and that all sources of materials
used for this dissertation have been duly acknowledged. This dissertation has been submitted
in partial fulfillment of the requirements for Ph.D. degree at Haramaya University and is
deposited at the University Library to be made available to borrowers under rules of the
Library. I solemnly declare that this dissertation is not submitted to any other institution
anywhere for the award of any academic degree, diploma, or certificate.
Brief quotations from this dissertation are allowable without special permission provided that
accurate acknowledgement of source is made. Requests for permission for extended quotation
from or reproduction of this manuscript in whole or in part may be granted by the head of the
major department or the Dean of the School of Graduate Studies when in his or her judgment
the proposed use of the material is in the interests of scholarship. In all other instances,
however, permission must be obtained from the author.
Name: Tolesa Alemu Signature:_______________
Place: Haramaya University, Haramaya
Date of Submission: November 2014
v
ACRONYMS AND ABBREVIATIONS
AE Allocative Efficiency
AP Average Product
ANOVA Analysis of Variance
ATE Average Treatment Effect
ATT Average Treatment Effect on the Treated
ATU Average Treatment Effect on the Untreated
CGIAR Consultative Group for International Agricultural Research
CIMMYT International Maize and Wheat Improvement Center
CSA Central Statistical Agency
DEA Data Envelopment Analysis
DMU Decision Making Unit
DSA Development Studies Associates
EAAPP Eastern African Agricultural Productivity Project
EEA Ethiopian Economics Association
EEPRI Ethiopian Economic Policy Research Institute
EIAR Ethiopian Institute of Agricultural Research
ESSP Ethiopian Strategy Support Program
ETB Ethiopian Currency, Birr
FAO Food and Agriculture Organization
GDP Gross Domestic Product
GoE Government of Ethiopia
HU Haramaya University
ICARDA International Center for Agricultural Research in the Dry Areas
IFAD International Fund for Agricultural Development
IFPRI International Food Policy Research Institute
KARC Kulumsa Agricultural Research Center
MC Marginal Cost
MLE Maximum Likelihood Estimation
MoFED Ministry of Finance and Economic Development
vi
MP Marginal Product
NONIE Network of Networks for Impact Evaluation
OLS Ordinary Least Squares
PSM Propensity Score Matching
SFA Stochastic Frontier Analysis
TE Technical Efficiency
TLU Tropical Livestock Unit
VMP Value of Marginal Product
vii
BIOGRAPHICAL SKETCH
The author was born on April 22, 1971 in Arsi Zone of Oromia National Regional State,
Ethiopia. He attended his elementary, junior and high school education at the then Gumguma
elementary, Eteya junior, and Asella comprehensive secondary schools. He joined Haramaya
University (Ethiopia) and Yashwant Singh Parmar University (India) and graduated with
B.Sc. in forestry and M.Sc. in agricultural economics in 1993 and 1999, respectively. In 1994,
he joined the then Oromia natural resource development and environmental protection
authority and served as state forest development and protection officer and project coordinator
for 3.5 years in Illu-Ababor zone. After completion of M.Sc. degree, he joined Oromia Bureau
of Agriculture at Addis Ababa in 2000 and served in agricultural inputs and credit promotion
department as agricultural inputs and credit promotion officer and department head until
2002. He then joined Oromia Credit and Saving Institution at Addis Ababa and served as
research, planning and programming department coordinator until 2004. Since 2005, he
joined Ethiopian Institute of Agricultural Research, Kulumsa Research Center, and has served
as socioeconomics researcher and coordinator of socioeconomics and research-extension
department at Kulumsa Agricultural Research Center till he joined Haramaya University for
his PhD study in 2011/12 academic year.
viii
ACKNOWLEDGMENTS
I am very grateful to several individuals and institutions that have contributed a lot to the
successful completion of this study. I am highly indebted to my major advisor, Dr. Bezabih
Emana, Co-advisors Dr. Jema Haji and Dr. Belaineh Legesse for their valuable and
constructive ideas, suggestions, and comments that have helped to largely improve the
dissertation. I would like to warmly thank them. I sincerely express my thanks to Ethiopian
Institute of Agricultural Research and Kulumsa Agricultural Research Center for granting me
the chance of study; and Eastern Africa Agricultural Productivity Project (EAAPP) for
sponsoring my study and research work. Thanks go to EAAPP‘s technical coordinators for
their facilitation of the matters related to EAAPP during my stay at Haramaya University. I
would also like to express many thanks to Dr. Fekadu Fufa and Dr. Firdisa Iticha from whom
I received a lot of advice and encouragement for pursuing my study. I also greatly enjoyed the
support of School of Agricultural Economics and Agribusiness Management, and School of
Graduate Studies of Haramaya University. I would like to express my gratitude for their kind
help and facilitation for the successfully completion of my study.
In the course of field survey for data collection, I received warm support and cooperation
from Lemu-Bilbilo, Hetosa and Dodota districts agricultural development offices and
development agents of my sample kebeles. I express my sincere thanks and appreciation for
their kind cooperation. I especially thank the then development agents Eyob Ketema,
Adanech Gizaw, Muleta Regasa, Legese Girma, Solomon Hirpo, Girma Dibaba, Sisay
Mersha, Lamirot Tibebe, Teshome Bedada, Haji Abduro, Kelil Eda‘o and Addisu Sokora for
their painstaking involvement and cooperation in field data collection. Thanks go to also
Tarekegn Itana, the field research assistant at Kulumsa Research Center, for his cooperation
and assistance during data entry. I appreciate and recognize the assistance I received from all
of them.
Last, but not least, I am highly grateful to my family for their moral support and patience that
gave me strength for successful completion of my study.
ix
TABLE OF CONTENTS
APPROVAL SHEET ............................................................................................................. ii
DEDICATION...................................................................................................................... iii
STATEMENT OF THE AUTHOR ...................................................................................... iv
ACRONYMS AND ABBREVIATIONS .............................................................................. v
BIOGRAPHICAL SKETCH ............................................................................................... vii
ACKNOWLEDGMENTS .................................................................................................. viii
TABLE OF CONTENTS ..................................................................................................... ix
LIST OF TABLES............................................................................................................... xii
LIST OF FIGURES ............................................................................................................. xv
LIST OF TABLES IN THE APPENDIX ........................................................................... xvi
ABSTRACT ...................................................................................................................... xvii
1. INTRODUCTION ............................................................................................................. 1
1.1. Background ..................................................................................................................... 1
1.2. Statement of the Problem ................................................................................................ 4
1.3. Research Questions ......................................................................................................... 5
1.4. Objectives of the Study ................................................................................................... 6
1.5. Significance of the Study ................................................................................................ 6
1.6. Scope and Limitations of the Study ................................................................................ 7
1.7. Organization of the Dissertation ..................................................................................... 7
2. LITERATURE REVIEW .................................................................................................. 8
2.1. Definition of Terms ......................................................................................................... 8
2.2. Theoretical Models for Efficiency Analysis ................................................................. 10
2.3. Theoretical Models for Adoption Analysis ................................................................... 17
x
TABLE OF CONTENTS (Continued)
2.4. Theoretical Models for Impact Analysis ....................................................................... 23
2.5. Empirical Literature ...................................................................................................... 27
2.5.1. Efficiency analysis ................................................................................................. 27
2.5.2. Adoption analysis ................................................................................................... 29
2.5.3. Impact analysis ....................................................................................................... 34
2.6. Conceptual Framework ................................................................................................. 38
3. RESEARCH METHODOLOGY .................................................................................... 40
3.1. Description of the Study Area ....................................................................................... 40
3.2. Sampling Methods ........................................................................................................ 43
3.3. Methods of Data Collection .......................................................................................... 45
3.4. Methods of Data Analyses ............................................................................................ 46
3.4.1. Efficiency analysis ................................................................................................. 46
3.4.2. Analysis of adoption of wheat row planting and crop rotation .............................. 55
3.4.3. Analysis of impact of wheat row planting ............................................................. 64
4. RESULTS AND DISCUSSION ...................................................................................... 73
4.1. Socioeconomic Profile of Sample Households ............................................................. 73
4.1.1. Sex of household head ........................................................................................... 73
4.1.2. Age and educational status ..................................................................................... 74
4.1.3. Marital status and household size .......................................................................... 76
4.1.4. Land use types ........................................................................................................ 78
4.1.5. Crop production ..................................................................................................... 80
4.1.6. Livestock ownership .............................................................................................. 83
4.1.7. Household income .................................................................................................. 84
xi
TABLE OF CONTENTS (Continued)
4.1.8. Wheat production and adoption of farming practices ............................................ 85
4.2. Wheat Production Efficiency ........................................................................................ 96
4.2.1. Technical efficiency ............................................................................................... 96
4.2.2. Factors affecting technical efficiency .................................................................. 103
4.2.3. Allocative efficiency ............................................................................................ 107
4.2.4. Economic efficiency............................................................................................. 111
4.3. Determinants and Impact of Farming Practices .......................................................... 113
4.3.1. Factors affecting adoption of wheat row planting................................................ 113
4.3.2. Factors affecting choice of precursor crop for rotation ........................................ 116
4.3.3. Impact of wheat row planting on yield ................................................................ 123
5. SUMMARY AND CONCLUSSIONS .......................................................................... 136
5.1. Summary ..................................................................................................................... 136
5.2. Conclusions and Recommendations ........................................................................... 138
6. REFERENCES .............................................................................................................. 140
APPENDICES ................................................................................................................... 150
Appendix A. Descriptive statistics of some socioeconomic variables of households ....... 151
Appendix B. Survey questionnaire for field data collection .............................................. 164
xii
LIST OF TABLES
Table Page
Table 1. Agro-ecological zones of the study area .................................................................... 42
Table 2. Selected kebeles and their sample sizes ..................................................................... 45
Table 3. Definitions and measurement of output, input and inefficiency variables ................ 52
Table 4. Definitions of variables of the Cobb-Douglas cost function...................................... 54
Table 5. Definitions of independent variables assumed to affect wheat row planting ............ 60
Table 6. Definitions of independent variables used in multinomial logit model ..................... 63
Table 7. Distribution of households by sex of the heads (%) .................................................. 74
Table 8. Descriptive statistics of age, education and farming experience of household head . 74
Table 9. Percentages of household heads in different levels of education............................... 75
Table 10. Distribution of sample household heads by marital status (%) ................................ 76
Table 11. Household size of the study area.............................................................................. 77
Table 12. Average household size of districts in adult equivalent ........................................... 78
Table 13. Households in different land holding class (%) ....................................................... 78
Table 14. Average land use types of sampled households in hectares..................................... 79
Table 15. Distribution of household heads by occupation (%) ................................................ 80
Table 16. Average area cultivated (ha) and yield (q/ha) for major crops ................................ 81
Table 17. Average livestock number per households .............................................................. 83
Table 18. Average annual household income in ETB.............................................................. 84
Table 19. Households in different wheat farm sizes (%) ......................................................... 85
Table 20. Average cultivated area, output, sale and yield of wheat in 2012/13 season ........... 86
Table 21. Average cost, income and profit of wheat production (birr/ha) ............................... 87
Table 22. Percentage of households that sold wheat output .................................................... 88
Table 23. Proportion of household heads that were aware of the usefulness of wheat row
planting (%) .............................................................................................................................. 89
Table 24. Analysis of variance for mean difference between planting methods ..................... 92
Table 25. Proportion of households with different choices of precursor crop to wheat planting
(%) ............................................................................................................................................ 93
xiii
LIST OF TABLES (Continued)
Table 26. Proportion of households by what they planted on wheat farm before 2012/13 (%)
.................................................................................................................................................. 94
Table 27. Number of crop types planted by a household in 2012/13 ...................................... 95
Table 28. Households practicing crop rotation and having input services (%) ....................... 96
Table 29. Descriptive statistics of input and output variables used in efficiency analysis ...... 97
Table 30. Descriptive statistics of inefficiency variables......................................................... 98
Table 31. Estimates of elasticities of output for each agro-ecology ...................................... 100
Table 32. Summary statistics of technical efficiency scores.................................................. 101
Table 33. Distribution of households in different technical efficiency ranges (%) ............... 102
Table 34. Estimates of sources of technical inefficiency variables for total sample ............. 105
Table 35. Estimates of inefficiency variables for each agro-ecology .................................... 106
Table 36. Descriptive statistics of variables used in allocative efficiency ............................. 108
Table 37. Maximum likelihood estimates of allocative efficiency for total sample .............. 109
Table 38. Maximum likelihood estimates of allocative efficiency for each agro-ecology ... 110
Table 39. Summary of allocative efficiency scores of households in wheat production ....... 111
Table 40. Summary of mean technical, allocative and economic efficiencies ...................... 112
Table 41. Analysis of variance of economic efficiency scores of districts ............................ 112
Table 42. Determinants of adoption of wheat row planting................................................... 115
Table 43. Probability estimate in the logit model for the use of wheat row planting ............ 116
Table 44. Determinants of the choice of precursor crop to wheat planting ........................... 117
Table 45. Likelihood ratio tests for independent variables .................................................... 119
Table 46. Hausman tests of IIA assumption .......................................................................... 120
Table 47. Probability estimates for choice categories after multinomial logistic regression 120
Table 48. Marginal effect after multinomial logistic regression ............................................ 122
Table 49. Logit estimates for propensity score for Lowland ................................................. 123
Table 50. Performance of matching estimators for Lowland ................................................. 125
Table 51. Estimates of average treatment effects for Lowland .............................................. 126
Table 52. Logit estimate for propensity score for Midland ................................................... 127
Table 53. Summary of propensity scores for participants and non-participants of Midland . 128
xiv
LIST OF TABLES (Continued)
Table 54. Performance of matching estimators for Midland ................................................. 129
Table 55. Estimates of average treatment effects for Midland .............................................. 130
Table 56. Logit estimate for propensity score for Highland .................................................. 131
Table 57. Summary of propensity scores for participants and nonparticipants of Highland . 132
Table 58. Performance of matching estimators for Lemu-Bilbilo district ............................. 133
Table 59. Estimates of average treatment effects for Highland ............................................. 134
xv
LIST OF FIGURES
Figure Page
Figure 1. Production frontier and technical efficiency............................................................. 13
Figure 2. Arsi zone administrative divisions............................................................................ 41
Figure 3. Comparisons of wheat yield at different levels for 2012/13 cropping season .......... 82
Figure 4. Average yield of different methods of planting ........................................................ 90
Figure 5. Kernel density estimate of study area ..................................................................... 103
Figure 6. Graph of propensity score by treatment status for Dodota district ......................... 125
Figure 7. Graph of propensity score by treatment status for Hetosa district.......................... 129
Figure 8. Kernel density of propensity scores of Lemu-Bilbilo district ................................ 132
Figure 9. Graph of propensity score by treatment status for Lemu-Bilbilo district ............... 134
xvi
LIST OF TABLES IN THE APPENDIX
Appendix Table Page
Appendix Table 1. Total area cultivated, production and yield of grain crops for private
holdings in Meher season of 2012/13 .................................................................................... 151
Appendix Table 2. Age, educational status and farming experience of household heads ..... 152
Appendix Table 3. Distribution of marital status of sample household heads by district ..... 153
Appendix Table 4. Descriptive statistics of household size by district.................................. 154
Appendix Table 5. Land use by district (hectares) ................................................................ 155
Appendix Table 6. Number and proportion of sample households planted wheat in row and
broadcast ................................................................................................................................ 156
Appendix Table 7. Average productivity per hectare for wheat planting methods, in quintals
................................................................................................................................................ 156
Appendix Table 8. Livestock population conversion factor into TLU .................................. 157
Appendix Table 9. Conversion factors for adult equivalents ................................................. 157
Appendix Table 10. Thermal zone classification of Arsi zone .............................................. 158
Appendix Table 11. Sensitivity analysis for the impact of row planting in Dodota district .. 159
Appendix Table 12. Sensitivity analysis for the impact of row planting in Hetosa district .. 160
Appendix Table 13. Sensitivity analysis for the impact of row planting in Lemu-Bilbilo
district ..................................................................................................................................... 161
Appendix Table 14. Tests of mean and variance differences among districts for some
variables ................................................................................................................................. 162
xvii
ADOPTION AND IMPACT OF IMPROVED AGRICULTURAL
PRACTICES AND WHEAT PRODUCTION EFFICIENCY OF
SMALLHOLDERS IN ARSI ZONE OF ETHIOPIA
ABSTRACT
This study was carried out in Arsi zone of Ethiopia to measure the level of production
efficiency, identify sources of inefficiencies, factors affecting adoption of wheat row planting
and choice of precursor crop to wheat planting, and the impact of wheat row planting on
yield. Cross-sectional data were collected from 381 randomly selected farm households in
2012/13 cropping season. Descriptive and inferential statistics, Cobb-Douglas Stochastic
Frontier Analysis, Binary Logit, Multinomial Logit, and Propensity Score Matching methods
were employed to achieve the objectives of the study. The average technical efficiency
estimates for lowland, midland and highland agro-ecologies were 57, 82 and 78 percents,
respectively. The allocative efficiency estimates for the lowland, midland and highland agro-
ecologies were 89, 88 and 87 percents, respectively; and the economic efficiency estimates for
the lowland, midland and highland agro-ecologies were about 51, 73, and 69 percents,
respectively. The efficiency estimates show that there is a potential to increase wheat yield in
all agro-ecologies given the current state of technology and input levels. Wheat output
elasticities associated with land, labor, chemical fertilizers and, seed and pesticides were
positive and significant in the lowland whereas in mid and highland agro-ecologies, only
output elasticities of land and chemical fertilizers were significant. Age of household head,
livestock holding size, practice of crop rotation, wheat row planting and access to improved
seed were significant factors that affected technical efficiency in wheat production. The study
also identified that access to improved seed, agricultural extension services, education, and
livestock size positively and significantly affected adoption of wheat row planting. Farming
experience, off-farm income and access to pesticides had positive and significant effects on
farmers’ choices of pulse and vegetable crops being precursors to wheat planting. On
average, the probabilities of choosing cereals and pulses as precursors to wheat planting
were 56 and 36 percents, respectively. The yield of row planted wheat was higher
xviii
significantly by 14.6 and 13.9 percents in midland and highland agro-ecologies, respectively
compared to broadcast planting method. The findings imply that full exploitation of the
potentials of farm inputs increase wheat output in all agro-ecologies; improving efficiency
should be considered through promotion and scaling up of wheat row planting, crop rotation,
livestock production, and increased farm inputs utilization. Agro-ecology and socio-economic
contexts of the farmers and access to farm inputs and institutional services should be integral
of the promotion and scaling up wheat row planting and crop rotation.
Keywords: Wheat in Ethiopia, Production efficiency, Wheat row planting, Crop rotation,
Impact of wheat row planting.
1
1. INTRODUCTION
1.1. Background
Wheat was one of the first domesticated food crops in the world. For 8,000 years, it has been
the basic food of major civilizations in Europe, West Asia, and North Africa. Wheat is grown
on more than 240 million hectares exceeding other commercial crops and continues to be the
most important food grain source for humans (Dixon et al., 2009). Accounting for a fifth of
humanity‘s food, wheat is second only to rice as a source of calories in the diets of consumers
in the developing countries, and it is the first source of protein (Braun et al., 2010). Wheat is
an especially critical ―stuff of life‖ for approximately 1.2 billion ―wheat dependent‖ and 2.5
billion ―wheat consuming‖ poor men, women and children who live on less than USD 2 per
day, and for approximately 30 million poor wheat producers and their families (CIMMYT,
2012).
If population growth continues at double the growth of wheat production, there will likely be
serious difficulties in maintaining a wheat food supply for future generations (Dixon et al.,
2009; CIMMYT, 2012). In 2010 African countries spent more than US$ 12.5 billion on
importing 32 million tons of wheat. Demand for wheat in Africa is growing faster than for any
other food crops. Demand for wheat in the developing world is expected to increase by 60%
by 2050. This will be a major challenge, particularly in cities, where urban population growth
is forecasted to increase by higher percents by 2050. Several countries could achieve wheat
yields exceeding 6 t/ha but are faced with many challenges in realizing this potential
(Rosegrant and Agcaoili, 2010; CIMMYT, 2012). The challenges of globally low and
fluctuating wheat production, rising consumer demand and higher food prices effects require
efforts that dramatically boost farm-level wheat productivity and reduce global supply
fluctuations. Productivity growth is considered to be one of the long term solutions to these
challenges (Diao et al., 2008).
Agriculture is the foundation of rural development and economic growth in developing
countries; and it recently returned to the forefront of development issues because of the
2
impacts of agricultural productivity change on economic growth and poverty reduction
(Headey et al., 2010). Given that wheat is such a prominent cereal in many regions, it is of
critical importance to identify ways to enhance wheat productivity growth in major producing
areas at global, regional, national, and local levels. Increasing productivity requires efficient
utilization of farm inputs, adoption of improved agricultural practices and technologies
(Dorosh and Rashid, 2013).
In Ethiopia, the major challenges facing agriculture are low productivity, low use of improved
farm inputs, and dependency on traditional farming and rainfall. As a result, food insecurity
and poverty are prevalent in the country. The agricultural sector is mainly composed of
smallholder farmers characterized by subsistence production with low input use and low
productivity. The sector supports about 12.7 million smallholder farmers that provide
approximately 95 per cent of agricultural GDP (IFAD, 2011). Smallholders‘ agriculture has
potential to stimulate growth and alleviate poverty, and particularly growth in staple food
sector is important for most low income countries to ensure food security and reduce poverty
more than growth driven by agricultural exports (Diao et al., 2006).
Despite Ethiopian government‘s policy to expand crop production for exports, domestic
consumption and universal food security (MoFED, 2006), the productivity of food crops is the
lowest compared to global average. Agricultural productivity is about 2 tons per hectare for
major crops (CSA, 2013) and it is vulnerable to drought and other adverse natural conditions.
As a result, even though major efforts have been made over the past decade to reduce poverty
in the country, farmers and other rural residents remain poor (IFAD, 2011). The issue of
increasing agricultural productivity has become the main concern to the government
following considerable increase in food price over the last years that followed decades of low
food price (Conradie et al., 2009).
Wheat in Ethiopia is one of the major food crops. It is the fourth important cereal crop with
annual production of about 3.43 million tons produced on area of 1.63 million hectares (CSA,
2013). Based on CSA data of 2013, wheat occupied about 17% of the total cereal area in
Ethiopia with an average yield of 2.11 tons per hectare in 2012/13 cropping season (Appendix
3
Table 1). This is low productivity compared to the world average of 4 tons/ha (FAO, 2009).
The low productivity has made the country unable to meet the high demand and it remained
net importer of wheat despite its huge potential to produce it (Rashid, 2010). The demand for
wheat has been increasing due to growing population, urbanization and the expansion of food
processing industries in the country. It has become a very important staple in recent decades
throughout the country, in part due to the massive wheat food aid shipments (Guush et al.,
2011). Therefore, if the country is to feed the rapidly growing population and meet the high
demand, it needs to increase the production and productivity of wheat in potential producing
areas.
One of the major wheat producing areas in Ethiopia is Arsi zone which has different agro-
ecologies with different wheat producing potentials. Wheat is one of the major food and cash
crops for the farmers in the zone. Even though the zone has high potential in wheat
production, low productivity, that is 2.40 tons/ha (CSA, 2013), is a major concern. Improving
efficiency in production, and adoption of improved agricultural practices are some of the
factors for productivity enhancement. Agricultural extension activities in the country focused
on improving production efficiency and promotion and scaling-up of improved farming
practices. The farming practice, which is believed to increase productivity and currently
popularized through agricultural extension in Ethiopia, is wheat row planting. Wheat row
planting is considered one of good agronomic practices for improving productivity, and it has
become a major extension activity of agricultural development offices of different regional
states. Therefore, impact study on wheat row planting helps to decide whether or not to scale-
up the practice, and to inform the farmers and other stakeholders about the effect of the
planting method in enhancing yield. Crop rotation is another agronomic practice promoted by
agricultural extension activities to improve soil fertility, suppress diseases and pests and
thereby improve crop yield. However, agro-ecology based comparative empirical knowledge
on production efficiency, factors influencing adoption of crop rotation and wheat row planting
and the impact of wheat row planting on yield in the country in general and in the study area
in particular are limited.
4
1.2. Statement of the Problem
Wheat is the most important food and cash crop in Arsi zone. The zonal average yield of
wheat exceeds the national and regional average yields (Appendix Table 1). Different reports
of Arsi zone agricultural offices show that there are farmers who are able to get yield over 4
tons per hectare whereas others get much below the zonal average (2.4 tons/ha). The average
yield is very low compared to global average. To enhance wheat yield, adoption and scaling-
up of improved farming practices have become a major concern of agricultural extension
activities in the study area. One of the major focuses of agricultural extension activities in
recent years is wheat row planting. Agricultural extension services have promoted the practice
and farmers have been using the row planting in the study area. In addition to this, the
government requires farmers to adopt improved farm inputs and be more efficient in crop
production to ensure food security and growth.
However, agricultural productivity depends partly on successful adoption of improved
farming practices as well as on efficient use of resources in the production process. Efficient
production is the basis for achieving overall food security and poverty reduction objectives
particularly in major food crops producing potential areas of the country. Adoption of
improved farm inputs and farming practices are also the basis to trigger the overall economic
growth. Low yield of crops is partly due to poor agronomic practices. Some of yield
improving agronomic practices are crop rotation and row planting. Proper sequence of crops
on specific farm can improve soil fertility, suppress weeds and plant diseases and thereby
improve yield. Row planting at reduced seed rate can reduce plant competition and encourage
tillering. Resource poor farmers could use the practices and improve crop yield. Impact study
on wheat row planting gives information whether or not to continue or scale-up the practice,
and inform the beneficiaries about its effects on wheat yield. Therefore, to enhance wheat
yield in the study area, the level of production efficiency and sources of inefficiencies as well
as factors influencing adoption of improved farming practices and yield effect of row planting
need to be known.
5
Several studies on production efficiencies of selected crops and adoption of agricultural
technologies have been conducted in the country. Most studies (e.g., Kaleab and Brehanu,
2011; Mesay et al., 2013) show inefficiency of farmers in wheat production. However,
documented empirical study is limited on comparative analysis of wheat production efficiency
among agro-ecologies and the sources of inefficiencies in the study area. Studies on
socioeconomic and institutional factors affecting adoption of crop rotation and wheat row
planting, and the impact of row planting on wheat yield are lacking. Adoption studies (Bekele
et al., 2000; Doss et al., 2003; Hailu, 2008; Cavatassi et al., 2011; Solomon et al., 2011;
Teame, 2011; Yu et al., 2011; Hassan et al., 2012; Mesfin et al., 2012) in the country mainly
focused on improved seed and/or chemical fertilizers. They did not analyze and identify
factors influencing the adoption of crop rotation and wheat row planting. The reviewed
literature on impact studies also did not consider the impact of wheat row planting on yield of
wheat. Therefore, this study is designed to fill these research gaps.
1.3. Research Questions
Arsi zone has a potential in wheat production. However, wheat yield (2.4tons/ha) (CSA,
2013) is low compared to about 6 to 7tons/ha of potential yield from research fields of Kulusa
Agricultural Research Center (KARC) released wheat varieties. The low yield is partly due to
inefficient production and poor agronomic practices. Cereal mono-cropping and low use of
chemical fertilizers due to higher price is a problem. The conventional broadcast planting
method at high seed rate creates problem on plant growth and management. Besides, changes
in land productivity and availability of farm inputs, and improved farming practices can affect
famers‘ decision making process. Under such circumstances and low yield, it is important to
address the following questions: How efficient are farm households in wheat production?
What factors are responsible in determining production efficiency? What are the factors
influencing adoption of wheat row planting and choice of precursor crop for rotation in wheat
production? What is the impact of wheat row planting on yield? These questions need
answers. Therefore, this study attempts to answer these questions and fill the knowledge gap.
6
1.4. Objectives of the Study
The general objective of this study was to assess household level of production efficiencies
and sources of inefficiencies as well as the factors influencing adoption of improved farming
practice and its impact on yield of wheat. The specific objectives are:
1. to evaluate the level of technical, allocative, and economic efficiencies in wheat
production of smallholders in major agro-ecologies;
2. to identify factors affecting technical efficiency of smallholder farmers in wheat
production in the study area;
3. to identify factors influencing adoption of crop rotation and wheat row planting in wheat
production;
4. to measure the impact of adoption of wheat row planting on the yield.
1.5. Significance of the Study
Addressing the challenges facing wheat production and productivity is vital for the future of
millions of Ethiopians producing and consuming wheat. Empirical evidences on production
efficiency, sources of inefficiencies and factors influencing adoption of crop rotation and
wheat row planting and their impacts will help for targeted agricultural extension activities.
That is, agricultural extension services will have adequate and evidence based information to
increase efficiency of production, promote and scale-up agronomic practices to improve the
livelihoods of the poor. Therefore, the study provides information on level of production
efficiency and sources of inefficiency, and factors influencing adoption of improved farming
practices and their impact which will help for decisive action to improve production
efficiency, promote and scale-up improved farming practices to increase wheat yield in the
study area and meet the high demand. The study also contributes to the Eastern Africa
Agricultural Productivity Project‘s (EAAPP) objective of enhancing wheat productivity in
Ethiopia in particular and in project‘s mandate countries in general. Overall, the information
that has been generated helps wheat producers, policy makers, agricultural extension workers,
7
and other stakeholders to make evidence based decisions on efficient resource use, promotion
and scaling-up of yield improving agronomic practices.
1.6. Scope and Limitations of the Study
The study focused on technical, allocative, economic efficiencies, and adoption of crop
rotation and row planting in wheat production as well as impact of wheat row planting on
yield. For this purpose, three districts where one district falls in highland, midland, and
lowland were selected. Taking into account the resources available, the study covered and
used cross-sectional data collected from 381 randomly selected farm households of the three
districts of the zone.
As a result, the results obtained from the analysis of one year cross-sectional data of
households farm input and output may not necessarily hold true for other years. Moreover,
since most farmers did not keep their farm input and output records, it was difficult to obtain
accurate data on farm input and output records or utilization. Because of these factors and the
methodological limitation of some analytical tools (e.g. PSM), the results of the study would
not be absolutely free from error.
1.7. Organization of the Dissertation
This dissertation consists of seven categories including five main chapters to address various
issues and objectives of the study. The remainder of the chapters is organized as follows:
Chapter two presents review of literature related to the study. Chapter three deals with the
research methodology that includes description of the study area, sampling techniques, data
collection and data analysis methods. Chapters four and five are concerned with results and
discussion, and summary and conclusions, respectively. The last two parts present lists of
references cited and appendices in the dissertation, consecutively.
8
2. LITERATURE REVIEW
The review of literature covers the theoretical models of efficiency, agricultural technology
adoption, and technology impact analyses as well as empirical literature on analyses of
production efficiency, improved agricultural technology adoption and impact of technology
adoption on agricultural production and productivity.
2.1. Definition of Terms
Efficiency: The term efficiency may be defined as the ability to produce at the potential level
of the farm. Alternatively, efficiency is the ability to produce more output with the same level
of inputs or producing the same level of output with fewer inputs (Coelli et al., 2005).
Technical efficiency reflects the ability of a farm to obtain maximum output from a given set
of inputs. It takes value between zero and one, and a value of one shows fully or 100 percent
technically efficient farm. Inefficiency is defined as the difference between the actual output
achieved and the output achieved by potential or best producers on the production frontier.
Productivity of a farm is defined as the ratio of the output(s) to the input(s) that the farm
produces and uses respectively. Productivity is a level concept and measures of productivity
can be used in comparing performance of farms at a given point of time. When productivity is
measured as the ratio of output to single input, it is called partial productivity measures like
output per worker or per hour worked; and the ratio of output to all inputs combined is total
factor productivity (Coelli et al., 2005).
Efficiency is measured as the average cost for producing a given yield, relative to the lowest
cost option. Economists generally distinguish technical and allocative efficiency. Technical
efficiency refers to the ability to operate on the yield frontier. Allocative efficiency refers to
the ability to meet the marginal conditions for profit maximization where the marginal value
of applying an additional unit of input is equal to the price of the input (Fischer et al., 2009).
Economic efficiency is the product of technical and allocative efficiencies. In this study
technical, allocative, and economic efficiencies in wheat production and the determinants of
technical efficiency were analyzed.
9
Adoption: A simplistic definition of adoption is basically the use of a technology. This is
further elaborated as the incidence/pattern and intensity of adoption. The incidence indicates
whether a farmer has used a technology or not and the latter explains the degree of use of a
technology. An example of pattern or incidence would be whether a farmer has used fertilizer
or not (Yes or No), or percent of farmers that have applied fertilizers on their wheat fields in a
given year in a region or district. The intensity would show the amount or rate of fertilizer that
the farmer applied per hectare (Langyintuo, 2008).
The measurement of adoption of agricultural technologies depends on whether the technology
is divisible or not. Based on these, there are two measures of adoption of a given technology.
The first one is by the share of the farm area under the new technology at a given period of
time. The second measure of adoption at the farm level at a given period of time is
dichotomous (adoption and non adoption), and is measured by the percentage of farmers
using the new technology within a given area (Langyintuo, 2008). Such definition must
distinguish between individual (farm level) adoption and aggregate adoption. Final adoption
at the individual farmer's level is defined as the degree of use of a new technology in long-run
equilibrium when the farmer has full information about the new technology and its potential
(Feder et al., 1982). Therefore, this study focused on factors influencing the adoption of
improved farming practices where adoption is measured as dummy or categorical variable i.e.
whether or not farm households used or practiced yield enhancing farming practices, namely,
wheat row planting and crop rotation i.e. choice of precursor crop to wheat planting for
rotation.
Impact: Impact is defined as the positive and/or negative, intended and/or unintended, direct
and/or indirect, primary and/or secondary effects produced by an intervention. Impact
evaluation investigates the changes brought about by an intervention; and the expected results
of an impact evaluation are important part of the evaluation (Rogers, 2012). Impact evaluation
can be undertaken on interventions at any scale. However, impact evaluation in the present
study only include evaluation containing an estimate of what would have happened on yield if
wheat row planting had not been practiced by households. It includes the comparison of group
who received intervention with a group who did not receive the intervention. The major
10
reasons for doing impact evaluation are to decide whether or not to continue or expand an
intervention, to learn how to scale up and successfully adapt a successful intervention to suit
another context, and to inform intended beneficiaries about the benefits of the intervention.
Agro-ecological zone: Agro-ecological zone is a physical land area grouped together based
on major homogenous physical conditions and similar agricultural land uses (EIAR, 2011). In
Ethiopia, elevation of an area above sea level is the basis for the traditional agro-ecological
divisions. This is due to the association of elevation with temperature, rainfall and agricultural
land use practices. Though the country has many agro-ecological zones, there are six major
traditionally divided agro-ecological zones. The two extremes are the hot lowland and very
cool highland agro-ecological zones with an altitude range of less than 500 and above 3,700
meters above sea level, respectively (EIAR, 2011). Agricultural crops are mainly grown in the
lowland, midlands and highland (dega) agro-ecological zones, that is, within the altitude
range of about 500 to 3200 meters above sea level.
2.2. Theoretical Models for Efficiency Analysis
In economic theory of production, one of the basic economic problems is ―how to produce‖,
means which combination of resources is to be used for the production of goods and which
technology is to be used for their production. Scarcity of resources demands goods should be
produced with the most efficient method. The physical relationship between inputs and
outputs, and this physical relationship along with prices of factors goes to determine the cost
of production. The cost of production governs the supply of goods which together with
demand for them determines the prices of goods (Varian, 2005; Perloff, 2008). The
production is said to be efficient if the productive resources are utilized in such a way that
through any reallocation it is impossible to produce more of one good without reducing the
output of any other good. Thus, the theory of production is about the study of production
functions.
The neoclassical theory is based on individual behavior and typically assumes a single
individual household. However, in empirical work particularly in households with more than
11
one member, individual preferences are aggregated and the household is assumed to behave
as if they were a single individual. While such aggregation is convenient, such restrictive
assumptions may not be capable of investigating intra-household resource allocation and
gender differences in agriculture. However, many empirical studies of farming households
assume that decisions are made at household level and that no conflict arises in the decision
making process of households (Udry et al., 1995; Udry, 1996). Therefore, household level
characteristics are used in determining the efficiency of production as it has been widely
employed in the analysis of agricultural production efficiency in developing countries
including in Ethiopia.
The foundation of the revealed preference approach to production analysis (Farrell, 1957) and
the refinement of this approach to production modeling in the context of static decision
making (Varian, 1984) provide a consistent static theoretical framework to measure and
evaluate production efficiency. The neoclassical theory of production postulates that firms
maximize profits and/or minimize costs subject to certain technological constraints. The
conventional analysis of production proceeds by postulating a parametric form for the
production function or some equivalent parametric representation and then using standard
statistical technique to estimate the unknown parameters from the observed data (Varian,
1984).
The empirical estimation of production functions had begun long before Farrell‘s work,
arguably with Cobb and Douglas (1928). However, until the 1950s, production functions were
largely used as devices for studying the functional distribution of income between capital and
labor at the macroeconomic level (Greene, 1993). Farrell‘s arguments provided an intellectual
basis for redirecting attention from the production function specifically to the deviations from
that function, and re-specifying the model and the techniques accordingly. A series of papers
including Aigner and Chu (1968) and Timmer (1971) proposed specific econometric models
that were consistent with the frontier notions of Debreu (1951) and Farrell (1957). The
contemporary line of research on econometric models begins with the nearly simultaneous
appearance of the canonical papers of Aigner, Lovell, and Schmidt (1977) and Meeusen and
12
van den Broeck (1977), who proposed the stochastic frontier models to combine the
underlying theoretical propositions with a practical econometric framework.
The estimation of production function and measurement of production performance is an
essential component in production function analysis. The natural measure of production
performance is a productivity ratio. Productivity of a firm has been commonly defined as the
ratio of the output(s) of the firm to the inputs it uses, where larger values of this ratio are
associated with better performance (Coelli et al., 2005). However, productivity varies
according to differences in production technology, production process and differences in the
environment in which production occurs. Efficiency analysis is to measure the contribution of
efficiency to productivity. Producers are efficient if they have produced as much as possible
with the inputs they have actually employed and if they have produced that output at
minimum cost. There are two measures of productivity namely, partial productivity and total
factor productivity (TFP). Partial productivity is measured as the ratio of output to a single
input. The ratio of output to all inputs combined is the total factor productivity (Coelli et al.,
2005). The overall productivity performance involves the measurement of efficiency and
effectiveness (Greene, 1997).
There are several methods for performance measurements. The methods differ according to
the type of measure they produce and the data they require; and the assumptions they make
regarding the structure of the production technology. Some methods require data on the
quantities of inputs and outputs, while others require price data and various behavioral
assumptions such as cost minimization, profit maximization, etc (Debreu, 1951; Farrell, 1957;
Aigner and Chu, 1968; Timmer, 1971; Varian, 1984; Coelli et al., 2005).
Figure 1 represents production frontier that shows the relationship between single input and
output. The frontier shows the maximum output level produced with the current level of
technology.
13
Y
F
B
A C
0 X
Figure 1. Production frontier and technical efficiency
Farm households operating on the frontier (F) at points A and B are technically efficient
whereas farm households operating at point C are inefficient because they could increase
output to the level of point B without requiring more input. Points below the frontier are not
technically efficient, and technical efficiency shows the physical relationship of input and
output. If the actual production point lies on the frontier it is perfectly efficient. If it lies below
the frontier then it is technically inefficient. However, according to Coelli et al. (2005) if
information on prices is available, and a behavioral assumption, such as cost minimization or
profit maximization, is appropriate, then performance measures can be devised which
incorporate this information. In such cases it is possible to consider allocative efficiency, in
addition to technical efficiency. Allocative efficiency in input selection involves selecting that
mix of inputs (e.g., labor and capital) that produces a given quantity of output at minimum
cost (given the input prices which prevail). Allocative and technical efficiency combine to
provide an overall economic efficiency measure. Both can be measured by input or output
approach. In the input approach, we evaluate the ability to minimize inputs keeping outputs
fixed, whereas in output approach we evaluate the ability to maximize outputs keeping inputs
fixed (Debreu, 1951; Farrell, 1957).
14
According to Forsund et al. (1980); Battese (1992); Coelli et al. (2005), approaches to
measuring productive efficiency are grouped into non-parametric and parametric frontiers.
The methods based on the econometric (or parametric) approach, and the mathematical (or
non-parametric) approach. The two techniques use different methods to envelop data, and in
doing so they account for random noise and for flexibility in the structure of production
technology. Non-parametric frontiers do not impose a functional form on the production
frontiers and do not make assumptions about the error term. They use linear programming
approaches, and the most popular nonparametric approach has been the Data Envelopment
Analysis (DEA). Parametric frontier approaches impose a functional form on the production
function and make assumptions about the data (Headey et al., 2010). Stochastic frontier
methodology is designed to estimate the underlying production technology represented by
production, cost or other functions while accounting for random noise and incorporating
inefficient behavior of firms (Kumbhakar et al., 2013). The frontier production function
specifies what output can be achieved, if all decisions were taken according to their best
practices. In smallholder farming, a farm‘s technical efficiency is a measure of its ability to
produce relative to the smallholder‘s best-practice frontier, the maximum output possible
from a given set of inputs and production.
If farm households wish to estimate the maximum possible production, but not average
production or average cost, given a set of inputs, an ordinary least square (OLS) regression
cannot be used. This is because the OLS estimates the mean of the dependent variable
conditional on the explanatory variables, but not the maximum possible outputs given a set of
inputs or the minimum possible cost of a set of outputs. To estimate the maxima or the
minima of the dependent variable given explanatory variables, the frontier functions, either
econometric stochastic frontier analysis (SFA) or linear programming data envelopment
analysis (DEA) is used. However, DEA is a non-parametric method that estimates the
efficiency of production of a group of farms, sometimes called decision-making units (DMU).
Since DEA is a non-parametric method no information is provided by the analysis as to the
reasons or sources of inefficiencies (Coelli et al., 2005). Also, DEA does not make allowance
in the analytical method for measurement error or missing data or information in estimating
the efficiencies of production.
15
On the other hand, SFA is a parametric method of analysis that is estimated using maximum
likelihood, which is similar to standard ordinary least squares (OLS) analysis. The use of this
approach enables missing information or data and measurement errors to be captured in the
error term (Coelli et al., 2005). The most important potential advantage of SFA is that it can
separate noise in the data from genuine variations in efficiency, whereas the DEA attributes
all measurement errors or omitted variable effects to inefficiency. This can lead to DEA
results that are difficult to interpret (Headey et al., 2010). Stochastic frontier analysis has
stochastic frontier, i.e. there is a probability distribution which is the basis of maximum
likelihood estimation. SFA usually has only one output but DEA can include more than one
output of a producer. In frontier functions, the disturbance has a distribution all on one side of
zero. Models with one-sided errors that represent inefficiency are known as stochastic frontier
models. The maximum production must be greater than or equal to any value in the sample,
and the minimum cost must be less than or equal to any value in the sample. Taking this view
in to account, this study used SFA to analyze wheat production efficiency of smallholder
farmers as well as the determinants of inefficiency across agro-ecologies. The frontier
provides information about maximum output relative to a best practice frontier through the
inclusion of an error term representing farm household inefficiency.
The models of stochastic production frontier address technical efficiency and recognize the
fact that random shocks beyond the control of farm households may affect agricultural output.
Therefore, in these models, the impact of random shocks on output can be separated from the
impact of technical efficiency variation. This is the most important potential advantage of
SFA since it can separate noise in the data from genuine variations in efficiency. As a result,
only deviations caused by controllable decisions are attributed to inefficiency. But, the DEA
attributes all measurement errors or omitted variable effects to inefficiency. The problem with
SFA is when the data contain zero observation because it is impossible to have the logarithm
of a variable for zero observation (Coelli et al., 2005).
The most common functional forms used in SFA include the Cobb-Douglas, constant
elasticity of substitution and translog production functions. The Cobb-Douglas and translog
models overwhelmingly dominate the applications literature in stochastic frontier and
16
econometric inefficiency estimation (Greene, 1993). In a production (or cost) model, the
choice of functional form brings a series of implications with respect to the shape of the
implied isoquants and the values of elasticities of factor demand and factor substitution. The
Cobb-Douglas production function has universally smooth and convex isoquants. The implied
cost function is likewise well behaved. This good behavior is due to the strong assumption
that demand elasticities and factor shares are constant for given input prices for all outputs
(Greene, 1993). The choice of flexible second order functional form (e.g. translog) relaxes the
restrictions on demand elsticities and elasticity of substitution but it is not monotonic or
globally convex as in the Cobb-Douglas model (Coelli et al., 2005). Moreover, if the interest
rests on efficiency measurement and not on the analysis of the general structure of the
production technology, the Cobb-Douglas production function provides an adequate
representation of the production technology (Taylor et al., 1986).
The Cobb-Douglas function has been the most commonly used function in the specification
and estimation of production frontiers in empirical studies. It is attractive due to its simplicity
and because of the logarithmic nature of the production function that makes econometric
estimation of the parameters a very simple matter. Yin (2000) points out, that this function
may be criticized for its restrictive assumptions such as unitary elasticity of substitution and
constant returns to scale and input elasticities, but alternatives such as translog production
functions also have their own limitations such as being susceptible to multicollinearity and
degrees of freedom problems. Kopp and Smith (1980) suggest that functional specification
has only a small impact on measured efficiency. Furthermore, Coelli and Perelman (1999)
points out that if an industry is not characterized by perfectly competitive producers, then the
use of a Cobb-Douglas functional form is justified.
The modeling and estimation of production efficiency of a farm relative to other farms or the
'best' practice in farming has become an important area of agricultural production study.
Productivity is generally measured in terms of the efficiency with which factor inputs, such as
land, labor, fertilizer, herbicides, tools, seeds and equipment etc are converted to output
within the production process. Fischer et al. (2009) points out that yield gaps and efficiency
gaps are often measuring the same things. However, efficiency gaps may exist even where
17
there are no yield gaps. Farmers may achieve the economically attainable yield, but using
above optimum input levels. As for yield gaps, factors related to farmer characteristics and
system wide constraints explain variation in efficiency across farmers and fields. Technical
efficiency relates largely to timing and technical skills in using inputs and is often explained
by farmer specific knowledge and skills. However, system-level factors, for instance,
management of farm land and irrigation systems can also explain technical inefficiency.
There are two methodological approaches to analyze the sources of technical efficiency using
stochastic production functions. The first approach is the two-stage estimation procedure in
which the stochastic production function is first estimated, and then efficiency scores are
derived. Then, in the second stage, the derived efficiency scores are regressed on explanatory
variables using ordinary least squares method or tobit regression. This approach has been
criticized on grounds that the firm‘s knowledge of its level of technical inefficiency affects its
input choices; hence inefficiency may be dependent on the input variables. The second
approach advocates a one stage simultaneous estimation approach as in Battese and Coelli
(1995), in which the inefficiency effects are expressed as an explicit function of a vector of
farm-specific variables. Therefore, the study used the one stage simultaneous estimation
approach of Cobb-Douglas stochastic frontier production to analyze the relationship between
inputs and output and the sources of inefficiencies.
2.3. Theoretical Models for Adoption Analysis
Adoption of technological innovations in agriculture in less developed countries attracts
attention of development economists. This is because majority of the population drives its
livelihood from agricultural production; and successful adoption of improved technologies
offers opportunity to increase production and productivity and income sustainability (Dorosh
and Rashid, 2013). Therefore, it is useful to investigate the theoretical results of adoption of
agricultural innovations in order to define adoption variables, set precise relationships for
estimation, and suggests hypotheses which can be tested empirically. Moreover, theoretical
analysis can lead to better understanding of interdependence among adoption decisions (Feder
et al., 1982).
18
Besides, the classification of innovations (that is, new methods, customs, or devices used to
perform new tasks) according to form is useful for considering policy questions and
understanding the forces behind the generation and adoption of innovations. Categories in this
classification include mechanical innovations (tractors and combines), biological innovations
(new seed varieties), chemical innovations (fertilizers and pesticides), agronomic innovations
(new management practices), biotechnological innovations, and informational innovations
that rely mainly on computer technologies. Each of these categories may raise different policy
questions. For example, mechanical innovations may negatively affect labor and lead to farm
consolidation. Chemical and biotechnological innovations are associated with problems of
public acceptance and environmental concerns (Sunding and Zilberman, 2000).
Another categorization of innovation is according to form which distinguishes between
process innovations (e.g., a way to modify a gene in a plant) and product innovations (e.g., a
new seed variety). Innovations can also be distinguished by their impacts on economic agents
and markets which affect their modeling; these categories include yield-increasing, cost
reducing, quality-enhancing, risk-reducing, environmental-protection increasing, and shelf-
life enhancing. Most innovations fall into several of these categories. For example, a new
pesticide may increase yield, reduce economic risk, and reduce environmental protection
(Sunding and Zilberman, 2000). Therefore, the present study considered agronomic
innovation category specifically crop rotation and row planting farming practices.
The adoption/diffusion model of agricultural technologies, developed in the United States by
rural sociologists (Rogers, 1983), is a very important model describing a process of change,
i.e. the diffusion of an innovation into a community. It attempted to predict the adoption
behavior of individuals by looking at their personal characteristics, the time factor and the
characteristics of the innovation itself. The model was, for a long time, the main theoretical
model for agricultural extension and the development of agricultural advisory services
(Vanclay and Lawrence, 1994).
According to Feder et al. (1982), adoption decisions in a given period are assumed to be
derived from the maximization of expected utility (or expected profit) subject to land
19
availability, credit, and other constraints. Profit is a function of the farmer's choices of crops
and technology in each time period. It, therefore, depends on his discrete selection of a
technology from a mix of technologies including the traditional technology and a set of
components of the modern technology package. Most adoption models assume that the utility
function of the farmer has one argument, for example, perceived income or perceived
consumption; but in some situations the utility function is assume to have other elements such
as leisure time. Of course, maximization of temporal expected utility represents an
oversimplification of the dynamic considerations that could be made by a sophisticated
planner. But intuition suggests that this "myopic optimization" approach may be a reasonable
representation of decision making by farm households. In point of fact, it has been proved
analytically that, under reasonable circumstances, the myopic optimization outcomes are good
approximations of the outcomes of the more complex inter-temporal optimization problem
(Leigh, 1980).
Given the discrete technology choice, income is a continuous function of land allocation
among crop varieties, the production functions of these crop varieties, the variable usage
inputs, the prices of inputs and outputs and the annualized costs associated with the discrete
technological choice. The discrete technology choice and land and variable input values, the
perceived income may be regarded as a random variable embodying objective uncertainties
with respect to yields (and prices) and the subjective uncertainties associated with the farmer's
incomplete information about the production-function parameters. Most adoption studies
assume that the amount of land a farmer can operate each period is given; and, thus, he or she
maximizes his or her expected utility subject to land availability. Constraint imperfections in
the credit and labor markets may also result in credit and labor availability constraints that
affect the farmer's choice. The solution to the temporal optimization problem at the beginning
of each period determines the type of technology the farmer will use in the period, his
allocation of land among crops, and his use of variable inputs. At the end of each period, the
actual yields, revenues, and profits are realized; and this added information, as well as the
experience accumulated during the period and information on outcomes obtained by other
farmers tends to update the parameters the farmer will use in his decision making for the next
period (Feder et al., 1982).
20
A farmer may become more proficient with his technology as he accumulates information by
using it. Measures of experience with a technology include the length of time the farmer
under consideration and other farmers in the region have used the technology or the total
cumulative amounts of land utilized with the technology by the farmer and other farmers in
the region over time. Variables describing extension efforts and human capital may play the
same role as measures of learning by doing. Similarly, cost and price changes may result from
technological improvements in the production of capital goods or from improvements in the
marketing network of inputs associated with the new technologies. Output prices may be
affected by expanded production of the crop if the innovation is adopted on a wide scale
(Kislev and Bachrach, 1973).
Generally, adoption study is essential because farming practices change continually. Farmers
build on their own experience and that of their neighbors to refine the way they manage their
crops. Changes in natural conditions, resource availability, and market development also
present challenges and opportunities to which farmers respond. In addition, farmers learn
about new technologies from various organizations, programs, and projects dedicated to
research, extension, or rural development. These organizations develop and promote new
varieties, inputs, and management practices. It is essential that such organizations be able to
follow the results of their efforts and understand how the technologies they promote fit into
the complex pattern of agricultural change in which all farmers participate (CIMMYT, 1993).
In adoption of technologies, the first thing is to consider whether adoption is a discrete state
with binary variables or whether adoption is a continuous measure. Many studies use
proportion of farmers that use the technology as a measure of adoption. The proportion of
land allocated to new technologies is also used as a measure of adoption (Langyintuo, 2008).
There are several statistical and econometric methods that can be used in the analysis of
adoption. They include frequency tables, contingency tables, correlation analysis, linear
regression, and binomial choice models. The use of tables helps to compare adopters and non
adopters of technology. The use of binomial choice models involves qualitative response or
regressand variable which, in this case, is adoption of agricultural technology or practice.
There are various methods for the study of qualitative response models. They are the linear
21
probability models, logit model, probit model, multinomial logit and probit, and the tobit
models (Gujarati, 2004; Greene, 2012). In these models, the dependent variable takes on
values which are qualitative or dummy in nature such as adopting or choosing a technology or
not adopting or choosing the technology. The qualitative variables usually indicate the
presence or absence of a quality or an attribute such as adopter or non adopter, row planter or
non row planter etc. To quantify such attributes the presence and absence of the attribute takes
on values 1 and 0, and the variables that assume these values are called dummy variables
(Gujarati, 2004).
Therefore, in models where the dependent variable is qualitative, the objective is to find the
probability of something happening, such as adoption of improved agronomic practices.
Hence, qualitative response regression models are often called probability models (Maddala,
1992; Gujarati, 2004). In regression models, when the dependent variable is continuous, the
linear regression model is probably the most commonly used statistical method. However,
when the dependent variable is discrete or categorical outcome, linear regression models are
not appropriate (Wooldridge, 2010).
The use of linear probability model in adoption study has problems. The problems include
non-normality of the error term, heteroscedasticity of the error term, possibility of estimated
value of the qualitative dependent variable lying outside the 0 - 1 probability range, and the
lower R2 values. However, these problems can be overcome by the use of weighted least
square to resolve heteroscedasticity problem or by increasing the sample size to minimize the
non-normality problem. But the major problem with the linear probability model is that it
assumes probability or expected value of the dependent variable equal to one, given the
independent variable, increases linearly with the independent variable, indicating the marginal
effect of the independent variable remains constant (Gujarati, 2004).
To model adoption study, therefore, a model needs to have two basic features that include:
First, as the independent variable(X) increases, probability Pi = E(Y = 1 | X) increases but
never lie outside the 0-1 probability interval, and second, the relationship between Pi and Xi is
non linear. The model with probability that lie between 0 and 1, and varies nonlinearly with X
22
variable is sigmoid or S-shaped curve that resembles the cumulative distribution function of a
random variable (Gujarati, 2004; Greene, 2012).
The cumulative distribution functions that represent the 0 and 1 response model are the
logistic and the normal that give rise to logit and probit models respectively. The two models
differ in the specification of the distribution of the error term, and in most cases the models
are similar except the logistic distribution (the logit) has flatter tails (Maddala, 1992). The two
models will produce similar results if the distributions of the sample values of Yi not too
extreme. However, a sample in which the proportion Yi = 1 (or the proportion Yi = 0) is very
small will be sensitive to the choice of cumulative distribution function; and the estimation of
the parameters of these non linear models is by using the technique of maximum likelihood
(Greene, 2012). ―In the logit model the slope coefficients of a variable gives the change in the
log of the odds associated with a unit change in the variable, holding all other variables
constant. The rate of change in the probability of an event happening is given by βjPi(1- Pi),
where βj is the partial regression coefficient of the jth
regressor. But in evaluating Pi, all the
variables included in the analysis are involved‖ (Gujarati, 2004).
According to Wooldridge (2010) and Greene (2012), there are a wide variety of non linear
models for categorical dependent variables. The model specification which is most commonly
used for nominal outcomes with more than two categories that are not ordered is the
multinomial logit model (Hausman and McFadden, 1984; Greene, 2012). It is used in a
variety of situations in applied econometrics and social sciences. Multinomial logistic
regression jointly maximizes the likelihood that the estimates of the parameters predicting
each category of the dependent variable could generate the observed sample data. When the
dependent variable has only two categories, the multinomial logistic regression estimates
reduce to binary logistic regression estimates. The procedure of estimation of all categorical
dependent variable models involves the method of maximum likelihood, since qualitative
response models cannot consistently be estimated with linear regression methods
(Wooldridge, 2010; Greene, 2012).
23
2.4. Theoretical Models for Impact Analysis
Development policies and interventions are typically aimed at changing the behavior or
knowledge of households, individuals, and organizations. Underlying the design of the
intervention is a theory which is either explicit or implicit with social, behavioral, and
institutional assumptions indicating why a particular policy intervention will work to address
a given development challenge. For evaluating the nature and direction of an impact,
understanding this theory is critical (Leeuw and Vaessen, 2009). According to the same study,
theories partly require reconstruction and articulation. After articulating the assumptions on
the effect of an intervention on outcomes and impacts, these assumptions will need to be
tested. This can be done by carefully constructing the causal story about the way the
intervention has produced results or by formally testing the causal assumptions using
appropriate methods.
Interventions are embodiments of theories. They comprise an expectation that the introduction
of a program or policy intervention will help ameliorate a recurring social problem; and also
they involve an assumption or set of assumptions about how and why program activities and
resources will bring about changes for the better. Program or intervention theory can be
identified (articulated) and expressed in a graphic display of boxes and arrows, a table, a
narrative description, and so on. The methodology for constructing intervention theory, as
well as the level of detail and complexity, also varies significantly (Trochim, 1989; Rogers et
al., 2000).
Methods for reconstructing the underlying assumptions of policy theories include a policy-
scientific method, which focuses on interviews, documents, and argumentation analysis,
strategic assessment method, which focuses on group dynamics and dialogue as well as
elicitation method, which focuses on cognitive and organizational psychology (Leeuw, 2003).
The overall aim is to reduce the uncertainty about the contribution the intervention is making
to the observed results through an increased understanding of why the observed results have
occurred (or not) and the roles played by the intervention and other factors. The central
question is to what extent changes in outcomes of interest can be attributed to a particular
24
intervention (Leeuw and Vaessen, 2009). Program impact evaluation studies the effect of an
intervention on final welfare outcomes, rather than the program implementation process.
More generally, program impact evaluation establishes whether the intervention had a welfare
effect on individuals, households, and communities, and whether this effect can be attributed
to the concerned intervention.
To know the effect of an intervention on a participating individual, we must compare the
observed outcome with the outcome that would have resulted had that individual not
participated in the program. However, two outcomes cannot be observed for the same
individual. In other words, only the factual outcome can be observed. Thus, the fundamental
problem in any intervention evaluation is the missing data problem (Bryson et al., 2002;
Ravallion, 2005). Estimating the impact of a program requires separating its effect from
intervening factors which may be correlated with the outcomes, but not caused by the
program. To ensure methodological rigor, an impact evaluation must estimate the
counterfactual, that is, what would have happened had the intervention never taken place
(Baker, 2000).
Impact evaluation methods can either be experimental, as when the evaluator purposely
collects data and designs evaluations in advance, or quasi-experimental, as when data are
collected to mimic an experimental situation. Multiple regression analysis is an all-purpose
technique that can be used in virtually all settings (provided that data are available); when the
experiment is organized in such a way that no controls are needed, a simple comparison of
means can be used instead of a regression, because both will give the same answer. Several
designs exist of combinations of ex-ante and ex-post measurements of participant and control
groups. Randomization of intervention participation is considered the best way to create
equivalent groups. Random assignment to the participant and control groups leads to groups
with similar average characteristics for both observables and non-observables, except for the
intervention.
Quasi-experimental methods can be used to carry out an evaluation when it is not possible to
construct treatment and control groups through experimental design. These techniques
25
generate comparison groups that resemble the treatment group, at least in observed
characteristics. According to Jalan and Ravallion (2003), a quasi-experimental method is the
only alternative when neither a baseline survey nor randomizations are feasible options. The
main benefit of quasi-experimental designs is that they can draw on existing data sources and
are thus often quicker and cheaper to implement, and they can be performed after a project
has been implemented, given sufficient existing data. It includes matching methods, double
difference methods and reflexive comparisons.
There are several matching techniques that can be used to create control groups that are as
similar to participant groups as possible (Leeuw and Vaessen, 2009). Propensity score
matching (PSM) method is one of quasi-experimental method to estimate causal treatment
effects. PSM is a method to match program participants with non-participants typically using
individual observable characteristics. Each program participant is paired with a small group of
non-participants in the comparison group that are most similar in the probability of
participating in the program (Becker and Ichino, 2002). It matches control groups to treatment
groups on the basis of observed characteristics or by a propensity scores; the closer this score,
the better the match.
Unlike econometric regression methods, PSM compares only comparable observations and
does not rely on parametric assumptions to identify the impacts of programs and it does not
impose a functional form of the outcome, thereby avoiding assumptions on functional form
and error term distributions, e.g. linearity imposition, multicollinearity and heteroskedasticity
issues. In addition, the matching method emphasizes the problem of common support, thereby
avoiding the bias due to extrapolation to non-data region. Results from the matching method
are easy to explain to policy makers, since the idea of comparison of similar group is quite
intuitive.
Becker and Ichino (2002) and Caliendo and Kopeining (2008) proposed that propensity score
matching as a way to correct for the estimation effects of the program controlling for the
existence of confounding factors based on the idea that the bias is reduced when the
comparison is performed using participants and control subjects who are similar as possible.
26
Difference in difference method is another quasi experimental impact evaluation method.
According to Becker and Ichino (2002), it entails comparing observed changes in outcome
before and after the project for a sample of participants and nonparticipants. Typically, one
collects outcome data of both participants and nonparticipants using a baseline survey before
the program and repeats this survey at some later point after the program is implemented. For
this reason, this approach incurs higher costs and spends more time as compared to propensity
score approach.
Regression discontinuity design is also the other method for impact evaluation. This method
compares outcomes of a group of individuals just above the cut-off point for eligibility with a
group of individuals just below it. The major technical problem of this method is that it
assesses the marginal impact of the program only around the cut-off point for eligibility and
nothing can be said of individuals far away from it (Caliendo and Kopeining, 2008). For the
aforementioned reasons, propensity score matching method is used to measure the impact of
agricultural extension‘s scaling up of wheat row planting on the wheat yield of smallholders.
The problems that quantitative impact evaluation methods attempt to address are the
establishment of a counterfactual: What would have happened in the absence of the
intervention(s)? The elimination of selection effects, leading to differences between the
intervention group (or treatment group) and the control group, and a solution for the problem
of unobservable: the omission of one or more unobserved variables, leading to biased
estimates.
When no comparison group has been created at the start of the project or program, a
comparison group may be created ex-post through a matching procedure; for every member of
the treatment group, one or more members in a control group are selected on the basis of
similar observed (and relevant) characteristics. An alternative way to create a control group
for this case is the method of propensity score matching. This technique involves forming
pairs, not by matching every characteristic exactly, but by selecting groups that have similar
probabilities of being included in the sample as the treatment group. The technique uses all
available information to construct a control group (Leeuw and Vaessen, 2009).
27
2.5. Empirical Literature
2.5.1. Efficiency analysis
Arega and Zeller (2005) studied the empirical performance of the parametric distance
function and data envelopment analysis with application to adopters of improved cereal
production technology in Eastern Ethiopia. Both approaches reveal that there are substantial
technical inefficiencies of production among sample farmers. Technical efficiency estimates
from the two methods are positively and significantly correlated. But the data envelopment
analysis approach is found to be very sensitive to outliers and choice of orientation. The
parametric distance function results are very robust and revealed that adopters of improved
technology have average technical efficiency of 79 percent implying that the farmers could
increase food crop production on average by 21 percent if they fully exploit the potentials of
improved seeds and chemical fertilizers. The authors also revealed that the single output
production frontier approach has been the standard approach to farm level efficiency analysis
and most have revealed substantial inefficiencies of production in developing countries.
Uaiene and Arndt (2009) used a tanslog stochastic frontier production function to estimate
farm household efficiency and its determinants among smallholder farmers in Mozambique.
Panel data analysis shows that variation in farm household efficiency indicates that access to
agricultural technology is the major constraint, and farm household‘s efficiency varied widely
across households and agro-ecologies. Accesses to advisory services, rural credit, and
membership to agricultural association, use of improved agricultural technology (irrigation,
seed, animal traction, and chemicals) are found to significantly reduce the level of farm
households‘ inefficiencies.
Fernandez and Nuthall (2009) analyzed the technical efficiency of sugar cane farmers in
Philippines to identify the source of input use inefficiency. They find out that technical
efficiency of farmers is positively related to farmers‘ age and experience, access to credit,
fertilizer application, and soil type and farm size.
28
Headey et al. (2010) estimated TFP growth in agriculture for 88 countries over the 1970 to
2001 period using DEA and SFA methods. They argue that the SFA results are more
appropriate, and SFA based series are less volatile, more correlated with labor productivity,
and also avoid the technical regress often found with DEA estimates. They reported that SFA
method show advantages over the commonly used DEA technique.
Arega (2010) measured and compared total factor productivity growth in African Agriculture
over the period 1970 to 2004 under contemporaneous and sequential technology frontier. The
study shows that average productivity growth rate of 0.3% per year; the improved technology
show that African agricultural productivity grow at a higher rate of 1.8% per year. Technical
progress is the principal source of productivity compared to efficiency change. Agricultural
research and development, weather, and trade reforms have significant effects on productivity
in African agriculture.
Commercial farms' wheat production efficiency was analyzed in Ethiopia by Kaleab and
Berhanu (2011) using stochastic frontier model on cross-sectional data by fitting Cobb-
Douglass production function. Elasticities of output with respect to seed, chemicals, tractor
hours, and fertilizer are positive and significant whereas area and labor are negative and
significant. The mean technical efficiency is 82 percent; and 99 percent wheat output
variations from the best practice are due to inefficiency. Experience of managers, distance of
farm from main road, value of machinery owned and mechanization services are found to
affect efficiency significantly.
Okpe et al. (2012) assessed resource use efficiency and rice production in Nigeria using gross
margin analysis and stochastic production function on cross-sectional data. The results
indicate that yield and profit of small farmers is low compared to large farmers; the estimated
coefficients using maximum likelihood show positive sign implying increase in quantities of
input use would result in increase of rice output. The inefficiency model also shows that
resource is not fully utilized in all farm categories. It can be noted from this finding that most
farmers are inefficient in farm resource utilization which is also in line with the findings of
several production efficiencies in developing countries.
29
Similarly, technical and scale efficiency analysis in rice production using data envelopment
analysis was conducted in Nigeria by Ogisi et al. (2012). They found out that most rice
farmers (about 77 percent) operate with increasing returns to scale; education level, farmers
experience and extension agents visit significantly influenced the efficiency level of farmers
in rice production. However, farm size is found to be negatively correlated and has no effect
on resource use efficiency.
Mesay et al. (2013) investigated technical inefficiency and the factors affecting wheat
production in two selected waterlogged areas of Ethiopia using stochastic frontier and
translog production function analysis. The study found that wheat producers were inefficient
in production, and their inefficiencies were related to education, gender, land fragmentation,
access to input and output markets. The study also indicated that scaling up of best farmers‘
practices is essential for improving wheat yield.
The reviewed literature shows that there is a limited comparative analyzes of production
efficiency in different agro-ecologies in the country in general and in the study area in
particular. The review also shows that SFA technique is widely used in efficiency analysis,
and the technique shows advantage over DEA technique in efficiency analyses. The studies
also show that adoption of appropriate techniques or farming practices and adoption of
improved technologies can affect efficiency of production; and farmers are inefficient in
agricultural production in developing countries. The sources of inefficiencies are related to
farmers‘ socioeconomic characteristics and access to institutional services and improved farm
technologies. Generally, most of the empirical studies show that socio-economic and farm
characteristics are important sources of technical inefficiency among farmers.
2.5.2. Adoption analysis
Agricultural productivity change, especially food crops productivity, has recently become an
issue of developing countries because of its greater impact on economic growth and poverty
reduction both in rural and urban areas (Headey et al., 2010). In view of this, Ethiopia has
placed substantial emphasis on improving the productivity of smallholder agriculture through
30
increased use of a package of improved agricultural technologies. Smallholder agriculture
producers are increasingly able to select economically viable technologies and practices for
maximum and efficient production (MoFED, 2010). For increased agricultural production and
productivity adoption of improved technologies and farming practices is vital. Therefore,
analyzing the adoption level and identifying the factors that determine the adoption of
technology are essential for targeting agricultural policies and extension services.
The limited use of modern inputs is a major characteristic of crop production in Ethiopia and
it seems to be a major explanation for its current low productivity (Alemayehu et al., 2011).
Increased use of fertilizers has been a major factor explaining perhaps one third to one half of
yield growth in developing countries. Fertilizer use per hectare in Sub-Saharan Africa is low
due to high prices and poor markets; and the low fertilizer use explains a large part of the
lagging productivity growth in Sub-Saharan Africa (Fischer et al., 2009).
In areas where farmers are unable to afford high fertilizers and pesticides prices, crop rotation
can have positive impact on the yield of crops. Crop rotation is one of the oldest and most
fundamental agronomical practices, and is thought to have great impact on increasing crop
yield. Crop rotation means changing the type of crop grown on a particular piece of land from
year to year. It is primarily a management decision based on a desire to optimize financial,
agricultural or environmental objectives through profit and yield maximizations as well as
through minimized pesticide use (Castellazzi et al., 2008). Rotations primarily help in weed
control, improve soil fertility, and increase wheat grain yield when compared to mono-
cropping (Harris et al., 2007; Moghaddam et al., 2011). A well planned rotation reduces weed
pressure by eliminating the constant niche that mono-cropping provides. A leguminous crop
usually precedes cereals for the aim of improving soil fertility. Therefore, the benefits of
rotations could arise from increased nitrogen supply, soil organic matter, and improvement in
soil structure, and decreased pests, disease or weed competition. Hence, choice of appropriate
precursor crop to wheat planting for rotation can affect wheat yield.
The other agronomic practice which currently believed to increase crop yield in Ethiopia is
row planting. The conventional planting method that is broadcasting seed by hand at high
31
seed rates reduce yield because uneven distribution of the seeds makes hand weeding and
hoeing difficult, and plant competition with weeds lowers wheat growth and tillering. This
causes wheat yield reduction. However, row planting with proper distance between rows and
plant density allows for sufficient aeration, moisture, sunlight and nutrient availability leading
to proper root system development. For instance, in United States, Planting wheat in wide
rows in combination with inter-row cultivation reduced weed density by 62 percent and
increased yield 16 percent (Lauren et al., 2012). Hence, to increase wheat yield, promotion
and scaling-up of the planting technique has been carried out in wheat producing areas
through agricultural extension offices. Though its impact linked to problems in
implementation of the program and its recommendations, methodological issues, and over
optimism of the potential of row planting in real farm setting (Vandercasteelen et al., 2013),
the planting technique is seen as best agronomic practice by agricultural policy makers and
extension personnel.
Earlier studies on agricultural technology adoption in Ethiopia (Bekele et al., 2000; Doss et
al., 2003) have focused on the adoption of improved seeds (mainly wheat and maize) and
chemical fertilizers. These studies have been conducted by the support of CIMMYT in
different parts of Ethiopia including Bale and Arsi zones, northern, southern and central
Ethiopia. The study showed that the percentage of seed and fertilizer adopters were increased
overtime.
Hailu (2008) analyzed the influences of farmers‘ learning and risk on the likelihood and
intensity of adoption of improved tef and wheat technologies using probit and tobit and
random effect models and panel data. The study gives adoption level of improved wheat and
tef varieties in terms of the proportion of farm land allotted to these varieties; and adopters of
improved varieties have raised their production by 20 and 39 percents for wheat and tef
respectively.
Adoption study conducted by Solomon et al. (2011) examined that the farmers‘ decisions to
adopt agricultural technologies using cross-sectional sample data from farmers in Ethiopia.
They estimated a Double-Hurdle model to analyze the determinants of the intensity of
32
technology adoption conditional on overcoming seed access constraints. The study reveals
that knowledge of existing varieties, perception about the attributes of improved varieties,
household wealth (livestock and land) and availability of active labor force are major
determinants for adoption of improved technologies.
Teame (2011) has investigated fertilizer adoption and the intensity of fertilizer use in Tigray
region, Ethiopia using a panel data set of a sample of 307 households. The random effect
Panel probit and panel tobit models have been used to examine factors that determine the
probability of fertilizer adoption and the intensity of fertilizer use, respectively. The
likelihood of fertilizer adoption has been explained by the head of the household‘s education
status, labor endowment, farm size, the number of plots that the farmer used, the distance to
plots from homesteads, oxen ownership and the distance to market from residence. On the
other hand, the intensity of the input use is explained by the household head‘s education
status, farm size, manure use, the number of plots the farmer used, the distance to plots from
homesteads, and oxen ownership. The study indicates that geographical locations of
households significantly affect both the likelihood of adoption and the intensity of the input
use. Time also has its own significant impact in determining the intensity of the input use, and
it has less effect on the likelihood of fertilizer adoption in the region.
Cavatassi et al. (2011) also have analyzed adoption of improved sorghum varieties in relation
to weather, market and other social factors, and have found that farmers vulnerable to weather
shocks are less likely to use improved varieties implying modern variety adoption does not
necessarily represent a means of copping with drought.
Yu et al. (2011) examined the extent and determinants of the adoption of fertilizer and seed
technology in Ethiopia using a double hurdle model of fertilizer use for four major cereal
crops: barley, maize, teff, and wheat. The study found out that access to fertilizer and seed is
related to access to extension services and that production specialization together with wealth
play a major role in explaining crop area under fertilizer and improved seed. Inefficiency in
farm input utilization is found to be the most important factors limiting adoption of improved
varieties and fertilizer.
33
Hassan et al. (2012) assessed the determinants of the probability of adoption and intensity of
use of inorganic fertilizer in south Wollo zone of Ethiopia using cross-sectional data. They
have analyzed the effect of farmers‘ demographic, socioeconomic and institutional setting,
market access and physical attributes on the probability and intensity of use of inorganic
fertilizer. A double hurdle model has been employed using data collected from randomly
selected 252 farmers in 2009. The study shows that extension and credit services, age, farm
land size, education, livestock, off/non-farm income and gender have positive impact in
enhancing the adoption of inorganic fertilizer. The study also indicates that physical
characteristics like distance from farmers‘ home to markets, roads, credit and input supply
played a critical role in the adoption of inorganic fertilizers.
Negassa et al. (2012) used Heckman two-stage estimation procedure to investigate factors
influencing the adoption of modern and/or landrace wheat varieties and spatial diversity of
wheat varieties in Turkey. Multinomial logit choice model is used in the first stage to
determine factors influencing farmers‘ adoption of modern varieties and/or landrace varieties
of wheat. Tobit regression model is used in the second stage, conditional on the choice of a
given wheat variety, to assess the determinants of on-farm spatial diversity of wheat varieties.
The analysis of cross-sectional data based on random utility model shows that household size,
the number of owned cattle, the number of buildings on farm, farm size, farm land
fragmentation, the percentage of irrigable farm plots and regional variations are the important
factors in determining the farmers‘ first-stage choice of wheat variety types. In the second-
stage, the farm size and land fragmentation are found to be the key variables influencing the
level of on-farm spatial diversity of wheat varieties.
Some adoption studies conducted in other countries also use similar techniques of analysis.
For instance, farm-specific and market factors affecting the adoption of herbicides and the
level of herbicide use by rice farmers in the Philippines has been analyzed by Beltran et al.
(2013) using a modified version of Heckman‘s two-step method to estimate a random-effects
double-hurdle model for unbalanced panel data. The age of the farmer, household size, and
irrigation use are significant determinants of the decision of farmers to adopt herbicides as an
alternative to manual weeding, while economic variables such as the price of herbicides, total
34
income, and access to credit determine the level of herbicide use. Determinants of both
adoption and level of use are land ownership, farm area, and the method of crop
establishment.
The review of literature indicates that productivity of crops can be improved through adoption
of improved farm inputs and use of improved farming practices that is, improved agronomic
practices which includes method of crop establishment. In areas where smallholders use very
low farm inputs especially chemical fertilizers, the use of cultural practices like crop rotation
and appropriate planting techniques can substantially enhance crop productivity. However,
the reviewed literature on adoption of technologies in Ethiopia mainly focuses on limited
areas as well as on the adoption of some farm inputs. Adoption studies on agricultural
technologies largely or solely focused on crop variety and/or fertilizers. They did not consider
adoption of crop rotation and row planting practices, and the factors influencing the adoption
of these practices. The reviews also show the existing knowledge from what have been done
and the knowledge gaps on the factors influencing adoption of improved agronomic practices
especially wheat row planting and choice of crop planted preceding wheat production for
rotation which are believed to have great potential for increasing wheat productivity.
2.5.3. Impact analysis
Mendola (2007) assessed the potential impact of agricultural technology adoption on poverty
alleviation strategies. The study does so through an empirical investigation of the relationship
between technological change, of the Green Revolution type, and wellbeing of smallholder
farm households in two rural Bangladeshi regions. As technology adoption is not randomly
assigned but there is ‗self-selection into treatment‘, the paper tackles a methodological issue
in assessing the ‗causal‘ effect of technology on farm-household wellbeing through the non-
parametric ‗p-score matching analysis‘. It pursues a targeted evaluation of whether adopting a
modern seed technology causes resource-poor farmers to improve their income and decrease
the propensity to fall below the poverty line. It finds a robust and positive effect of
agricultural technology adoption on farm household wellbeing suggesting that there is a large
35
scope for enhancing the role of agricultural technology in ‗directly‘ contributing to poverty
alleviation.
Todo and Takahashi (2011) estimated the effect of farmer field schools in rural Ethiopia on
income from agriculture. The study employs a difference-in-differences propensity score
matching approach to correct for possible biases due to selection of participants. They find
that by participating in the farmer field schools, agricultural households increased their real
income per worker. They find evidence that the large increase in income is due to the use of
new agricultural practices, such as new varieties, taught and promoted in the farmer field
schools.
Solomon et al. (2011) examined the driving forces behind farmers‘ decisions to adopt
agricultural technologies and the causal impact of adoption on farmers‘ integration into output
market using cross-section data of farmers in Ethiopia. They estimate a Double-Hurdle model
to analyze the determinants of the intensity of technology adoption conditional on overcoming
seed access constraints. They estimate the impact of technology adoption on farmers‘
integration into output market by utilizing treatment effect model, regression based on
propensity score as well as matching techniques to account for heterogeneity in the adoption
decision, and for unobservable characteristics of farmers and their farm. Results show that
knowledge of existing varieties, perception about the attributes of improved varieties,
household wealth (livestock and land) and availability of active labor force are major
determinants for adoption of improved technologies. The results suggest that the adoption of
improved agricultural technologies has a significant positive impact on farmers‘ integration
into output market and the findings are consistent across the three models suggesting the
robustness of the results. This confirms the potential direct role of technology adoption on
market participation among rural households, as higher productivity from improved
technology translates into higher output market integration.
Vandercasteelen et al. (2013) assessed the impact of the promotion of a new agricultural
technology, i.e. row planting at reduced seed rate, on farmers‘ teff yields in Ethiopia. The
results of a randomized control trial show that the program to scale-up row planting on
36
average has a positive effect on teff yield. Depending on the measure of yield used, the study
finds increases between 2 percent—but not statistically significant—and 17 percent. The
findings are in contrast with larger yield increases found on village demonstration plots and in
more controlled settings, as well as with the yield increase expected by teff farmers. The
differences seemingly are linked to problems in implementation of the program and of its
recommendations, methodological issues, and likely over-optimism on the potential of row
planting in real farm settings.
Yenealem et al. (2013) evaluated the impact of integrated land management interventions on
crop production value per hectare and annual gross income of smallholder farm households in
West Harerghe Zone of Oromia National Regional State. Results of the descriptive statistics
showed that before matching there was difference between program and non-program
households in terms of sex, education, farming experience, land holding and livestock
ownership. Estimates of propensity score matching (PSM) indicate the existence of a positive
additional significant crop production value premium per hectare and annual gross income for
program groups compared to non-program groups. This indicates that on average participant
households earned more crop production value per hectare and more gross household income
than their matches. The independent analysis result of the data also revealed that the value of
crop production was fairly higher on moisture stress program kebeles than in the high rainfall
areas of the program.
Gashaw et al. (2014) used household survey data from Ethiopia and evaluated the impact of
agricultural cooperatives on smallholders‘ technical efficiency in crop production. They used
propensity score matching to compare the average difference in technical efficiency between
cooperative member farmers and similar non cooperative farmers. The results show that
agricultural cooperatives are effective in providing support services that significantly
contribute to members‘ technical efficiency. These results are found to be insensitive to
hidden bias and consistent with the idea that agricultural cooperatives enhance members‘
efficiency by easing access to productive inputs and facilitating extension linkages. According
to the findings, increased participation in agricultural cooperatives should further enhance
efficiency gains among smallholder farmers.
37
Ahmed et al. (2014) evaluated the impact of soil conservation interventions on technical
efficiency of smallholder farm households in Arsi Negelle district of Ethiopia using cross
sectional data collected from randomly selected sample households. Stochastic frontier
method has been used to extract the efficiency scores from the production function and to
measure the impact of conservation practices on technical efficiency; Propensity Score
Matching (PSM) was employed. The logistic regression estimation of factors affecting
participation in soil conservation revealed that educational level of the household head,
farming experience and frequency of extension contact significantly affected the participation
decision of households in soil conservation. In matching processes, kernel matching was
found to be the best matching algorism. This method was also checked for covariate balancing
with a standardized bias, t-test, and joint significance level tests. The results revealed that
households that participate in conservation practice have got an improvement in technical
efficiency than those households that were not participated.
Nyangena and Juma (2014) investigated the impact of package adoption of inorganic
fertilizers and improved maize seed varieties on yield among smallholder households in
Kenya. They used a quasi-experimental difference-in-differences approach combined with
propensity score matching to control for both time invariant and unobservable household
heterogeneity. The findings show that inorganic fertilizers and improved maize varieties
significantly increase maize yields when adopted as a package, rather than as individual
elements. The impact is greater at the lower end of the yield distribution than at the upper end.
A positive effect of partial adoption is experienced only in the lower quartile of yield
distribution. The policy implication is that complementary agricultural technologies should be
promoted as a package, and should target households and areas experiencing low yields.
Asres et al. (2014) investigated the effect of agricultural extension service and other factors
on the technical efficiency (TE) of teff producers in northern Ethiopia. Using cross sectional
data they compare TE level of teff producers who are participants and non-participants of
Agricultural Extension (AE) program. They address self-selection in to AE program
participation using propensity score matching method. Translog stochastic frontier production
function is used for TE analysis. The empirical results reveal that, AE program participants‘
38
and non-participants‘ farms have an average TE of 72 and 71% respectively. Both groups of
farms have considerable overall technical inefficiencies, suggesting the existence of immense
potentials for enhancing production through more efficient use of available technology and
resources. Determinants of TE are explained significantly by livestock ownership, credit and
improved seed. Based on the results, they derive policy recommendations to improve farmers‘
teff production performance. The policy measures include the provision of extension services
related to technical skill and farm management capacity of the farmers, demand driven
livestock extension service, greater access to credit and increasing the availability, quality and
adoption of improved seed.
The review of literature on impact of agricultural technology adoption shows that propensity
score matching is the commonly used statistical technique for evaluating the impact of various
agricultural technology adoption, development programs, treatments and policies. Nearest
neighborhood, caliper and kernel matching estimators, and logit model are widely employed
in the estimation of PSM. Most studies used the covariates that include age and educational
level of household, farming experience of household head, household size, and land and
livestock holding sizes, number of oxen owned and cultivated crops, off-farm income,
extension services, number of plot of land owned, etc. These variables are used as
independent variables in the logit model.
2.6. Conceptual Framework
Generally, the background information on wheat production and productivity problems and
the review of literature provide the idea context or the main key factors, concepts and or
variables (i.e. conceptualization of issues) that need to be included in the study. The
interlinked key concepts and/or variables include improving wheat yield, farm inputs
utilization and efficiency in production, adoption of improved farm technologies and farming
practices for enhancing wheat yield, the impact of improved farming practices on yield, and
farm households‘ socioeconomic, institutional as well as environmental factors. These
interlinked factors and concepts directly or indirectly influence one another and have impact
on wheat yield. The state of wheat yield with current inputs and technology utilization can be
39
measured by technical, allocative and economic efficiencies. Efficiency in production is
influenced by various households‘ socioeconomic, institutional, and environmental factors.
Adoption of improved farm inputs and farming practices are key factors for enhancing
agricultural productivity. However, adoption of improved farming practices are affected by
farm households related socioeconomic, institutional, and ecological factors. Successful
adoption of improved farming practices and technologies finally impacts wheat yield. For
targeted agricultural policies and extension activities to enhance wheat yield, the government
and other stakeholders require identifications of the level of efficiency of production, farm
input utilizations, adoption of improved farming practices, and the factors that influence
efficiency of production and adoption of improved farming practices as well as the impact of
these practices on wheat yield.
40
3. RESEARCH METHODOLOGY
This chapter presents the research methodology adopted for conducting the study. It includes
description of the study area, sampling methods, methods of data collection and analyses as
well as specification of the analytical models for efficiency, adoption and impact analyses.
3.1. Description of the Study Area
Arsi zone is found in the central part of Oromia National Regional State of Ethiopia.
According to Oromia Regional State Bureau of Finance and Economic Development report of
2010, the zone astronomically lies between 70
08‘ 58‘‘ N to 80 49‘ 00‘‘ N latitude and 38
0 41‘
55‘‘ E to 400 43‘ 56‘‘ E longitude. It shares border lines with East Shewa zone in the north
and northwest (303km), West Arsi zone in the south and southwest (182km), Bale zone in the
south, southeast and east (189km), and West Harerge zone in the north and northeast (209km)
directions. The zone shares the longest borderline with East Shewa administrative zone
followed by West Harerge and Bale zones. It shares the shortest boundary line with Afar
National Regional State (5km) at the north extreme (www.oromiabofed.org). The highest
point in Arsi is Mount Kaka; other notable mountains in this zone include Mount Chilalo and
Mount Gugu. The total length of the boundary line is about 888km. The administrative center
of the zone is Asella town; other towns in the zone include Abomsa, Bokoji, Sagure, Kersa,
Dhera, Etaya, Arsi Robe, Huruta, etc. Asella is located at 175 km from Addis Ababa along
Adama-Bale Robe main road. It is located at 75 km south of Adama town. Having the total
area of 21,009 km2, Arsi accounts for about 5.8 percent of the total area of the Regional State
of Oromia (Figure 2).
41
Source: Arsi zone profile: www.oromiyaa.com, 25 February 2011.
Figure 2. Arsi zone administrative divisions
Arsi Zone is divided into five agro-climatic zones mainly due to variation in altitude. It is
dominantly characterized by moderately cool (about 40 percent) followed by cool (about 34
percent) annual weather. Cool/cold type of weather is found in the highland areas of Kaka,
Chilalo, Hankolo and Gugu mountains with altitudes of 4245, 4005, 3850 and 3625 meters
above sea level, respectively. Moderately warm weather is found in the lowland areas of
Dodota, Ziway-dugda, Jeju, Amigna, Seru and Merti districts. Some highland districts include
Lemu-bilbilo, Chole and Hankolo-wabe, whereas Hetosa and Tiyo districts mainly fall in
midland. The mean annual temperature of the zone is found between 20-250c in the low land
and 10-150c in the central highland. About 74 percent of zonal land area falls in moderately
cool to cool temperature within the altitude range of 1,500 to 3,200 meters above sea level.
About 20 and 6 percents of the area are moderately warm (20 to 25oc) and cold, respectively.
The cold areas have altitude of 3,200 meters above sea level (www. oromiya.com, 25 January
2013; Appendix Table 11).
42
The rainy season starts in June with the highest concentration in July and August while Belg
season extends from February to April. There is a variation in amount from one place to
another. The central highlands receive a mean annual rainfall between 1000-1400mm where
as the lowland receives a mean annual rainfall between 700-1000mm. Moreover, the rainy
days of the zone varies from 120 to 200 days in the highland parts and slightly decreases as
one goes down to the lowland areas. On average, the zone gets a monthly mean rainfall of 85
mm and an annual mean rainfall of 1020 mm (www.oromiabofed.org). These characteristics
make the zone good potential for production of various agricultural crops.
In addition to wheat, the major cereal crops grown in the zone are barley, maize, teff and
sorghum. However, the contribution of wheat to annual total cereal output is the highest
(45%) followed by barley (20%). It also accounts for 42% of the total cereal area cultivated in
2012/13 (CSA, 2013). This shows the higher relative importance of wheat in food crop
cultivation in the area. However, nationally wheat accounted for 17% of cereal area cultivated
in 2012/13 cropping season (CSA, 2013).
Wheat is grown in lowlands, midlands, and highlands agro-ecological zones of Arsi; and this
study focused on these three major agro-ecologies (Table 1).
Table 1. Agro-ecological zones of the study area
No Agro-ecological zone Local name Altitude above sea level, meters
1 Hot lowland Bereha < 500
2 Lowland Kolla 500 – 1,500
3 Midland Woinadega 1,500 – 2,300
4 Highland Dega 2,300 – 3,200
5 Highland Wurch 3,200 – 3,700
6 Highland Kur >3,700
Source: EIAR, 2011
43
3.2. Sampling Methods
A combination of purposive and probability sampling procedures were used for sample
selection. In the first case, Arsi zone which is a high potential wheat producer in Ethiopia and
three major wheat producing districts representing major agro-ecologies within the zone (one
from highland, midland and lowland areas) were selected. The main reason for purposive
selection of the zone was due to its representativeness in wheat production both in regional
and national perspectives. The zone accounted for about 25 percent of Oromia Regional
State's total wheat production (Appendix Table 1). For sample districts selection, the criteria
used were wheat production potential in the respective agro-ecology; strong research and
extension intervention programs embracing wheat producers; adoption of newly released
improved wheat varieties for high, mid and lowland agro-ecologies that were distributed by
district agricultural offices and Kulumsa Agricultural Research Center; and better extension
activities involving wheat row planting and crop rotation practices in the districts. From the
lists of districts of each agro-ecology, one district was randomly selected, and the total
selected districts were three. The three districts selected were Lemu-Bilbilo from the
highland, Hetosa from the midland, and Dodota from the lowland agro-ecology.
In the second stage of the probability sampling, a list of major wheat growing Farmers‘
Associations (kebeles) within the selected districts were obtained from district agricultural
offices, and then two kebeles were randomly selected from each district from the lists
prepared. If the purpose of a study includes comparing differences among groups (in this case
agro-ecologies), equal number of kebele selection from each district is more efficient
(Kothari, 2004). In the third and final stage, a list of farm households that planted wheat on
their farm and harvested in 2012/13 cropping season was obtained from the selected kebeles‘
agricultural extension offices. The list of farm households was assigned consecutive serial
numbers. Sample farm households were selected by simple random sampling technique.
To determine a representative sample size for the study, population size of wheat farm
households was taken from respective district agricultural offices. Hence, using 95% level of
confidence and chi-square value for one degree of freedom, and proportion of population
44
assumed to be 0.5 with degree of accuracy of 0.05, the sample size was determined based on
the formula given by Krejcie and Morgan (1970):
𝑛 =χ2NP 1−P
𝑑2 𝑁−1 +χ2P 1−P (1)
Where:
n = required sample size
χ2 = tabulated value of chi-square for 1 degree of freedom at 5% significance level (3.841)
N = the population size which is the size of wheat farm households
P = proportion of population assumed to be 0.5 since this would provide maximum sample
size
d = the degree of accuracy expressed as proportion (0.05) i.e. standard error
The total number of wheat producer households in the three districts was 49,546 (Arsi zone
and respective districts agricultural development offices, unprocessed data obtained through
personal communication). The proportions of wheat producers in Lemu-bilbilo (highland),
Hetosa (midland) and Dodota (lowland) districts were 43.3, 34.9 and 21.8 percents,
respectively. Using equation (1) and the given values, the sample size (n) was calculated to be
381. Allocation of sample size to each district was determined proportional to the size of
wheat farm household population of each district. However, to identify participant and non
participant farmers in wheat row planting, selected households were asked and identified in
each kebele if they planted wheat in row in 2012/13 cropping season. Finally, the participant
and non-participant households in wheat row planting were used for the analysis of impact of
wheat row planting on wheat yield of households.
The sample size for each district was determined as:
𝑛𝑖 = 𝑛 𝑁𝑖
𝑁 , and 𝑛 = 𝑛𝑖 (2)
where 𝑛𝑖 is the sample size from each selected district (i = highland, midland, and lowland
districts, n is total sample size of the study which is the sum of the sample size of the three
districts, and 𝑁𝑖 is total wheat farm households in respective district, and N is the total
45
population or wheat farm households of the three districts combined. Accordingly, the
proportional allocation resulted in sample size of 165 in highland, 133 in midland and 83 in
lowland districts (Table 2). Similarly, allocation of sample size to each kebele was made
proportional to the household size of each kebele within their respective districts.
Table 2. Selected kebeles and their sample sizes
District/Kebele Household size Sample size
Lemu-Bilbilo 21,457 165
Lemu-Dima 749 86
Chiba-Michael 684 79
Hetosa 17,296 133
Gonde-Finchema 533 54
Boru-Lencha 781 79
Dodota 10,793 83
Amigna-Debeso 502 36
Dodota-Alem 656 47
Total 49,546 381
3.3. Methods of Data Collection
The data for this study was collected from both primary and secondary sources. Cross-
sectional data was collected from the survey of randomly selected sample farm households.
Structured and pre-tested questionnaire was used to collect primary data. Enumerators were
employed and trained and the interview was conducted in the months of May and June 2013.
Both quantitative and qualitative information were collected. The data collection included
households‘ demographic and socioeconomic characteristics (household sizes, age and sex
structures, education, etc), land holding (agricultural, grazing, wheat land, and others), farm
inputs utilization (seeds, fertilizers, herbicides and fungicides, labor utilization, credit,
extension services), farm outputs, input and output prices, livestock holding, income sources,
agronomic practices including crop rotation, row planting and hand weeding, etc.
46
Secondary information on rainfall amounts (annual mean and cropping season), temperature,
etc. were also collected. Published and unpublished documents were used as sources of
secondary data.
3.4. Methods of Data Analyses
The data were analyzed using descriptive and inferential statistics and econometric methods.
Descriptive and inferential statistics such as mean, percent, standard deviation, t-test, F-test,
Chi-square test, likelihood ratio test, and others as well as analysis of variance (ANOVA)
were used to analyze the data collected from sample households. Analyses of efficiency,
adoption of improved crop production practices, and impact of adoption of farming practices
were carried out using econometric methods. Stochastic Frontier Cobb-Douglas Production
and Cost Functions, Binary logit, Multinomial logit, and Propensity Score Matching
analytical techniques were employed to analyze efficiency, wheat row planting, choice of
crop for rotation and impact of wheat row planting on yield, respectively. For analyses,
STATA software version 11 was used.
3.4.1. Efficiency analysis
3.4.1.1. Specification of the stochastic frontier model
Stochastic Frontier Model was introduced by Aigner et al. (1977) and Meeusen and Van den
Broeck (1977); and for n sample farms, it can be written as:
𝑌𝑖 = 𝑓 𝑋𝑖 ; 𝛽 + ε (3)
Where 𝑌𝑖 is wheat output of the jth
household's farm, i = (1, 2, 3,……, n) are sample
household farms, Xij is the ith
input used by the jth
household and β is a vector of unknown
parameters and ε is composed of error term which can be written as:
ε = vi – ui, (4)
47
where vi is a symmetric random error which represents random variations, or random shocks
in the production of the ith
household, outside the control of the farmer assumed
independently and identically distributed as N(0, 𝜎2). The error term ui is a one-sided non-
negative variable which measures technical inefficiency of the ith
household, the extent to
which observed output falls short of the potential output for a given technology and input
levels. The method helps to decompose deviation of the actual observed wheat output from
the estimated frontier into random variations and inefficiency. Hence,
𝑢𝑖 = Zi𝛿 + wi (5)
Where,
Zi is a vector of variables that explain inefficiency of i th
household. 𝛿 is a vector of unknown
coefficients that are to be estimated in the model, and wi ≥ −𝑍𝑖𝛿 to ensure that ui ≥
0 (Battese and Coelli, 1995).
The technical efficiency of production of the ith
farm in the data set, given the level of inputs,
is defined by the conditional expectation evaluated at the maximum likelihood estimates of
the parameters in the model, where the expected maximum value of Y is conditional on u =0.
The measure of technical efficiency (TE) must have a value between zero and one. Following
from equations (3) and (5), technical efficiency will be estimated as:
TEi = E 𝑌𝑖 𝑢𝑖 , 𝑋𝑖 ∕ E 𝑌𝑖 𝑢𝑖 = 0, 𝑋𝑖 = 𝑒𝑥𝑝(−𝑢𝑖) = 𝑒𝑥𝑝 −𝑍𝑖 𝛿 − 𝑤𝑖 (6)
Given the specifications of the stochastic frontier model expressed in equations (6), the
stochastic frontier output (potential output) for the ith
farm is the observed output divided by
the technical efficiency, and TEi is given by:
Y* = Y i
TE i=
E X iβ+v i−ui
E −ui = exp Xiβ + vi (7)
The parametric specification of frontier in the Cobb-Douglas form for one output and n inputs
is given by:
48
𝑙𝑛𝑦𝑖 = 𝛽0 + 𝛽𝑖 𝑙𝑛𝑥𝑖𝑛𝑖=1 + 𝑣𝑖 − 𝑢𝑖 (8)
Where,
yi is wheat output of ith
household
xi represents vector of farm inputs used as listed in Table 3
𝛽0 is intercept
𝛽𝑖 is vector of production function parameters to be estimated.
To identify the factors in Eq. 8, a linear model was simultaneously estimated with the
coefficients of farm inputs variables. Given the level of technical inefficiency derived from
equation (7) and the above specified X vector inefficiency explanatory variables (Table 3), the
coefficients of inefficiency variables were simultaneously estimated along with the
coefficients of input variables. The linear form of the equation is:
(In) efficiency = βX + ε (9)
3.4.1.2. Variables definition and hypothesis for efficiency analysis
The dependent variable: The dependent variable in technical efficiency analysis is total
wheat output in kilogram. It is influenced by a number of improved farm inputs and
households‘ socio-economic contexts, institutional and environmental factors. The effects of
independent variables or inputs have been hypothesized as follows:
Land: Total land planted with wheat is one of the conventional inputs in efficiency analysis.
Since land is a basic input in farming, its relation to output is positive. That is an increase in
land area cultivated is usually positively related to the total output. More land means more
total output, and less land means less total output, implying the importance of land in
production processes.
Labor: The availability of labor (which included family, exchange, and hired labors) in
production processes is one of the basic inputs. Production theory shows that labor and land
49
are the factors of production that determine production or the output to be realized. Without
labor there will be no output, and labor is usually positively related to output.
Chemical fertilizers: Fertilizers provide soils with essential nutrients that improve the
fertility of soils. Urea and DAP are the basic chemical fertilizers that are widely used in
Ethiopia to provide soils with nitrogen and phosphorous. Low fertilizers use is one of the
reasons for low agricultural productivity in developing countries in general and in Ethiopia in
particular. Therefore, increased use of fertilizers up to the optimum level (the level that the
soil requires) is positively related to the level output.
Seed and Pesticides: Improved farm inputs utilization is a factor for enhancing agricultural
productivity. Seed is basic input, and without it there will be no production or no output.
Therefore, the utilization of optimum level of seed per hectare i.e seed rate is a prerequisite
for better production and productivity; and it is positively related to total output. Weeds and
plant diseases are also the major factors that reduce agricultural yield. Pesticides help to
control weeds and diseases and thereby help in enhancing output. Therefore, these inputs are
positively related to agricultural output. To overcome estimation problems, in this study the
quantity value of seed and pesticides were combined and used as one input variable, and
estimated as the value of seed and pesticides deflated by weighted price of the inputs, the
weights being the share of each input in total cost. This approach of farm input categorization
(that is, independent variables) in stochastic frontier analysis has been applied by other
authors such as Arega (2006) for the study of smallholder maize production efficiency in
eastern Ethiopia. Various farm household socioeconomic and institutional characteristics were
also assumed to affect wheat production efficiency, and they were included in the model as
inefficiency factors.
Age: Age affects farm households‘ experiences in farming processes. Older people usually
have accumulated knowledge that helps them to improve their agricultural production. In this
regard, age is positively related to agricultural productivity. However, sometimes older people
are resistant to changes and unwilling to accept new innovations and test them. In this case,
50
age is negatively related to agricultural production efficiency. Therefore, the effect of age of
household head on efficiency of production could be either positive or negative.
Education: it affects farmers‘ technical skills in farm operation and decision makings.
Educated farmers can easily understand agricultural extension and apply their technical skills
and experiences in production decisions and processes. Therefore, education is assumed to
positively affect farmers‘ wheat production efficiencies or negatively related to production
inefficiency.
Experience: with increased farming experience, farmers are generally better able to assess the
relevance of new farming practices. This often comes from their interactions with their
neighbors and the outside people. Because of their experience, they also tend to be better
placed to acquire the needed skills to use the farm technologies compared with younger ones.
As a result farming experience is usually positively related to efficiency in production and
productivity.
Access and use of improved seed: Farmers make subjective inter-varietal comparisons of the
attributes of new and local varieties and they adopt modern varieties only when they are
perceived as having better characteristics than the locals. Moreover, improved seeds have
better yield due to their resistance to various pests and environmental factors. Therefore,
access to improved seed has positive effect on output and efficiency of production.
Livestock holding: Livestock can be a source of traction power and manure for wheat
production. Besides, income generated from sales of livestock and livestock products is used
for the purchase of improved seeds, fertilizers and pesticides that can influence efficiency of
production and productivity positively.
Household size: household members are sources of labor for agricultural activities, and it was
assumed to affect production efficiency of wheat. Various farm operations such as plowing,
planting, weeding, hoeing, harvesting and threshing are usually performed by households.
51
Women and children are mainly involved in weeding. Therefore, household size is usually
positively related to production efficiency.
Crop types: cultivation of different types of crops may cause competition for farm inputs, or
could show access to seed. Crops compete for land, labor, fertilizers, chemicals and
management practices. Growing of different crops in the same season is also a sign of land
fragmentation. Therefore, the number of different crops cultivated by a farmer can positively
or negatively affect wheat production efficiency through their competition for limited farm
inputs.
Credit: Access to credit services usually enables farm households to acquire various farm
inputs such as improved seeds, chemical fertilizers, pesticides, farm tools, etc that are
essential in improving production and productivity. In the literature, it has been argued that
the lack of credit is a constraint to farm inputs utilization in developing countries. Farmers can
invest in new improved farm inputs either from past accumulated capital or through
borrowing from capital markets. The lack of sufficient accumulated savings by smallholder
farmers prevents them from having the necessary capital for investing in new technologies.
Therefore, access to credit improves production efficiency. However, if the credit is not used
for farming and used for other activities such as petty trade, animal fattening credit may
negatively influence production efficiency.
Off-farm income: participation of households in off-farm income helps the households to
generate additional income that can help to improve their agricultural practices. In this case
such income may have positive impact on efficiency of production. However, participation in
off-farm income usually competes for labor that is used for agricultural activities, and in this
case it may have negative effect on efficiency of production. Therefore, the effect of off-farm
income on efficiency could be positive or negative.
Row planting: row planting at optimum seed rate reduces plant competition for nutrients and
encourages plant growth and tillering. Other management practices such as hand weeding
52
and hoeing can be easily conducted in row planted crop. As result, row planting along with
other agronomic practices can have positive effect on efficiency of production.
Crop rotation: Rotations primarily help in weed control, improve soil fertility, and increase
grain yield when compared to mono-cropping (Harris et al., 2007; Moghaddam et al., 2011).
A well planned rotation reduces weed pressure by eliminating the constant niche that mono-
cropping provides. Therefore, proper sequence of crops on a given land can positively
influence the efficiency of production. Table 3 provides the descriptions and measurements of
the above hypothesized variables that were used in efficiency analysis.
Table 3. Definitions and measurement of output, input and inefficiency variables
Variables Descriptions
ln (output) Natural logarithm of wheat output in kg
ln (area) Natural logarithm of cultivated wheat land (ha)
ln (labor) Natural logarithm of labor (man-days*)
ln (fertilizer) Natural logarithm of chemical fertilizers used in kg
ln (seed/pesticides) Natural logarithm of combined quantity of seed and pesticides in kg
Inefficiency variables
Age Age of household head in years
Education Educational level of household head in number of grades completed
Household size Household size in adult equivalent
Livestock holding Livestock size of household in tropical livestock unit (TLU)
Experience Farming experience of household head in years
Crops types Number of different types of crops cultivated in one season
Seed Access and use of improved seed (1 if yes, 0 otherwise)
Row planting Household practice of planting wheat in row (1 if yes, 0 otherwise)
Credit Access and use of credit service (1 if yes, 0 otherwise)
Crop rotation Household practice of crop rotation (1 if yes, 0 otherwise)
Off-farm income
Household annual income from off-farm activities
in thousand ETB
*A man-day is equivalent to 8 working hours in the study area
53
3.4.1.3. Specification of model for allocative efficiency analysis
Allocative efficiency measures farm‘s success in choosing optimal proportions, i.e. where the
ratio of marginal products for each pair of inputs is equal to the ratio of their market prices. If
a firm is technically and allocativelly efficient, then it is said to be cost effective (Coelli et al.,
2005). For the firm to realize allocative efficiency there should be an optimal combination of
inputs so that output is produced at minimal cost and profit could be increased by simply
reallocating resources. Therefore, the firm has to choose a combination of inputs to be used in
right proportions and technically efficient at low prices so that output is produced at minimal
costs. This results into profit maximization. This study followed the assumptions to maximize
profit as farmers choose the best combination (low costs) of inputs to produce profit
maximizing output level; there was perfect competition in input and output markets;
producers were price takers and assumed to had perfect market information; all inputs were of
the same quality from all producers in the market.
Thus, for the firm to maximize profit, under perfectly competitive markets, which requires
that the extra revenue (Marginal Value Product) generated from the employment of an extra
unit of a resource must be equal to its unit cost (Marginal Cost = unit price of input). In
general, if the firm is to allocate resources efficiently and maximize its profits, the condition
of MVP = MC should be achieved. Based on this theoretical framework, allocative
efficiencies of wheat farmers were established for the study area. The estimation of allocative
efficiency was achieved using the Cobb-Douglas cost function analysis. To maximize profit
from wheat production, famers have to choose the best combination of inputs given the prices
of inputs and output. For the present study, allocative efficiency was measured or estimated
from single output (wheat) four inputs Cobb-Douglas cost function. The four inputs were land
used for wheat production, labor utilization in production, chemical fertilizers used in
production, and seed and pesticides used in wheat production. These are the conventional
inputs commonly used in stochastic frontier production function analysis. It is standard
practice to include only the conventional inputs (land, labor, fertilizers, and other variable
inputs) in frontier analysis (Arega et al., 2006). Non-conventional inputs influence output
54
indirectly by raising the efficiency with which the conventional inputs especially land and
labor is used.
To estimate allocative efficiency, the following Stochastic Frontier Cobb-Douglas Cost
Function was used.
𝑙𝑛𝐶𝑖 = 𝛽0 + 𝛽𝑖 𝑙𝑛𝑥𝑖𝑛𝑖=1 + 𝑣𝑖 + 𝑢𝑖 (10)
Where
Ci is cost of production in ETB, Xi are prices of land, labor and, seed and pesticides. Wheat
output is also represented in Xi. 𝛽0 𝑎𝑛𝑑 𝛽𝑖 are parameters to be estimated. Vi and Ui are as
specified earlier but with positive sign of the inefficiency term since inefficiency factors raise
the cost of production. Price of land was estimated based on the rental value of land in ETB
per hectare in respective study kebeles. Labor wage was estimated in ETB per day, and price
of fertilizer was in ETB per kilogram. Average other input price (seed and pesticides) per
kilogram was estimated based on the proportionate weight of each input in total cost of
production. The allocative efficiency was estimated from the Cobb-Douglas Stochastic
Frontier Cost Function. Before fitting the Cobb-Douglas cost function, all data on each
variable were transformed into natural logarithms.
Table 4. Definitions of variables of the Cobb-Douglas cost function
Variables Descriptions of the labels of variables (all were in natural logarithms)
ln(cost) Cost of production in ETB
ln(1ha) Price (rental value) of land in ETB
ln(wage) Wage of human labor in ETB
ln(fertprice) Price of chemical fertilizers in ETB
ln(othprice) Average price of seed and pesticides used in production in ETB
ln(output) Wheat output in kilogram
The estimation of equation (10) gives 𝛽𝑖 . Then, 𝛽𝑖 can be expressed as:
55
𝜕𝑙𝑛𝑌
𝜕𝑙𝑛𝑋=
1
𝑌×𝜕𝑌
1
𝑋×𝜕𝑋
= 𝑋
𝑌×
𝜕𝑌
𝜕𝑋 = 𝛽𝑖 (11)
Using the coefficient estimates from (11), the marginal product (MPi ) of the ith
factor X was
calculated as:
MPi = 𝜕𝑌
𝜕𝑋𝑖= 𝛽𝑖
𝑌
𝑋𝑖 (12)
But average product (AP) = 𝑌
𝑋𝑖
Where Y is the mean of natural logarithm of wheat output; Xi is the mean of natural logarithm
of input i; 𝛽𝑖 is the estimated coefficient of input i. The value of marginal product of input i
(VMP) can be obtained by multiplying marginal physical product (MPi ) by the price of
output (Py ). Thus,
VMPi = MPi× 𝑃𝑦 (13)
Allocative Efficiency (A.E) = 𝑉𝑀𝑃 𝑖
𝑃𝑖 but Pi = Marginal cost of the i
th input. (14)
Allocative efficiency was determined by comparing the value of marginal product of input i
(VMPi) with the marginal factor cost (MICi). Since farmers were price takers in the input
market, the marginal cost of input i approximates the price of the factor i, Pxi. Hence, if
𝑉𝑀𝑃𝑖 > 𝑃𝑋𝑖 , the input was underused and firm profit could be raised by increasing the use of
this input. But, if 𝑉𝑀𝑃𝑖 < 𝑃𝑋𝑖 , the input was overused and to raise firm profits its use should
be reduced. The point of allocative efficiency (maximum profit) is reached when 𝑉𝑀𝑃𝑖 = 𝑃𝑋𝑖.
Finally, economic efficiency was determined by multiplying technical efficiency with
allocative efficiency.
3.4.2. Analysis of adoption of wheat row planting and crop rotation
Some agronomical practices such as crop rotation and row planting are expected to have
wheat yield increasing effects. This study examined the socioeconomic and institutional
variables that affect the adoption of these agricultural practices. For this purpose, the Logit
56
and Multinomial logit models were used for analyzing wheat row planting and crop rotation,
respectively.
Adoption of agricultural practice is a qualitative or categorical dependent variable (adopt or
not adopt), which is influenced by some explanatory variables. It is possible to compute
Ordinary Least Squares (OLS) for binary choice models, however, this results into
heteroscedastic error terms, that is, the variance term is not constant for all observations, so
that parameter estimates obtained are inefficient, thus classical hypothesis tests, such as t-
ratios, are inappropriate (Gujarati, 2004; Greene, 2012).
The possible solution recommended to overcome most of these problems is the use of Probit
or Logit models. All parameter estimates in these models are asymptotically consistent,
efficient and normal since the models use maximum likelihood estimation (MLE) procedures.
In this case, the analogy of the regression t-test can be applied. The t-ratio follows the normal
distribution and the chi-square test replaces the F-test when testing the significance of the
parameters in the model. Empirical evidence suggests that neither Logit nor Probit have
superiority over the other. The choice becomes a matter of preference (Gujarati, 2004).
Therefore, the logit model was used for the adoption of row planting of wheat. Thus, the
dependent variable took the value 1 (adopting row planting) and 0 (not adopting row planting)
and was regressed on a number of socio-economic characteristics of households.
3.4.2.1. Specification of logit model for factors affecting wheat row planting
Based on Gujirati (2004), the logit model can be specified as follows: If Pi is the probability
of adopting row planting and Xi is the factors influencing the adoption, then:
Pi = E 𝑌 = 1 𝑋𝑖 = 𝛽1 + 𝛽2𝑋𝑖 (15)
Equation (15) can be represented as:
Pi = E 𝑌 = 1 𝑋𝑖 = 1
1+𝑒−(𝛽1+𝛽2𝑋𝑖
) (16)
If Zi is equal to 𝛽1 + 𝛽2𝑋𝑖 , equation (16) can be written as:
57
Pi = 1
1+𝑒−𝑍𝑖 = 𝑒𝑍
1+𝑒𝑍 , and this represents logistic distribution function. (17)
If Pi is the probability of adopting row planting, then (1 – Pi) is the probability of not adopting
row planting which is:
1 - Pi = 1
1+𝑒𝑍𝑖 (18)
Therefore, 𝑃𝑖
1−𝑃𝑖=
1+𝑒𝑍𝑖
1+𝑒−𝑍𝑖= 𝑒𝑍𝑖 is odds ratio in favor of adopting row planting. (19)
Taking the natural logarithm of equation (19), gives
Li = ln 𝑃𝑖
1−𝑃𝑖 = 𝑍𝑖 = 𝛽1 + 𝛽2𝑋𝑖 (20)
Where L is the log of the odds ratio and it is called the logit.
To estimate the logit model, equation (20) can be written as:
Li = ln 𝑃𝑖
1−𝑃𝑖 = 𝛽1 + 𝛽2𝑋𝑖 + 𝑢𝑖 , where 𝑢𝑖 is stochastic error term. (21)
Analysis of variance (ANOVA) was also conducted to see the mean yield variation or
difference between row planting and broadcast planting methods as well as among agro-
ecologies. The idea behind the use of analysis of variance was to test for mean yield
differences by examining the amount of variation within each of the planting methods,
relative to the amount of variation between the methods i.e. planting in row and broadcast.
3.4.2.2. Variables definition and hypothesis for wheat row planting
The variables that were used in the model include:
The dependent variable: the dependent variable was adoption of row planting of wheat
which assumed values of 1 or 0. In this study, a farmer was considered adopter if he or she
used wheat row planting in 2012/13 cropping season and assigned 1, and 0 otherwise.
The explanatory variables: the independent variables that were assumed to affect the
practice of wheat row planting were age and educational level of household head, farming
experience, total land owned, household size, livestock holding size, total number of different
types crops cultivated in 2012/13 cropping season, off-farm income, access to improved seed
58
and credit services, access to chemical fertilizers and agricultural extension services, and
household agro-ecological location.
Age: Since row planting is labor intensive, the effect of age is hypothesized to be negative.
With increase in age farmers are getting older and unable to perform the labor intensive row
planting activity. Moreover, older people are resistant to accept new farming practices such as
row planting and continue to use the accustomed conventional broadcast planting method.
Education: It is often assumed that educated farmers are better able to process information
and search for appropriate technologies to alleviate their production constraints. The belief is
that education gives farmers the ability to perceive, interpret and respond to new information
much faster than their counterparts without education. From literature, it is the case in many
countries that the majority of farmers are illiterate. Nevertheless, it is significant to examine
the role education plays in technology adoption decisions (Langyintuo, 2008).
Experience: with increased years of experience in farming farmers are generally better able
to assess the relevance of new technologies. This often comes from their interactions with
their neighbors and the outside people. Because of their experience, they also tend to be better
placed to acquire the needed skills to use the technologies compared with younger ones.
Therefore, experience can influence the adoption of row planting positively.
Land holding: The size of household farm is a factor that is often argued as important in
affecting adoption decisions. It is frequently argued that households with larger farms are
more likely to adopt an improved technology compared with those with small farms as they
can afford to devote part of their fields to try out the improved technology. However, in
general, the directional effect of farm size on adoption is contradictory (Langyintuo, 2008).
Livestock holding: Livestock can be a source of traction power and manure for wheat
production. Besides, income generated from sales of livestock and livestock products is used
for the purchase of improved seeds, fertilizers and pesticides as well as use hired labor that
can positively influence adoption of row planting.
59
Household size: household members are sources of labor for agricultural activities, and it was
assumed to affect adoption of wheat row planting. Various farm operations such as plowing,
row planting, weeding, hoeing, harvesting and threshing are usually performed by households.
Women and children are mainly involved in weeding. Therefore, household size is usually
positively related to adoption of row planting.
Crop types: cultivation of different types of crops may cause competition for farm inputs; or
could show access to seed. Accessibility to seed positively affects adoption of row planting.
Crops compete for land, labor, fertilizers, chemicals and management practices. Growing of
different crops in the same season is also a sign of land fragmentation. Therefore, the number
of different crops cultivated by a farmer can positively or negatively affect adoption of row
planting through their competition for limited farm inputs.
Credit: Access and use of credit services usually enables farm households to acquire various
farm inputs such as improved seeds, chemical fertilizers, pesticides, farm tools, etc that are
essential in improving production and productivity. In the literature, it has been argued that
the lack of credit is a constraint to farm inputs utilization in developing countries. Farmers can
invest in new improved farm inputs either from past accumulated capital or through
borrowing from capital markets. The lack of sufficient accumulated savings by smallholder
farmers prevents them from having the necessary capital for investing in new technologies.
Therefore, access to credit improves adoption of technology. However, if the credit is not
used for farming and used for other activities such as petty trade, animal fattening, etc credit
may negatively influence the adoption of wheat row planting and yield.
Off-farm income: row planting is labor intensive since it is usually carried out manually. Off-
farm activities compete for households‘ labor used for the agricultural production activities.
However, the income generate may help to obtain farm inputs including labor. In this case,
off-farm income may positively affect adoption of row planting. Therefore, the effect of off-
farm income on adoption of row planting was hypothesized to be positive or negative.
60
Fertilizers: access and use of chemical fertilizers is essential to improve agricultural
productivity. Row planting of seed is usually carried out with fertilizers. In this case access to
fertilizers can positively influence adoption of row planting. However, since row planting is
meant for increasing yield and at the same time fertilizers are also used for increasing yield,
households having access to fertilizers may not go for labor consuming farming practice to
increase their yield. Therefore, the effect of chemical fertilizers on adoption of row planting
was hypothesized to be positive or negative. Table 5 provides measurement and descriptions
of variables used in the model.
Table 5. Definitions of independent variables assumed to affect wheat row planting
Variables Measurement Description
Age Years Age of household head
Education Grades Educational level of household in level completed
Experience Years Farming experience of household head
Land Hectares Total owned land
Household size Adult equivalent Household members involved in farming
Livestock TLU Livestock holding size in tropical livestock unit (TLU)
Crops Number Number of different types of crops cultivated
Off-farm income ETB Household annual off-farm income in thousands ETB
Fertilizers Yes/No Households use of chemical fertilizers, 1 if yes 0 otherwise
Seed Yes/No Access and use of improved seed, 1 if yes and 0 otherwise
Extension
Yes/No
Getting agricultural extension on planting wheat in row, 1
if yes, and 0 otherwise
Credit
Yes/No
Household access and use of credit service, 1 if yes 0
otherwise
Agro-ecology
low/mid/highland
Household agro-ecological setting, 1 if its agro-ecology
and 0 for other two remaining categories.
Seed: Seed is basic input in production. Households usually practice row planting if they get
improved seed. Local varieties of crops are not usually used in row planting. Therefore,
61
access to improved seed was hypothesized to positively affect adoption of wheat row
planting.
Extension: access to agricultural extension services enables households to know and
understand the usefulness of the new technology. Exposure to information of the improved
new technology reduces subjective uncertainty and therefore increases the likelihood of
adoption of the new technology. Therefore, access to agricultural extension service was
hypothesized to positively affect adoption of row planting.
Agro-ecology: different crops are grown in different agro-ecologies. Wheat is mainly grown
in lowland, midland and highland agro-ecologies if suitable variety released for specific agro-
ecology is used. However, for example, farmers in highland may prefer barley to wheat
because of high potential of barley production in the highland. Lowland farmers may prefer
other lowland crops. Moreover, the biophysical nature of the area may affect farmers‘
decision on adoption of new technology. Therefore, the effect of agro-ecological location of
households was hypothesized to be positive or negative depending on the biophysical nature,
farming system and crop preferences.
3.4.2.3. Specification of multinomial logit model for crop rotation
The multinomial logit model was used to analyze factors affecting choice of precursor crop
for rotation for wheat production. Crop rotation in this case is the use of different crops before
wheat planting on specific land by the farm households. The model specification which is
most commonly used for nominal outcomes with more than two categories that are not
ordered is the multinomial logit model (Hausman and McFadden, 1984; Greene, 2012). It is
used in a variety of situations in applied econometrics and social sciences. In the present
study, multinomial logit model was used for modeling smallholders‘ choices of precursor crop
for rotation in wheat production. Because a farmer chooses one crop category among more
than two crop categories as precursor for wheat planting, making the choice that provides the
greatest utility in production processes. Based on Wooldridge (2010) and Greene (2012), the
multinomial logit model can be specified as follows:
62
Let Yi denotes a random variable representing precursor crop chosen to wheat planting by a
farm household, which are discrete and mutually exclusive choices. Xi represents a set of
socioeconomic variables that are assumed to influence smallholders‘ choice of precursor crop
to wheat planting, then,
Prob (Yi = j) = 𝑒𝑋′ 𝑖𝐵𝑗
𝑒𝑋′ 𝑖𝐵𝑗𝐽
𝑗=0
, (22)
The model for precursor crop choice has 3 categories and hence the multinomial logit model
becomes:
Prob (Yi = j ∕ Xi) = 𝑒𝑋′ 𝑖𝐵𝑗
𝑒𝑋′ 𝑖𝐵𝑗2
𝑗=0
, j = 0, 1 and 2. (23)
Where βj is a vector of coefficients on each independent variables Xi, and j = 0,1, and 2 are
choices of precursor crops when the choices are pulse crops, vegetable crops and cereal crops,
respectively. The estimated equation provides a set of probabilities for the choices for a
decision maker with characteristics Xi. Equation (23) can be normalized to remove
indeterminacy in the model by setting β0 = 0 (Greene, 2012). This happens because the
probabilities sum to one. Therefore, the probabilities are:
Prob (Yi = j∕ 𝑋𝑖) = 𝑃𝑖𝑗 = 𝑒𝑋′ 𝑖𝐵𝑗
1+ 𝑒𝑋′ 𝑖𝐵𝑘 2𝑘=1
, j = 0,1,2. (24)
Equation (24) gives the J log-odds ratios:
𝑃𝑖𝑗
𝑃𝑖𝑘 = 𝑋′𝑖 𝛽𝑗 − 𝛽𝑘 = 𝑋′𝑖𝛽𝑗 if K = 0. (25)
The coefficients of multinomial logit model are difficult to interpret directly. It also misleads
to associate βj with jth
outcome. To interpret the effects of independent variables on the
probabilities, equation (24) was differentiated to find the partial effects of the variables on the
probabilities.
𝜕𝑃 𝑖𝑗
𝜕𝑋 𝑖= 𝑃𝑖𝑗 𝛽𝑗 − 𝑃𝑖𝑘𝛽𝑘
𝐽𝑘=0 = 𝑃𝑖𝑗 𝛽𝑗 − 𝛽 (26)
63
The marginal effects measure the expected change in probability of a choice being made with
respect to a unit change in independent variable (Wooldridge, 2010; Greene, 2012). The signs
of the marginal effects and the respective coefficients may be different, because the marginal
effects depend on the sign and magnitude of all other coefficients. Finally, model validity was
tested for the independence of irrelevant alternatives (IIA) assumption using Hausman test.
3.4.2.4. Variables definition and hypothesis for multinomial logit model
Table 6. Definitions of independent variables used in multinomial logit model
Variables Measurement Descriptions
Age Year Age of household head
Education Grades Educational level of household head in levels completed
Experience Years Household head farming experience
Land holding Hectares Household's total land holding size
Livestock
holding
TLU Household's livestock holding size in tropical livestock unit
Income ETB Household's annual off-farm income in thousand birr
Household size AE Household size in adult equivalent
Pesticides Yes/no Household's access to and use of pesticides, 1 if yes and 0
otherwise
Credit Yes/no Household's access to and use of credit service, 1 if yes and
0 otherwise
Fertilizers Yes/no Household's access to and use of chemical fertilizers, 1 if
yes and 0 otherwise
Disease
problem
Yes/no Facing of disease problems in past years, 1 if yes and 0
otherwise
Lowland
location
Yes/no Household's location, 1 if lowland and 0 otherwise
Midland
location
Yes/no Household's location, 1 if midland and 0 otherwise
64
Various socio-economic factors affect farmers‘ adoption of agricultural technologies in
Ethiopia. In this study, households related socio-economic and institutional factors which
were assumed to influence farmers choices of precursor crop include age and educational
level of household head, farming experience of household head, land and livestock holding
sizes, household size, access to and use of pesticides, chemical fertilizers and credit services,
and annual income from off-farm activities as well as households agro-ecological orientation.
The effects of all variables were hypothesized to be positive or negative because the effect of
a variable in multinomial logit model depends on the effects of all other variables in the
model. Details of definitions of variables (see section 3.4.2.2), measurement units and
descriptions of variables have been given in Table 6. The dependent variable (Yi ) and its
outcome categories (Ji) are as defined under equations (22) and (23) above.
3.4.3. Analysis of impact of wheat row planting on yield
For the last three to five years, wheat row planting has been promoted by government
agricultural extension offices in the study area, and farmers have been practicing wheat row
planting. In order to measure the impact of planting of wheat in row on yield of wheat, this
study used propensity score matching (PSM). Propensity score matching technique is widely
used in impact evaluation in the absence of baseline data and when randomization is very
unlikely.
Rosenbaum and Rubin (1983) have developed the PSM statistical tool as a method to reduce
bias in the estimation of treatment effects with observational data sets. Since then the
technique has attracted attention of social and economic program evaluators; and it has
become increasingly popular in medical trials and in the evaluation of economic policy
interventions. The PSM method can be used in many fields of social sciences to evaluate the
effects of public programs and policies (Janan and Ravallion, 2003). The technique enables us
to extract from the sample of non-adopters (non-participating) households a set of matching
households that look like the adopters (participating) households in all relevant pre-
intervention characteristics. In other words, PSM matches each adopter household with a non-
adopter household that has (almost) the same likelihood of adopting row planting.
65
The aim of matching is to find the closest comparison group from a sample of nonparticipants
to the sample of program participants. Closest is measured in terms of observable
characteristics. Individuals with similar propensity scores are paired and the average treatment
effect is then estimated by the differences in outcomes (Greene, 2012). Generally, in order to
infer the impact of an intervention or a program on individual outcome, it is necessary to
create a suitable comparison group among a large group of non-participants, which is
identical to the participating group, except in the attitude of treatment assignment (Caliendo
and Kopeinig, 2005; Raitzer and Kelly, 2008).
After propensity score is estimated, using an appropriate matching estimator is essential for
evaluating the impact of wheat row planting. An estimate of the propensity score is not
enough to estimate the average treatment effect on the treated (ATT) of interest using
(equation 29). The reason is that the probability of observing two units with exactly the same
value of the propensity score is in principle zero since p(X) is a continuous variable. Various
methods have been proposed in the literature to overcome this problem, and three of the most
widely used matching estimators are nearest neighbor, kernel, and caliper matching (Becker
and Ichino, 2002). In nearest neighborhood matching, an individual from a comparison group
is chosen as a matching partner for a treated individual that is closest in terms of propensity
score (Caliendo and Kopeinig, 2005). That is, each person in the treatment group chooses
individual(s) with the closest propensity score to them.
In Kernel based matching, each person in the treatment group is matched to a weighted sum
of individuals who have similar propensity score with greatest weight being given to people
with closer scores. All treated units are matched with a weighted average of all controls with
weights which are inversely proportional to the distance between the propensity scores of
treated and controls. The most common approach is to use the normal distribution (with a
mean of zero) as a kernel, where the weight attached to a particular comparator is proportional
to the frequency of the distribution for the difference in scores observed (Caliendo and
Kopeinig, 2005).
66
If the closest neighbor is far away, the nearest neighbor matching produces bad matches, and
in such situation caliper matching algorithm is used. In caliper matching an individual from
the comparison group is chosen as a matching partner for a treated individual that lies within a
given caliper (propensity score range) and is closest in terms of propensity score (Caliendo
and Kopeinig, 2005). If the dimension of the neighborhood is very small, it is possible that
some treated units are not matched because the neighborhood does not contain a control unit.
But, the smaller the size of the neighborhood the better is the quality of the matches.
The choice of a specific method depends on the data in question, and in particular on the
degree of overlap between the treatment and comparison groups in terms of the propensity
score. When there is substantial overlap in the distribution of the propensity score between the
comparison and treatment groups, most of the matching algorithms will yield similar results
(Dehejia and Wahba, 2002). A good matching estimator is the one that provides low pseudo-
R2 value (Sianesi, 2004), statistically insignificant likelihood ratio test of all regressors after
matching (a matching estimator which balances all explanatory variables between both
groups) (Smith and Todd, 2005) and also expected to retain relatively larger observations for
evaluating the impact of an intervention (i.e; relatively large matched sample size is
preferable). In particular, a rejection of the group means difference test after matching implies
a good balancing of the covariates.
Checking overlapping region and common support is required in using PSM. Imposing a
common support condition ensures that any combination of characteristics observed in the
treatment group can also be observed among the control group (Bryson et al., 2002). No
matches can be made to estimate the average treatment effects on the ATT parameter when
there is no overlap between the treatment and non-treatment groups. The common support
region is the area which contains the minimum and maximum propensity scores of treatment
and control group households, respectively. Only the subset of the comparison group that is
comparable to the treatment group should be used in the analysis i.e., observations which lie
outside this region are discarded from analysis (Caliendo and Kopeinig, 2008). Hence, an
important step is to check the overlap and the region of common support between treatment
67
and comparison group. One means to determine the region of common support more precisely
is by comparing the minima and maxima of the propensity score in both groups.
Balancing test need to be taken care of in using PSM. Because the primary purpose of the
PSM is to serve as a balancing method for covariates between the adopters and non adopters
groups. Balancing test is to check whether the propensity score is adequately balanced. In
other words, a balancing test seeks to examine if at each value of the propensity score, a given
characteristic has the same distribution for the treatment and comparison groups. The
propensity scores themselves serve only as devices to balance the observed distribution of
covariates between the treated and comparison groups. The success of propensity score
estimation is therefore assessed by the resultant balance rather than by the fit of the models
used to create the estimated propensity scores (Dehejia and Wahba, 2002). The parameter of
interest is the average treatment effect on the treated (ATT) (Krasuaythong, 2008). The true
ATT indicates the mean difference in wheat yield between participant and non participant of
row planting, who are identical in observable characteristics and adequately weighted by a
balanced probability of participation.
Sensitivity analysis helps to know whether the final evaluation results are sensitive with
respect to the choice of the balancing score. To test the sensitivity of matching estimators,
either t-test or F test can be used (Caliendo and Kopeinig, 2005). For estimating the scores, a
logit model was used using pre-intervention characteristics of sample wheat farmers and
matching was performed using propensity scores for each observation. The dependent
variable in the logit model was participation in wheat row planting which assumed value 1 if a
farmer practiced row planting and 0 otherwise.
In this study, the main pillars of PSM were wheat farmers, the treatment (wheat row planting)
and potential outcome (wheat yield). The idea was to match those wheat farmers that
practiced wheat row planting with that of a control group (non adopters of row planting)
sharing similar observable characteristics. Then, mean effect of row planting was calculated
as the average difference in yield between adopters (participant in row planting) and non
68
adopters (non participant in row planting) groups i.e. the impact was the change in wheat
yield as an outcome indicator.
3.4.3.1. Specification of model for propensity score matching
Based on Rosenbaum and Rubin (1983), propensity score can be defined as the conditional
probability of receiving a treatment given pretreatment characteristics.
Let YiT
and YiC are the outcome variable for participant (row planting) and non-participant
(broadcast), respectively. The difference in outcome between treated and control groups can
be seen from the following mathematical equation:
i =YiT-Yi
C (27)
YiT: Outcome of treatment (wheat yield in quintal per hectare of the i-th household, when
he/she is participant),
YiC: Outcome of the non-participant individuals (i.e wheat yield of the i-th household when
he/she is non participant in wheat row planting),
i: Change in the outcome as a result of wheat row planting for the i-th household.
Let the above equation be expressed in causal effect notational form, by assigning Di=1 as a
treatment variable taking the value 1 if the individual received the treatment (row planting)
and 0 otherwise.
Then the Average Treatment Effect of an individual i can be written as:
ATE=E (YiTDi=1)-E(Yi
CDi=0) (28)
Where ATE, Average Treatment Effect, which is the effect of treatment on the outcome
variables:
E (YiTDi=1): Average outcomes for individual, with treatment, if he/she would participant
(Di=1).
E (YicDi=0): Average outcome of untreated, when he/she would non-participant, or absence
of treatment (Di=0).
The Average Effect of Treatment on the Treated (ATT) for the sample households is given
by:
69
ATT=E (YiT
- YiCDi=1)=E(Yi
TDi=1)-E(Yi
CDi=1) (29)
The fundamental evaluation problem in estimation of impact is that it is impossible to observe
a person‘s outcome for with and without treatment at the same time. While the post-
intervention outcome E (YiTDi=1) is possible to observe, however, the counterfactual
outcome of the i-th household when she/he does not use the treatment is not observable in the
data.
Thus, estimation of ATE can give a seriously biased result, due to the fact that the population
can differ among the comparison group, not only in terms of treatment status, but even in
terms of other characteristics: this problem is often referred to as the ―fundamental problem of
causal inference‘‘. Thus, simple mean comparison between the treated and non-treated can be
misleading, yet taking the mean outcome of non-participants as an approximation is not
advisable, since participants and non-participants usually differ even in the absence of
treatment (Holland, 1986; Caliendo and Kopeinig, 2008). A solution to this problem is to
construct the unobserved outcome which is called the counterfactual outcome that households
would have experienced, on average, had they not participated (Rosenbaum and Rubin, 1983),
and this is the central idea of matching.
According to Rosenbaum and Rubin (1983), the effectiveness of matching estimators as a
feasible estimator for impact evaluation depends on two fundamental assumptions, namely:
Assumption 1: Conditional Independence Assumption (CIA)
It states that treatment assignment (Di) conditional on attributes, X is independent of the post
program outcome (YiT,Yi
C). In formal notation, this assumption corresponds to:
(YiT,Yi
C)(DXi) (30)
This assumption imposes a restriction that choosing to participate in a program is purely
random for similar individuals. As a consequence, this assumption excludes the familiar
dependence between outcomes and participation that lead to a self selection problem
(Heckman et al., 1998).
70
The conditional average effect of treatment on the treated has a problem, if the number of the
set of conditioning variables (X‘s) is high, and thus the degree of complexity for finding
identical households both from program users and non-users becomes difficult. To reduce the
dimensionality problem in computing the conditional expectation, Rosenbaum and Rubin
(1983) showed that instead of matching on the base of X‘s one can equivalently match treated
and control units on the basis of the ―propensity score‖ defined as the conditional probability
of receiving the treatment given the values of X‘s, notationally expressed as
P(Xi)=Pr(Di=1Xi)
Where Pr is the probability or the logistic cumulative distribution, Di = 1 if the subject was
treated, Xi is a vector of pre-treatment characteristics.
In estimating the propensity scores, all variables that simultaneously affect participation in the
program and outcome variables were included. Thus, the average treatment effect on the
treated conditional on propensity score P(X) can then be written as:
ATT=E (YiTP(X), Di=1) =E (Yi
CP(x), Di=1) (31)
Assumption 2: Assumption of common support
0P(X)1 (32)
The assumption is that P(x) lies between 0 and 1. This restriction implies that the test of the
balancing property is performed only on the observations whose propensity score belongs to
the common support region of the propensity score of treated and control groups (Becker and
Ichino, 2002). Individuals that fall outside the common support region would be excluded in
the treatment effect estimation. This is an important condition to guarantee improving the
quality of the matching used to estimate the ATT. Moreover, implementing the common
support condition ensures that a person with the same X values (explanatory variables) have a
positive probability of being both participant and non participants (Heckman et al., 1999).
This implies that a match may not be found for every individual sample. Rosenbaum and
Rubin (1983) describe assumption one and two together as strong ignore-ability assumption.
71
According to Caliendo and Kopeinig (2008), there are steps in implementing PSM. These are
estimation of the propensity scores, choosing a matching algorithm, checking the common
support condition, testing the matching quality and sensitivity analysis. Each step have been
reviewed and presented under review of theoretical models for impact evaluation in chapter
two. Finally, logistic regression (logit model) was fitted using method of planting as
dependent variable, the listed socioeconomic variables as explanatory variables which were
assumed to determine practice of wheat row planting and the outcome variable, yield.
Summary of Procedures for Propensity Score Matching
The steps in PSM and matching using propensity scores (Rosenbaum and Robin, 1983;
Heckman et al., 1998; Dehejia and Wahba, 2002) are as follows:
1. Representative sample of participants and nonparticipants in wheat row planting
should be obtained. To have better matching, larger sample of nonparticipants, same
questionnaire, same data enumerators training and same survey period should be used.
2. The two samples are combined and a logit model is estimated for participation in
wheat row planting as a function of all variables in the data that are likely to determine
participation and the outcome variables.
3. Predicted values of the probability of participation are generated from the logit
regression as propensity scores for every sample of participants and nonparticipants.
4. Propensity scores outside the range (too low) for nonparticipant farmers and too high
for participants are excluded and matching is allowed within the same agro-ecological
zone to help ensure that matches come from the same environmental orientation.
5. For each farmer in the participant sample, observation is found in the nonparticipant
sample that has the closest propensity score, as measured by the absolute difference in
scores.
6. The mean value of the outcome indicator is calculated for the nearest neighbors. The
difference between that mean and the actual value for the treated observation is the
estimate of the gain due to the practice of row planting for that observation.
7. The mean of individuals‘ gains is calculated to obtain the average overall gain.
72
Propensity scoring is to match participants of row planting with similar non-participant farm
households to estimate the conditional probability of becoming a participant in row planting
(i.e propensity score) given observed household characteristics using logit model, where
participation status in row planting is the dependent variable and covariates are introduced as
independent variables.
3.4.3.2. Variables definition and hypothesis for PSM
A number of socioeconomic variables or factors were assumed to explain farmers‘
participation in wheat row planting and the resulting yield increment in terms of output per
hectare. Propensity score matching was carried out separately for each district (Agro-ecology)
separately to ensure similarity within a given agro-ecology.
The dependent variable: the dependent variable was participation (treatment) in row
planting taking dummy value 1 if there is participation in row planting, and 0 otherwise; and
the outcome variable was wheat yield measured in quintals per hectare per household.
The independent variables: the explanatory variables that were assumed to influence both
the participation of households in row planting and the outcome variable were age, education,
farming experience, land holding, household size, livestock holding, access to improved seed,
access to pesticides, crop diversification, off-farm income, access and use of chemical
fertilizers, credit service, and crop rotation. Details of their definitions and hypotheses have
been given under section 3.4.2.2. The description and the measurement of the variables have
been given in Table 5.
73
4. RESULTS AND DISCUSSION
This chapter presents the results of the analyses of the cross-sectional survey data of the
study. Descriptive statistics on socio-economic profile of households and econometric
estimation results of the analyses of efficiency, adoption of improved farming practices and
impact of farming practice have been given under their respective sections of the chapter.
Results are presented based on agro-ecology i.e. lowland (Dodota), midland (Hetosa), and
highland (Lemu-Bilbilo) districts.
4.1. Socioeconomic Profile of Sample Households
4.1.1. Sex of household head
Sex of the household head can have influence on the agricultural production activity. Social
influences, experiences, flexibility in adopting new technologies, decision making in farm
operations, and participation in social organizations may be associated with sex of
households. The factors influenced by sex, in turn, have impact on households‘ adoption of
improved inputs and farming techniques, and thereby affect production and productivity.
According to results presented in Table 7, male headed households comprise 93 percent,
while 7 percent of households were female headed. The proportions of female headed sample
households in lowland, midland and highland districts are 7.2, 11.3 and 4.2 percents,
respectively. Pearson chi-square test for no association between sex of household head and
agro-ecology gives non-significant chi-square statistic at 5%, implying the non rejection of
the assumption of no association. That is, there appears to be no association between sex of
household head and agro-ecology. Though the percentage of female headed farm households
was very low, women in male headed households play a major role in all major agricultural
activities. Due to differences in asset bases and other factors affecting efficiencies, male and
female headed households could also show different level of performance in wheat
production.
74
Table 7. Distribution of households by sex of the heads (%)
Sex of household heads Lowland
n = 83
Midland
n = 133
Highland
n = 165
Total
n = 381
Female
Male
7.2 11.3 4.2 7.3
92.8 88.7 95.8 92.7
Pearson chi2 (2) = 5.356, pr = 0.069
4.1.2. Age and educational status
Age, educational status and farming experiences of household heads are also essential
variables in farm operations. These variables can influence access to and adoption of
improved agricultural technologies and recommended agronomic and farming practices. It is
usually assumed that older people may have higher accumulated capital, more contacts and
access to extension services and higher labor force. These characteristics may make older
people more prepared, adopt and use improved farm inputs and farming practices. However,
in some cases, younger famers are more active, better educated, have more access to
information and can adopt improved farm technologies. The effect of age may be positive or
negative on adoption of technologies.
Table 8. Descriptive statistics of age, education and farming experience of household head
Household head
characteristics
Lowland
n = 83
Midland
n = 133
Highland
n = 165
Total
N = 381
Deviation of
total (Std.)
Age, years 43.34 48.80 46.52 46.62 11.27
Education, grades 5.41 4.56 4.69 4.80 3.68
Experience, years 22.47 27.87 25.58 25.70 11.09
Source: Computation from own data; and the values are means
75
The average age, educational level completed and farming experience of sample household
heads were 46.62, 4.80 and 25.70 years respectively. The average age of household head was
the highest in midland district (48.8 years) followed by highland district (46.52 years). The
mean age was significantly different among districts with F-statistic of 6.19 with prob > F
equal to 0.0023; and average farming experience of household head was also significantly
different among the districts (F = 6.25, prob > F = 0.0021). Similarly, average educational
status of household heads in terms of grade completed was 5.41, 4.56 and 4.69 for lowland,
midland and highland districts respectively, and the mean levels were not significantly
different (Appendix Table 14). However, Bartlett‘s test for equal variance for education gave
significant chi-square statistic (7.76) at 5% level of significance (prob > chi2 = 0.021),
implying the variance of education was significantly different among agro-ecologies.
Educated and experienced farmers are better able to process information and search for
appropriate technologies to alleviate their production constraints. The assumption is that
education enables the farmer to perceive, interpret and respond to new information much
faster than an illiterate farmer.
Table 9. Percentages of household heads in different levels of education
Educational level Lowland
(n = 83)
Midland (n
= 133)
Highland
(n = 165)
Total
(N=381)
Illiterate 13.25 24.06 15.15 17.85
1 to 4 26.51 28.57 32.73 29.92
5 to 8 40.96 24.81 36.97 33.60
9 to 10 9.64 12.78 11.52 11.55
11 to 12 9.64 8.27 3.64 6.56
above 12
0.00
1.50
0.00
0.52
Total
100
100
100
100
Source: Computation from own data
Table 9 shows the percentage of household heads in different level of education based on the
level of educational grade completed by the household head. Accordingly, the majority (33.60
76
percent) of the sample household heads were in educational level of 5 to 8 grades. This was
followed by education level of 1 to 4 with 29.92 percent indicating that 63% of the household
heads attended 1-8th grade. It was found that only 6.56 percent of the sample farmers reached
11 and 12 grades. About 18 percent of sample household heads were illiterate and they did not
attend formal or informal education. There were only 2 sampled farmers whose educational
level was above 12 grades of schooling.
4.1.3. Marital status and household size
Table 10 depicts the distribution of household heads by marital status. About 91 percent of the
sample household heads were married; about 8 percent divorced and widowed and only 1
percent was single household heads. Analysis of marital status shows that the percentage of
married sampled household heads was about 94, 86 and 93 percents for lowland, midland and
highland districts, respectively.
Table 10. Distribution of sample household heads by marital status (%)
Marital status
Lowland
n = 83
Midland
n = 133
Highland
n = 165
Total
N=381
Married 94.0
85.7
92.7 90.6
Single 0.0
1.5
1.2 1.0
Divorced 3.6
3.0
2.4 2. 9
Widowed 2.4
9.8
3.6 5.5
Pearson chi2 (6) = 8.8789, pr = 0.181
Source: Computation from own data
The test whether marital status and agro-ecology are independent of one another shows non-
significant chi-square statistic (chi2 (6) = 8.8789, pr = 0.181), depicting that marital status and
77
agro-ecology are independent of one another i.e. no association among them. All family
members including children participate in farm activities; and this makes the large family
more advantageous than a household with one or two individuals.
Household is the major source of labor for farming. In rural areas, all household members
including children involved in different farming activities on full time or part time basis.
Plowing, sowing, weeding, harvesting and threshing of crops and livestock rearing are mainly
performed by household members. As a result, the size of household can affect the labor
availability and hence various operations in crop and livestock production and management.
Table 11. Household size of the study area
Sex of household
Lowland
Midland
Highland
Total
Std. Deviation
Male 3.43 3.35 3.52 3.44 1.68
Female 3.18 2.73 3.03 2.96 1.49
Total 6.63 6.08 6.53 6.40 2.43
Source: Computation from own data; and the values are means
Table 11 shows that the average household size for the study area. The average persons per
household were about 7. Male and female average family sizes were 3.44 and 2.96
respectively. The average household sizes for lowland, midland and highland districts were
6.63, 6.08 and 6.53, respectively. Households in lowland district have largest the household
size while households in midland district have relatively the smallest household size.
However, statistical test for mean difference (Appendix Table 14) shows non-significant
mean difference among agro-ecologies with F-value of 1.72 and prob > F equal to 0.181(i.e.
the p-value).
Table 12 shows average household size in adult equivalent. Based on the conversion factors
given in Appendix Table 9, the household sizes were converted in to adult equivalent.
Accordingly, the household size in adult equivalent ranges from 0.5 to 11.5 with an average
of 4.16 for the whole study areas. The average household sizes in adult equivalents were 4.18,
78
4.23 and 4.09 for lowland, midland and highland districts, respectively. The test for mean
difference of household size in adult equivalent (Appendix Table 14) shows statistically non-
significant difference among the agro-ecologies (F = 0.26; prob > F = 0.774).
Table 12. Average household size of districts in adult equivalent
District/Agro-ecology N Mean Std. Dev. Minimum Maximum
Lowland 83 4.18 1.85 0.75 10.75
Midland 133 4.23 1.73 1.00 9.25
Highland 165 4.09 1.71 0.50 11.50
Total 381 4.16 1.74 0.50 11.50
Source: Computation from own data
4.1.4. Land use types
Land is a basic asset for rural farm households. Their livelihoods directly depend on their
land. Land holding size affects both crop and animal productions. Households with large land
holding sizes have better chance of crop and animal production, and generate income for
family needs. Therefore, the percentage of farmers in different land holding sizes helps to
compare the relative land shortage problems among the study districts. Table 13 shows the
number and percents of households in different land holding class.
Table 13. Households in different land holding class (%)
Agro-ecology
(District)
Land holding size in hectares
< 1 1 to 2 2 to 4 > 4 Total
Lowland 16.8 32.5 32.5 18.2 83
Midland 12.8 26.3 54.9 6.0 133
Highland 9.1 15.2 43.6 32.1 165
Total 12.1 22.8 45.1 20 381
Source: Computation from own data
79
In all districts, larger proportion of households own land holding size ranging from 2 to 4
hectares. The household percentage in lowland, midland, and highland districts in this land
holding class was about 33, 55 and 44 percents, respectively. The highest percentage of
households (32.1 percent) was found in highland district, and they own land holding size
greater than 4 hectares.
Table 14 depicts mean area by land use types of sampled households of the three districts.
Average privately owned land was relatively higher in highland district (3.21 ha). The average
holding size for the three districts was 2.65 hectares. The average agricultural land holding
size of midland (1.87 ha) was the lowest compared to the other two districts. Highland district
was better in average land holding sizes in all land use types. Midland was the least in the
average sizes of all land use types except grazing land (0.13 ha) which was higher than that of
lowland district (0.05 ha). The maximum of total land holding (12 ha) and agricultural land
holding (8 ha) were reported in highland district, and the minimum landholding size (0.25ha)
and agricultural land size (0.15 ha) were noticed in midland district (Appendix Table 5).
Generally, the highland district is endowed with larger land that is used for different purposes
as compared to the other two districts.
Table 14. Average land use types of sampled households in hectares
Land use type
Lowland
n=83
Midland
n=133
Highland
n=165
Total
N=381
Land owned 2.29 2.19 3.21 2.65
Agricultural land 2.02 1.87 2.23 2.06
Grazing land 0.05 0.13 0.73 0.37
Woodlot 0.01 0.00 0.02 0.01
Homestead area 0.23 0.16 0.25 0.21
Source: Computation from own data.
Agriculture is the major occupation for the rural households. The agricultural activities
include both crop and animal production. However, the focus of farmers on these activities
80
varies from one agro-ecology to another. For instance, in midland agro-ecology, relatively
more percentages of households use crop alone as main occupation. Main occupation is the
primary work for farmers on which they spend much of their work time, and serves as a
means of livelihood.
Table 15 depicts that the majority of household heads (85 percent) were involved in both crop
and animal production as their main activities. Crop production was the main occupation for
14.7 percent of total sampled household heads. Similarly, about 97.6 percent, 60.2 percent and
98.8 percent of the famers in low, mid and highland districts, respectively, involved in crop
and animal production as their main occupation. Relatively, crop production was the main
occupation of farmers in midland district on which about 40 percent of sampled household
heads live.
Table 15. Distribution of household heads by occupation (%)
Main occupation Lowland Midland Highland Total
Crop production 1.2 39.8 1.2 14.7
Crop & animal productions
97.6
60.2
98.8
85.0
Government job 1.2 0.0 0.0 0.3
Source: Computation from own data
4.1.5. Crop production
This section provides area, production and yield of major crops of the study districts. The
major agricultural crops produced in the study districts are wheat, barley, faba bean, field
peas, tef, maize and potato. Onion was also planted by some sampled households in the
districts in the 2012/13 cropping season. The average agricultural land allotted to wheat in
2012/13 cropping season was 1.1 hectares (Table 16). Barley (malt and food barley) is the
second dominant crop with average area of 0.69 hectares. However, malt barley alone
accounted for average area of 0.8 hectares. Other crops like tef, faba bean and field peas were
also essential crops with average planted land of 0.5, 0.3 and 0.3 hectares respectively. Table
81
16 also shows that wheat was the major crop with average planted land of about 1.6 hectares
both in lowland and midland districts. But in highland district, malt and food barley were the
major crops followed by wheat in terms of land planted with these crops.
Table 16. Average area cultivated (ha) and yield (q/ha) for major crops
Lowland Midland Highland Total
Crops Area Yield Area Yield Area Yield Area Yield
Wheat 1.6 15.6 1.6 30.9 0.5 24.8 1.1 24.9
Malt barley 0.5 12.0 0.0 0.0 0.8 29.2 0.8 29.1
Food barley 0.6 19.1 0.3 22.0 0.7 33.8 0.6 27.5
Faba bean 0.1 8.0 0.3 22.1 0.4 22.0 0.3 22.0
Field pea 0.2 10.2 0.2 14.6 0.4 19.1 0.3 16.5
Tef 0.6 8.7 0.2 9.0 0.3 12.0 0.5 8.8
Maize 0.3 11.2 0.2 19.1 0.5 40.0 0.2 17.0
Potato 0.1 24.0 0.2 115.5 0.3 96.6 0.2 105.4
Cabbage 0.0 0.0 0.2 152.2 0.1 181.0 0.1 166.6
Onion 0.2 54.7 0.3 88.1 0.0 0.0 0.2 74.4
Source: Computation from own data
Average yield was the highest for both malt barley (29 q/ha) and food barley (27.5 q/ha). Faba
bean and field pea‘s yield per hectare were 22 and 16.5 quintals, respectively. Relatively, the
least yield was reported for tef (8.8q/ha). Malt barley was mainly planted by farmers in
highland district. Food barley was planted in the three districts and it had average yield of
about 19, 22 and 34 qt/ha in lowland, midland and highland agro-ecologies, respectively. The
yield of barley was highest (33.8qt/ha) in highland. But yield of wheat was highest (30.9qt/ha)
in midland district. Generally, the yield of almost all crops was relatively better in midland
and highland than in lowland district of Dodota.
The average yield of wheat for sampled study area was 24.9q/ha. The average yield of the
study areas was almost equal to the yield of the highland area (24.8q/ha) but greater than the
low land yield (15.6q/ha). The midland (Hetosa) yield (30.9q/ha) was higher than the lowland
82
and highland yield levels. Comparison of the mean yields using F-test shows that there was
statistically significant mean yield difference at 1 percent significance level among the
districts with significant F statistic of 152.62 and prob > F = 0.0000 (Appendix Table 14).
Oromia regional average wheat yield for 2012/13 cropping season was 23.2q/ha, and the
national average yield was 21.10q/ha (CSA, 2013). This depicts that average yield of wheat
for the study areas were greater than the national and regional yield levels. However, the yield
level of the lowland district (Dodota) was lower than the national, regional and zonal average
yield levels. Similarly, the average yield of the midland district (Hetosa) was higher than the
national, regional, Arsi zone, and sample study areas‘ average yield levels (Figure 3). This
shows that midland district is a potential wheat producing area both at national and regional
levels in the country. Generally, it can be noted from Figure 3 that the average wheat yield for
sample study areas (24.9q/ha) was equal to the CSA yield estimate for Arsi zone (24.45q/ha),
and this yield was higher than the national and regional average yield levels of the same
2012/13 cropping season.
Figure 3. Comparisons of wheat yield at different levels for 2012/13 cropping season
Source: CSA, 2013 and districts‘ values from own survey
0 5 10 15 20 25 30 35
Ethiopia
Oromia
Arsi zone
Study
districts
Dodota
Hetosa
Lemu-
Bilbilo
Yield (q/ha)
Yield (q/ha)
83
4.1.6. Livestock ownership
Livestock rearing is also one of the major agricultural activities of farm households of the
study districts. Sales of animals and animal products‘ are the major sources of cash income to
farm households. Oxen are mainly used for traction power in farming. Animal manure is used
as fertilizer for crop production, and dried dung is used as source of energy for cooking.
Therefore, livestock helps in farm operations and source of income and contribute to the
improvement of production efficiency. On average, a household owns 1.81 and 2.87 cows and
oxen, respectively. Livestock size was converted into Tropical Livestock Unit (TLU) based
on the conversion factor given by Jahnke (1982). The result shows that a household owns on
average 7.28 TLU. Households in highland district own relatively larger livestock number in
TLU. Average holding sizes of goat and donkey were lowest compared to the other two
districts. Midland district ranked second in average livestock holding sizes though the average
holding sizes of goats, sheep and chicken were lower than lowland‘s average holding sizes.
Table 17. Average livestock number per households
Livestock type Lowland Midland Highland Total
Cows 0.76 1.18 2.85 1.81
Oxen 2.33 2.71 3.27 2.87
Bulls 0.35 0.45 1.25 0.77
Heifer 0.31 0.67 1.70 1.04
Calves 0.25 0.48 1.31 0.79
Goats 0.98 0.40 0.73 0.67
Sheep 3.89 2.72 11.25 6.67
Horses 0.23 0.12 1.96 0.94
Donkeys 1.30 1.65 0.97 1.28
Chicken 6.37 4.73 6.75 5.96
Total TLU 4.18 5.13 10.58 7.28
Source: Own survey data. TLU = Tropical Livestock Unit
84
Average holding size of horses was higher in highland (1.96) followed by lowland (0.23)
district. Equines are useful for transportation of goods, agricultural produce and humans. The
test for mean difference of livestock holding in TLU shows statistically significant mean
difference among agro-ecologies with test statistic (F) value of 67.54 and p-value of 0.0000
(Appendix Table 14).
4.1.7. Household income
As stated earlier, agriculture is the major source of income for the households. However,
some sampled households involved in subsidiary occupations like off-farm and wage work
including farming to generate additional income. Sales of firewood and remittances were also
sources of cash income to some of the sampled households. Households were interviewed to
estimate their annual income from different sources other than crop. The estimation was made
for each source on monthly basis and finally 12 months estimates were summed up. Crop
income was also included in order to see the relative importance of each income sources to
the households. Table 18 presents income sources of farm households.
Table 18. Average annual household income in ETB
Income sources Lowland Midland Highland Total
Off-farm activities 1,318.80 949.62 1,066.33 1,080.59
Wage work 191.93 156.39 302.06 227.22
Sale of livestock 2,668.59 1,765.64 6,051.20 3,818.30
Sale of livestock products 661.01 146.05 1,760.05 957.21
Sale of firewood 18.07 0.00 1,216.76 530.88
Remittance 1,216.87 1,622.11 453.33 1,027.66
Total non crop 5,414.26 4,639.81 10,849.73 7,641.86
Crop (gross) 27,432.95 41,815.78 50,134.27 42,285.01
Source: Computation from own data
85
The average annual income generated from sales of livestock was the highest in all districts.
Households in highland district had the maximum average annual income from sale of
livestock (ETB 6,051.20). Off-farm activities and remittances were also considerable sources
of income for households in all districts. The figures in table 18 also show that there were
great disparities among farm households in annual income generation from the listed income
sources. It can be noted that the variability of each income sources was very high. There were
sample households with very low or even zero annual income generation from some or all of
the income sources. When average gross income from crop production is compared to the non
crop annual income, average gross crop income is higher than the non crop income in all
agro-ecologies indicating that crop production is the major source of income for farm
households in the study area. The highest average annual gross income from crop production
was found in highland agro-ecology, implying that barley and pulse crops are the major
sources of income since these crops are the dominant ones in the highland agro-ecology.
4.1.8. Wheat production and adoption of farming practices
4.1.8.1. Yield, cost, returns and sales of wheat produce
Table 19 and 20 show allocation of farmland to wheat production and yield.
Table 19. Households in different wheat farm sizes (%)
Agro-ecology
Wheat area in hectares Total
N <1 1-2 > 2
Lowland 30 40 30 83
Midland 24 49 27 133
Highland 90 9 1 165
Total sample 54 29 17 381
Source: Computation from own data
86
Wheat is the most widely grown food and cash crop of the study districts. However, its yield
varies from district to district. Besides, land allocation for wheat production varied from
district to district during the planting season. The highest percentage of sample households
(90 percent) planted wheat on less than one hectare in 2012/13 cropping season in highland
district. Only 24 percent of households in midland and 30 percent of households in lowland
districts allotted land area less than one hectare to wheat production in the same cropping
season. Land area ranging from one to two hectares was allotted by 49 percent of sample
households in midland district (Table 19). The table depicts that the highest percentage of
sample households of both lowland and midland had wheat farm sizes of greater than one
hectare. The overall percentage of farmers with wheat farm size less than one hectare was 54
percent; and that with greater than one hectare was 46 percent.
As shown in Table 20, the average cultivated wheat area was almost the same (1.6ha) in low
and midland areas. But wheat yield was different in the two areas. Yield was about 16 and 31
quintals per hectare in lowland and midland districts respectively. The average yield of
midland district was almost equal to double of average yield of lowland district.
Table 20. Average cultivated area, output, sale and yield of wheat in 2012/13 season
Average
Lowland
n=83
Midland
n=133
Highland
n=165
Total
N=381
Cultivated area (ha) 1.58 1.59 0.50 1.12
Production (quintal) 25.45 50.19 12.43 28.44
Yield (qt/ha) 15.63 30.89 24.84 24.95
Quantity sold (quintal) 12.12 26.06 4.71 13.78
Source: Computation from own data.
However, the yield of the highland area i.e. Lemu-Bilbilo (25q/ha) was higher than the
lowland (Dodota) yield (16q/ha). On average, about 12, 26 and 5 quintals of wheat output per
87
household was sold in lowland, midland and highland districts, respectively. The average sale
was about 48, 52 and 38 percents of per household average wheat output in lowland, midland
and highland, respectively. The highest output sale proportion was observed among farm
households of midland district. The one-way analysis of variance for mean difference among
the districts shows significant mean difference among districts with significant F-value at one
percent level of significance for cultivated area, output, quantity sold, and yield (Appendix
Table 14).
Farm inputs used for wheat production and wheat output were quantified. Prices of inputs and
output were used to estimate the total costs and returns per hectare. The inputs include land,
labor for various farm operations, and other variable inputs. Details are indicted in survey
questionnaire (Appendix B). Table 21 presents cost of wheat production and the net returns
per hectare. The average cost of wheat production ranges from ETB 6,807.89 to ETB
16,930.54 per hectare with an average of ETB 10,406.53. Average profit was ETB 1,083.89;
8,039.89 and 4, 547.29 per hectare for lowland, midland and highland districts, respectively.
Table 21. Average cost, income and profit of wheat production (birr/ha)
Variables
Lowland Midland Highland Total
n = 83 n = 133 n = 165 N = 381
Mean Mean Mean Mean
(Std.Dev) (Std.Dev) (Std.Dev) (Std.Dev)
Revenue
10,233.89 19,250.75 14,937.57 15,418.54
(3146.59) (4384.89) (3864.74) (5,135.50)
Cost
9,149.99 11,210.86 10,390.28 10,406.53
(1324.28) (1369.91) (1444.1) (1,581.70)
Profit
1,083.91 8,039.89 4,547.29 5,012.00
(3351.31) (4236.94) (3948.35) (4,696.84)
Source: Computation from own data, and figures in parentheses are standard deviations
88
The highest average profit was earned in midland, and the lowest profit was obtained in
lowland district (Table 21). Average cost was the highest in midland and the lowest in
lowland district. This indicates that more cost of production was associated with more returns
or profit. Increasing farm input utilizations will increase farm profit. Therefore, the disparity
in returns among districts could be due to variations in input utilizations, improved agronomic
practices and efficiency in production. There was statistically significant difference in average
profit among districts with F-value of 81.96 and p-value of 0.0000 (Appendix Table 14).
Bartlett‘s test also gave significantly different variance of profit among districts (Chi-square =
5.309, and prob > chi-square = 0.07).
Production of crops including wheat for market and the subsequent commercialization of
smallholder farmers is viewed by the government as the main factor for productivity
enhancement in Ethiopia. The proportion of output sold out of the total at the farm level
indicates the level of commercialization of production. Table 22 shows that the largest
proportion of households (46%) sold 26 to 50 percents of wheat output. The highest
percentage of farmers in highland (24percent) did not sell any wheat product. This might be
due to relatively more output of barley in the district. Overall, 12 percent of sample
households produced wheat only for consumption.
Table 22. Percentage of households that sold wheat output
Proportion of output sold (%)
Lowland
n = 83
Midland
n = 133
Highland
n= 165
Total
N = 381
0 4 4 24 12
1 to 25 13 11 13 12
26 to 50 53 44 44 46
51 to 75 24 38 16 25
76 to 95 5 4 3 4
96 to 100 1 0 0 0
Total 100 100 100 100
Source: Computation from own data
89
However, as shown in Table 22, 81percent of the sample farmers in highland sold less than
or equal to 50 percent of their wheat output, whereas in lowland and midland districts, 70
percent and 59 percent of the farmers sold less or equal 50 percent of the output, respectively.
This is because of more barley output and less wheat output in the highland, and more wheat
output in midland areas. If output is less, it is usually consumed at home by households and
little or no excess is left for sale.
4.1.8.2. Wheat row planting and crop rotation
Row planting
Agricultural research and extension offices have been working to increase wheat yield.
Improved farming practices are recommended and promoted through research and extension
systems. These improved inputs and farming practices need to be adopted by the end users to
have meaningful impact on the level of production and productivity and thereby improve the
living standards of the users. Planting wheat in-row is one of innovative farming practices in
Ethiopia that has been promoted by agricultural extension activities in recent years. It has
been practiced in various parts of the country.
Table 23. Proportion of household heads that were aware of the usefulness of wheat row
planting (%)
Awareness Lowland Midland Highland Total
No
8.4
3.0
17.0
10.2
Yes 91.6 97.0 83.0 89.8
Total (N) 83 133 165 381
Pearson chi2 (2) = 15.999, pr = 0.000
Source: Computation from own data
Table 23 shows percentage of farm households that were aware of the usefulness of planting
wheat in-row through agricultural extension services. About 90 percent of the sample farm
90
households are aware of plating wheat in-row. The highest percentage of the sample farmers
knew the usefulness of planting of wheat in-row in all study districts. The percentage of
sampled farmers who practiced planting wheat in row was 33.1 percent. Wheat row planters
were 41, 19.5 and 40 percents in lowland, midland and highland respectively (Appendix
Table 6). The chi-square test for independence of awareness of row planting and agro-ecology
gives significant statistic (15.999) at 1% level of significance, showing there is association
between awareness of row planting and agro-ecology. Yield data were collected for wheat
grown with row planting and broadcast methods. Figure 4 and Appendix Table 7 display
comparisons of average yield of the two planting methods. The average yield differences were
1.6, 11.1 and 4.6 quintals per hectare for row planting and broadcast methods in lowland,
midland and highland districts, respectively. Overall, the average yield difference was 3.1
quintal per hectare between the two planting methods in the study districts. This indicates the
yield of planting wheat in-row was higher than planting by broadcast method in all study
areas; and the average yield gain from row planting was 12.53 percent for whole sample size.
Figure 4. Average yield of different methods of planting
Source: Computation from own survey data.
0.00
5.00
10.00
15.00
20.00
25.00
30.00
35.00
40.00
45.00
Dodota Hetosa Lemu-Bilbilo Total
15.74
30.14
23.3524.73
17.37
41.23
27.95 27.83
Aver
age
pro
duct
ivit
y(q
t/ha)
Districts (Agro-ecology)
Broadcast
In-row
91
But the yield gain due to row planting in lowland, midland and highland districts was about
10, 37 and 20 percents, respectively (Figure 4 and Appendix Table 7). The significance of the
difference between the mean yields of the two methods of wheat planting was judged through
analysis of variance (ANOVA). Table 24 gives the results of one-way analysis of variance for
the study area as well as for each district. It can be noted from the table that wheat mean yield
of row and broadcast planting methods was significantly different at 1 percent level of
significance for the whole study area. However, when the mean yield variance was tested for
each district separately, lowland district showed non-significance of yield difference at 1
percent significance level between the two planting methods. The mean yield difference
between row and broadcast planting in midland and highland districts were statistically
significant at P<0.01 (Table 24).
However, placing wheat seed in-row alone might have not been a factor for yield advantage
over broadcast planting method. There was strong promotion and agricultural extension
services on wheat row planting in the study area. As a result, most farmers were motivated
and used recommended seed and fertilizer rates. Farmers added fertilizer along with seed in-
row rather than broadcast, and this could help plant growth. Moreover, agronomic practices
like early hand weeding and hoeing were practiced by most farmers of wheat planted in-row.
Generally, an integrated use of appropriate input levels and agronomic and management
practices could be the contributing factors for higher yield of wheat planted in-row.
92
Table 24. Analysis of variance for mean difference between planting methods
Study
area
Analysis of Variance
Source SS df MS F Prob > F
All
Between methods 809.83 1 809.83 9.38* 0.0023
Within methods 32716.46 379 86.32
Total 33526.30 380 88.23
Bartlett's test for equal
variances: chi2(1)= 8.89 Prob>chi2= 0.003
Lowland
Between methods 52.91 1 52.91 1.79 0.185
Within methods 2398.25 81 29.61
Total 2451.17 82 29.89
Bartlett's test for equal
variances: chi2(1)= 0.39 Prob>chi2 = 0.529
Midland
Between methods 2575.18 1 2575.2 39.3* 0.0000
Within methods 8594.54 131 65.61
Total 11169.72 132 84.62
Bartlett's test for equal
variances: chi2(1)=2.813 Prob>chi2 = 0.093
Highland
Between methods 837.82 1 837.82 22.5* 0.0000
Within methods 6061.38 163 37.19
Total 6899.20 164 42.07
Bartlett's test for equal
variances:
chi2(1) =
0.0292 Prob>chi2 = 0.864
*p < 0.01
93
Crop rotation
Crop rotation is another agronomic practice that has impact on the level of crop yield. It is the
sequence of crops grown in succession on a particular field for the purpose of optimizing
yield. In areas where farm input utilization, especially chemical fertilizers and pesticides, is
very low due to un-affordable high input prices, choice of crop rotation could affect
productivity of a given crop. In view of this, this study attempted to analyze famers‘ choices
of precursor crop to wheat cultivation. The type of choice of precursor crop i.e. crop that
precedes wheat could have impact on wheat yield by affecting nitrogen supply, soil structure
and organic matter, and by decreasing or increasing disease and weed competition. The major
crop categories that are cultivated in the study districts include pulses, vegetables, and cereal
crops (Table 16). Proper sequence of these crops for rotation can improve soil fertility. A
leguminous crop usually precedes cereals to increase cereal crop yield. Table 25 shows
percentage of sampled farmers who planted pulse, vegetable or cereal crop in 2011/12
cropping season on the farm planted with wheat in 2012/13 cropping season.
Table 25. Proportion of households with different choices of precursor crop to wheat planting
(%)
Precursor crop Lowland Midland Highland Total
Pulses 14.5 35.3 43.0 34.1
Vegetable 2.4 12.8 6.7 7.9
Cereal 83.1 51.9 50.3 58.0
Pearson chi2 (4) = 32.4053, Pr = 0.000
Source: Computation from own survey data
Results presented in Table 25 shows that 58 percent of farmers planted cereals (i.e. wheat,
barley, tef or other cereal) before wheat. About 34 and 8 percent planted pulses and
vegetables respectively preceding wheat planting. This was an indication of cereal mono-
cropping system where the majority of farmers were practicing planting of cereal after cereal.
Similarly, about 83, 52 and 50 percents of farmers chose cereals as precursor crop to wheat
planting in lowland, midland and highland districts, respectively. Cereal before wheat was a
94
choice for majority of farmers in all districts, and followed by pulses. The chi-square test of
the assumption of no association between choice of precursor crop and agro-ecology shows
larger value of chi-square (32.41) with p-value of 0.000. That is, the data suggest that there is
an association between choice of precursor crop and agro-ecological location of households.
Generally, use of leguminous crop (pulses) which is believed to improve soil fertility was
low; and cereals were mainly used as a precursor to wheat planting in all study districts.
Relatively, use of pulses was higher in highland district (43 percent of sample farmers) as
precursor to wheat planting.
Similarly, the farm land on which wheat was planted in 2012/13 cropping season was used for
planting pulse and cereal crops by 29.9 percent and 68.8 percent of sample farmers in 2010/11
cropping season, respectively (Table 26). About 34 and 58 percent of sample farmers planted
pulse and cereal crops on the same wheat farm in 2011/12 cropping season, respectively. This
shows that majority of farmers planted cereal after cereal at least for three consecutive
cropping seasons on the same farm (including wheat planted in 2012/13 cropping season). It
is well known fact that breaking cereal mono-cropping by crop rotation helps resource poor
farmers to improve soil fertility, reduce plant diseases and pests and thereby increase yield of
crops.
Table 26. Proportion of households by what they planted on wheat farm before 2012/13 (%)
Crop type 2010/11 season 2011/12 season (precursor)
Pulses 29.9 34.1
Vegetables 1.3 7.9
Cereals 68.8 58.0
Source: Computation from own survey data
Even though the major crop categories are cereals, pulses and vegetables, farmers plant
different types of crop varieties on different or adjacent plot of lands. Number of crop types
planted could indicate level of land fragmentation and crop rotation. But planting cereal after
another cereal crop, for example wheat after barley or tef, is not a proper sequence of rotation.
Table 27 shows average number of different crops planted by a household in 2012/13 crop
95
season. On average, a household planted about four different types of crops with minimum of
one crop which was wheat and maximum of ten crop types. The crops included different types
of cereals, pulses and vegetable crops. One way analysis of variance for Bartlett‘s test for
equal variance gives significant chi-square statistic (30.12) at 1 percent significance level
(prob > chi2 = 0.000), implying that there is significant variability in the number of crop types
planted by households in the study districts (Appendix Table 14).
Table 27. Number of crop types planted by a household in 2012/13
Agro-ecology N Mean Std. deviation Minimum Maximum
Lowland 83 4.2 1.5 1 9
Midland 133 4.4 1.7 1 10
Highland 165 4.4 1.1 1 7
Total 381 4.3 1.4 1 10
Source: Computation from own survey data
Though we observe differences in the number of crop types planted by households within a
district (shown by the standard deviations), the mean difference of crops types planted among
the district was not significant as indicated by F-value of 0.26 and prob > F equal to 0.7726
(Appendix Table 14).
Table 28 shows percentage of sample households practicing crop rotation especially the use of
pulse crops and having access to services of various farm inputs. The highest percentage of
households (87.9) adopted crop rotation in highland district. About 28 and 32 percents of
households in lowland and midland districts adopted the use of pulse crops in their farming
practices. The table also shows that access to agricultural credit and extension services were
relatively highest in lowland district. Similarly, access to the required chemical fertilizers and
pesticides were relatively highest in highland district. However, only 28.5 percent of sample
households in highland district had access to improved wheat seed. The test statistic of the
null hypothesis of no association between a dummy variable and agro-ecology is found to be
significant at 1 percent for all variables, suggesting that there is an association between the
dummy variables and agro-ecological orientation of households.
96
Table 28. Households practicing crop rotation and having input services (%)
Dummy
Variables
Lowland
n= 83
Midland
n=133
Highland
n= 165
Total
N=381
Chi-
square
p-value
Rotation 27.7 31.6 87.9 55.1 126.58* 0.000
Credit 54.2 29.3 18.2 29.9 34.23* 0.000
Extension 98.8 91.0 86.7 90.8 9.74* 0.008
Fertilizers 96.4 75.9 96.4 89.2 38.02* 0.000
Pesticides 15.7 31.6 57.6 39.4 45.84* 0.000
Seed 80.7 60.2 28.5 50.9 67.26* 0.000
Source: Computation from own data. *significant at 1% (p < 0.01)
4.2. Wheat Production Efficiency
4.2.1. Technical efficiency
Production efficiency is one of the methods of farm performance measurement. It can be used
for making comparisons among farms over time or across geographical regions or agro-
ecologies. In the present study, technical efficiency analysis was used to measure farm
households‘ wheat production efficiency and make comparisons among the selected agro-
ecological orientations of farm households. It measured the physical relationship between
output and inputs used in wheat production at the given level of technology.
Wheat production of sample famers was represented by a Cobb-Douglas Stochastic Frontier
Model, and half-normal distribution of inefficiency (Model equation 8). Because, a series of
preliminary likelihood ratio tests revealed that Cobb-Douglas stochastic frontier model
(Equation 8) best fit the data given a more flexible translog frontier model, and the
distribution of inefficiency best represented by the half-normal distribution. Furthermore, the
self dual nature of this production function and its cost function provides a computational
advantage in obtaining estimates of technical and allocative efficiency. The natural logarithms
of the data on the variables were taken for efficiency analysis. Definitions of inputs, output
and inefficiency variables have been given under methodology chapter in Table 6.
97
Table 29. Descriptive statistics of input and output variables used in efficiency analysis
Variables
Lowland Midland Highland Total
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.
Output (kg/ha) 1562.95 477.68 3089.42 707.06 2484.49 618.83 2494.90 837.64
Area (ha) 1.58 1.32 1.59 1.18 0.50 0.33 1.12 1.10
Labor (Man-days/ ha) 41.20 25.74 54.89 25.89 98.00 43.68 70.58 42.40
Fertilizers (kg/ha) 74.33 29.88 90.36 27.21 115.65 36.08 97.82 35.92
Seed & pesticide quantity (kg/ha) 766.02 284.30 937.84 412.89 554.72 300.84 734.49 380.00
Output value (ETB/ha) 10233.89 3146.59 19250.75 4384.89 14937.57 1364.74 15418.54 5135.50
Source: Computation from own data.
Table 29 and Table 30 present descriptive statistics of input, output and inefficiency variables used in the analyses. The natural
logarithms of the values of output and input variables were used in the Cobb-Duglas Stochastic Frontier analysis. Comparisons of
mean difference for all variables among agro-ecologies through one-way analysis of variance show significant mean difference
among agro-ecologies at 1 percent (prob > F = 0.0000) for all variables (Appendix Table 14).
98
Table 30. Descriptive statistics of inefficiency variables
Variables
Lowland Midland Highland Total
Mean St.Dev Mean St.Dev Mean St.Dev Mean St. Dev
Continuous variables:
Age of Household head (yr) 43.34 10.46 48.80 10.63 46.52 11.80 46.62 11.27
Education of household head(level) 5.41 3.36 4.56 4.16 4.69 3.39 4.80 3.68
Livestock ownership (TLU) 4.18 2.71 5.13 3.68 10.58 6.34 7.28 5.66
Planted crop number 4.24 1.50 4.36 1.68 4.37 1.07 4.34 1.40
Household size in adult equivalent 4.18 1.85 4.23 1.73 4.09 1.71 4.16 1.74
Farming experience (yr) 22.47 10.40 27.87 10.64 25.58 11.44 25.70 11.09
Off-farm income (in ‗1000‘ ETB/yr) 6.08 5.05 4.64 6.14 10.85 11.00 7.64 8.88
Discrete variables (%)
Crop rotation (%)
No 72.29
68.42
12.12
44.88
Yes 27.71
31.58
87.88
55.12
Access to seed (%)
No 19.28
39.85
71.52
49.08
Yes 80.72
60.15
28.48
50.92
Row planting (%)
No 57.83
54.14
59.39
57.22
Yes 42.17
45.86
40.61
42.78
Credit (%)
No 45.78
70.68
81.82
70.08
Yes 54.22 29.32 18.18 29.92
Source: Computation from own survey data.
99
Some inefficiency variables like access to farmers‘ marital status and sex of household head
have shown small variations among households. For example, most household heads were
married and male headed. As a result, these variables were not included as inefficiency
variables in the frontier model.
The likelihood ratio test was used to identify the functional form of production function which
properly fit the data. The test result revealed that Cobb-Douglas Stochastic Frontier model
best fit the data compared to the more flexible translog frontier model. The existence of
inefficiency factor was also tested by Wald test. The null hypothesis was that no systematic
inefficiency in the distribution. However, the test result showed significant chi-squared
statistic for the study areas (wald chi2 = 52.1, prob > chi2 = 0.0000), implying rejection of the
null hypothesis. This means that there was systematic inefficiency in the distribution. Based
on this, the parameters of Cobb-Douglas stochastic production model was estimated with the
maximum likelihood estimates. Table 31 shows the coefficient of land, labor, fertilizer and
seed and pesticides of stochastic frontier model of Cobb-Douglas production function. The
signs of all the slope coefficients of the production function are positive and significant. This
implies that all inputs (land, labor, fertilizers, seed and pesticides) have turned out to be
significant in determining wheat output; that is, wheat output is responsive to inputs
utilization. The coefficients associated with the inputs measure the partial elasticity of output
with respect to the respective inputs. The elasticity of chemical fertilizer was the highest
(0.229) for the study area, indicating that there was relatively more proportionate change in
output due to proportionate change in amount of chemical fertilizer use in kilograms.
The sum of elasticities of the four inputs (land, labor, fertilizers and seed and pesticides) is
0.616 i.e. scale elasticity is less than one. This indicates that the wheat production function
exhibits decreasing returns to scale that represents the second stage economic region of
production function. It means that proportionate increase in all inputs results in a less than
proportionate increase in wheat output. The maximum likelihood estimate of the model for
each agro-ecology shows that wheat output elasticities associated with land, labor, chemical
fertilizers and other inputs (seed and pesticides) were positive and significant in lowland
district. The elasticity of output due to other inputs was the highest (0.247), followed by
100
elasticity of output due to labor (0.211) in lowland district. In midland district, elasticities of
output due to land and fertilizers were positive and significant, with the highest being
elasticity of output due to chemical fertilizers (0.163). Similarly, the elasticities associated
with land and fertilizers were significant in highland district with negative elasticity of output
due to land. This might be due to more suitability of the highland agro-ecology to barley
production, and more land allocation to barley production was observed in the highland
district. Relatively, there was less cultivated wheat area in the highland (Table 19). This might
have led to negative proportionate change of output due to proportionate change in the land
area planted to wheat.
Table 31. Estimates of elasticities of output for each agro-ecology
Variable
Maximum Likelihood Estimates (Standard error)
Lowland
(n = 83)
Midland
(n= 133)
Highland
(n = 165)
Total
N=381
Constant
4.316***
(0.726)
7.045***
(0.358)
6.955***
(.489)
5.688***
(0.418)
ln (land)
0.192***
(0.065)
0.086**
(0.037)
-0.586*
(0.033)
0.124***
(0.026)
ln (labor)
0.211**
(0.086)
0.059
(0.046)
-0.011
(0.041)
0.132**
(0.041)
ln (fertilizers)
0.182**
(0.084)
0.163***
(0.046)
0.163**
(0.076)
0.229***
(0.048)
ln (other inputs)
0.247***
(0.069)
0.024
(0.028)
0.040
(0.041)
0.131***
(0.032)
Wald χ2 statistic
25.78***
19.27***
8.89*
52.01**
Log-likelihood
-6.195
37.3
15.269
-87.99
***p< 0.01; **p< 0.05; *P< 0.10, and figures in parentheses are standard errors
101
The mean technical efficiency was 75 percent for the whole study area with minimum and
maximum technical efficiency of about 24 and 94 percents respectively. The mean technical
efficiency estimates for lowland, midland and highland districts were 57, 82, and 78 percents,
respectively. Given the current state of technology and input levels, there was a scope of
increasing wheat output by up to 25 percents on average. However, on average, the scope of
wheat output increment in lowland, midland, and highland districts were about 43, 18, and 22
percents respectively. The technical efficiency ranges from 24 to 87 percent in the lowland,
52 to 94 percents in the midland, and 35 to 94 in the highland agro-ecologies. The results
show that there is significant technical efficiency difference in wheat production among and
between agro-ecologies. The results are more or less consistent with what have been found by
Arega and Zeller, 2005; Arega, 2006; Kaleab and Brehanu, 2011 and Mesay et al., 2013 in
their smallholder or commercial farms production efficiency analyses carried out in the
country. The study indicates that improving technical efficiency in wheat production can
improve productivity of wheat in all agro-ecologies.
The mean technical efficiency difference among agro-ecologies was tested using one way
ANOVA. The response variable was technical efficiency scores and the factor variable was
district or agro-ecology. The test statistic (F-value) was 159.72 with prob > F equal to 0.0000
implying significance mean technical efficiency difference among districts. Similarly,
Bartlett‘s test of equal variance gave significant chi-square statistic of 19.29 with
corresponding p-value of 0.0000, implying rejection of the assumption that variances are
homogeneous. That is, the variances of the technical efficiency of agro-ecologies were
statistically different.
Table 32. Summary statistics of technical efficiency scores
Agro-ecology Obs Mean Std. Dev. Min Max
Total efficiency level 381 0.749 0.143 0.244 0.944
Lowland 83 0.569 0.126 0.244 0.886
Midland 133 0.820 0.083 0.516 0.944
Highland 165 0.784 0.111 0.345 0.943
Source: Computation from own survey data
102
Multiple comparison test of mean technical efficiency estimate was conducted between a pair
of agro-ecologies using Bonferroni normalization. The test result gave significant difference
of mean technical efficiency estimate between any paired two agro-ecologies at one percent of
significance level. It has to be noted that comparing the mean efficiency scores is misleading
if they are computed separately for different agro-ecologies. The scores only reflect the
dispersion of efficiencies within each agro-ecology. The scores say nothing about the
efficiency of one agro-ecology relative to the other. Therefore, for the purpose of comparison,
efficiency scores were estimated for combined sample and their distribution were sorted out
on agro-ecology basis. The percentage distribution of farm households in different technical
efficiency ranges has been given in Table 33. Thirty percent of households found within
technical efficiency range of 0 to 50 in lowland district, and 4 percent were in this range in
highland district. There was no farm household within 0 to 50 technical efficiency ranges in
midland district. The highest percentage of farm households (65 percent) was found in
efficiency range of 51 to 75 percent in lowland district, whereas the highest percent farmers
(46 percent) of midland district were found in efficiency range of 86 to 100 percent. In
highland district, 47 percent of farmers were within efficiency range of 76 to 85 percent.
Relatively, higher proportion of Households of midland district was found in higher percent
of technical efficiency range.
Table 33. Distribution of households in different technical efficiency ranges (%)
Technical efficiency range
(%)
Lowland
(n =83)
Midland
(n =133)
Highland
(n = 165)
Total
(n = 381)
0 - 25 1 0 0 0
26 - 50 29 0 4 8
51 - 75 65 17 22 30
76 - 85 2 38 47 34
86 - 100 2 46 27 28
Total 100 100 100 100
Source: Computation from own data
103
Figure 5 shows the distribution of households by their technical efficiency scores. The highest
density of households was found within higher efficiency scores, indicated by two vertical
lines at 0.75 and 0.90 technical efficiency scores.
4.2.2. Factors affecting technical efficiency
Various households‘ socioeconomic factors can affect the efficiency of wheat production.
These factors include age and educational level of household head, total livestock holding
size, household size, access to improved seed, and adoption of crop rotation and planting of
different types of crops, practice of wheat row planting and income generation from other
sources other than crop cultivation. Details of variable descriptions and their effects on
efficiency have been given under methodology chapter in Sub-section 3.4.1.
Using STATA version 11 and models in equations (6) and (8), the coefficients of inefficiency
variables were estimated along with the elasticities of wheat output with respect to inputs. In
0
1
2
3
4
Density
.2 .4 .6 .8 1 Technical efficiency scores of sampled households
kernel = epanechnikov, bandwidth = 0.0375
Figure 5. Kernel density estimate of study area
104
the model, output and inefficiency scores were dependent variables; and the independent
variables were farm inputs and the inefficiency variables. The existence of inefficiency factor
was tested by Wald test. The null hypothesis for the test was no systematic inefficiency in the
distribution. However, the test result showed significant chi-square statistic, depicting
rejection of the null hypothesis of no inefficiency. There was inefficiency factor in the
distribution, and the famers were inefficient in wheat production.
Table 34 shows that age of household head, livestock holding size, practice of crop rotation
and access to credit were significant factors at p <0.05, and practice of wheat row planting
was significant factor at p < 0.01 whereas household size and access to improved wheat seed
where significant at p < 0.10 level. Therefore, these significant variables were determinant
factors of households‘ wheat production efficiency. The signs of the coefficients reveal that
age, livestock holding size, practice of crop rotation and planting wheat in row were
negatively related to the deviation of observed output from the frontier. With increasing age, a
farmer usually becomes more experienced and has accumulated knowledge and resources to
overcome production constraints. This helps the farmer to improve his production and
productivity and thereby reduces the inefficiencies by decreasing deviation of output from the
optimal output represented by the frontier. More size of livestock is associated with
availability of traction power for plowing, threshing of harvested crop and transporting inputs
and outputs. Therefore, size of livestock increases the level of observed output to the frontier
and reduces its deviation from the frontier i.e. reduces inefficiencies. Similarly, wheat row
planting and crop rotation have yield improving effects and raise the level of observed output
of a household. This means, when the observed output increases the deviation between the
observed output and the output represent by the frontier decreases i.e. the percentage
deviation (measured by the coefficients of row planting and crop rotation variables) between
the two outputs decreases.
Likewise, the positive coefficients show that the percentage deviation between observed
output and the output represented by the frontier increases with the increase in the variables
with positive coefficients. The coefficients of access and use of credit indicates that credit
might not be used for wheat production but rather for other activities which in most cases for
105
petty trade and fattening of small ruminant animals in rural areas, and moreover these
activities compete for the available labor for wheat production. The significant positive
coefficient of access to improved seed would be due to the wrong perception the farmers have
for improved seed. That is, improved seed alone does not increase yield unless proper amount
of other inputs is used along with the improved seed. Most farmers perceive that improved
seed gives higher yield and therefore they do not focus on the use of other recommended
package of inputs that should be used along with the seed. This decreases the output and
increases inefficiency in production.
Table 34. Estimates of sources of technical inefficiency variables for total sample
Inefficiency
variables Coef. Std. Err. z P>z
Age -0.044** 0.021 -2.060 0.040
Education -0.004 0.030 -0.130 0.900
Family size 0.120* 0.068 1.760 0.079
Livestock size -0.063** 0.029 -2.200 0.028
Crop numbers 0.067 0.074 0.900 0.368
Access to seed 0.407* 0.217 1.880 0.060
Crop rotation -0.514** 0.211 -2.440 0.015
Income 0.020 0.016 1.300 0.193
Farming experience 0.035 0.022 1.590 0.112
Row planting -0.600*** 0.208 -2.890 0.004
Credit 0.416** 0.205 2.030 0.042
Constant -1.076 0.699 -1.540 0.124
sigma_v 0.201 0.023
Wald chi2 52.01***
Prob > chi2 0.0000
loglikelihood -87.99
*p< 0.10; **p< 0.05; ***p< 0.01
106
The test statistic for each agro-ecology gave significant chi-square at P < 0.01 level for
lowland and midland districts, and at P < 0.1 level for highland districts. This indicates the
presence of inefficiency in production. The sources of inefficiencies were attributed to some
socioeconomic variables which are found significant in Table 35. For instance, in highland
district, household size, practice of crop rotation and row planting and access to credit were
found to be significant factors in affecting the variation of output from the frontier in the
district. Similarly, access to improved seed, farming experience of farm household head, and
wheat row planting were significant factors influencing the efficiency of wheat production in
midland district.
Table 35. Estimates of inefficiency variables for each agro-ecology
Inefficiency
variables
Lowland Midland Highland
Coef St.err Coef St.err Coef St.err
Age -0.222** 0.106 -0.034 0.039 0.003 0.041
Education 0.007 0.009 0.093 0.066 -0.018 0.067
Household size 0.022 0.047 -0.211 0.147 0.353** 0.162
Livestock holding -0.023 0.038 -0.024 0.064 -0.098 0.064
Crop numbers 0.030 0.038 -0.490 0.129 0.144 0.209
Seed 0.793 1.845 1.851*** 0.572 -0.430 0.674
Rotation -1.060 1.227 -0.584 0.483 -1.283*** 0.490
Income 0.069 0.119 0.028 0.029 -0.023 0.035
Experience 0.132 0.113 0.109*** 0.042 0.000 0.042
Row planting -0.411 1.020 -0.993** 0.432 -1.962*** 0.652
Credit -1.305 1.275 -0.119 0.435 1.090** 0.449
Constant 1.783 3.070 -4.633 1.536 -2.502* 1.469
sigma_v 0.205 0.025 0.134 0.016 0.169 0.019
Wald χ2 24.85***
18.70***
8.39*
Prob > χ2 0.0001
0.0009
0.07
Log likelihood 1.580 37.42 18.25
*p < 0.10, **p < 0.05, *** p < 0.01
107
However, age of household head was significant inefficiency factor in the lowland district.
The difference in sources of inefficiency factor in lowland would be due the difference in the
agro-ecological setting of the district; and production efficiency might be more related to the
biophysical nature of the area than to the socioeconomic characteristics of households. The
study found that the sources of technical inefficiency of households are more or less similar to
the findings of most technical efficiency studies cited in literature review chapter (Solomon et
al., 2011; Team, 2011; Yu et al., 2011; Hassan et al., 2012 and Negassa et al., 2012). Age and
educational level of household head, farming experience of household head, household access
to improved seed and credit service, household and livestock holding sizes are the most
identified sources of production inefficiencies by several studies (Arega and Zeller, 2005;
Arega, 2006; Kaleab and Brehanu, 2011 and Mesay et al., 2013). Moreover, this study
indicates that crop rotation and row planting are the sources of technical inefficiency that are
found to be significant in determining the level of technical efficiency.
4.2.3. Allocative efficiency
Allocative efficiency also measures farm efficiency. To attain allocative efficiency, a firm or
farm should choose the optimal combination of inputs so that output is produced at a minimal
cost. To maximize profit from wheat production, farmers have to choose the best combination
of inputs given the prices of inputs and output. For the present study allocative efficiency was
estimated from a single output (wheat) four inputs Cobb-Douglas cost function (Equation 10).
The four inputs were land used for wheat production, labor utilization in production, chemical
fertilizers used in production and, seed and pesticides used in wheat production.
Cost of production was measured in birr per hectare; price of land was estimated based on the
rental value of land in birr per hectare in the respective study agro-ecology; and the value
varied from agro-ecology to agro-ecology based on the fertility of land and demand for rental
land. Daily wage rate per day was used to value labor, and price of fertilizer was in birr per
kilogram. Average price of seed and pesticides per kilogram was estimated based on the
proportionate weight of each input in total cost of production. All the data on each variable
were transformed into natural logarithms. Stochastic Frontier Cobb-Duglas cost function was
108
estimated by maximum likelihood method. Table 36 depicts descriptive statistics of the
dependent and independent variables used in Cobb-Douglas cost function. The F-tests for
mean difference among districts for all variables resulted in a significant mean difference at 1
percent level of significance (Appendix Table 14).
Table 36. Descriptive statistics of variables used in allocative efficiency
Variable
Lowland
n = 83
Midland
n = 133
Highland
n = 165
Total,
N = 381
Mean
Std.
Dev Mean
Std.
Dev Mean Std. Dev Mean
Std.
Dev
lncost 9.11 0.13 9.32 0.12 9.24 0.14 9.24 0.15
lnland 8.04 .03 8.23 0.02 7.98 .02 8.09 0.13
lnwage 3.79 0.17 3.61 0.11 3.32 0.14 3.52 0.23
lnfertprice 2.68 0.01 2.69 0.02 2.69 0.02 2.69 0.02
lnothprice 1.67 0.27 1.55 0.29 1.32 0.28 1.48 0.32
ln(output) 7.31 0.30 8.01 0.23 7.78 0.28 7.76 0.37
For mean comparison test, see Appendix Table 14.
Table 37 gives the maximum likelihood estimates of the coefficients of Cobb-Douglas
stochastic frontier cost function. Except wage of labor, all the coefficients are statistically
significant at 1 percent significance level. The wald test gives significant chi-square statistic
(167.21) and proves the rejection of the null hypothesis that all the coefficients except the
constant are equal to zero. That is, the effects of the coefficients are significantly different
from zero. The effects of prices of land, chemical fertilizers and output quantity were positive
on the cost of production per hectare. However, the effect of labor wage was negative but
insignificant, and the effect of prices of seed and pesticides was negative and significant. Seed
and pesticides are basic inputs in wheat production. When prices‘ of these inputs increase,
farmers tend to use less of them and allocate less of other inputs (land, fertilizers, etc) to
produce wheat. This reduces the total cost of wheat production per hectare.
109
Table 37. Maximum likelihood estimates of allocative efficiency for total sample
ln (cost) Coefficients Standard error z-value p >z
ln (land price) 0.31*** 0.062 4.740 0.000
ln(wage) -0.273 0.033 -0.830 0.406
ln (Fertilizer price) 1.216*** 0.291 4.180 0.000
ln (seed/pesticides) -0.064*** 0.023 -2.740 0.006
ln (output) 0.126*** 0.021 6.060 0.000
Constant 2.542*** 0.907 2.800 0.005
Log likelihood 254.240
Wald chi-squared 167.21***
Prob > chi-squared 0.0000
lnsig2v -4.969*** 0.191 -25.940 0.000
lnsig2u -3.702*** 0.187 -19.760 0.000
sigma_v 0.083 0.008
sigma_u 0.157 0.015
sigma2 0.031 0.004
lambda 1.883 0.021
Likelihod-ratio test of
sigma_u = 0: chibar2 17.19***, prob >= chibar2 = 0.000
*p < 0.10, **P < 0.05, ***p <0.01
The maximum likelihood estimation of the coefficients Cobb-Douglas stochastic cost function
shows that the coefficients (elasticites of cost due to change in prices of inputs) of seed and
pesticide, and output were statistically significant in their effects on the cost of production in
lowland district. That is, proportionate change in cost of production decreases due to
proportionate change in prices of seed and pesticides in lowland. It seems that households
tend to use less or no of these inputs if their prices increase. In midland, except the
coefficients of labor wage, all input prices of cost function were statistically significant. It is
only the effect of seed and pesticides price that was negatively related to the cost of
production in midland district. All other input prices had positive effects on the level of cost
of wheat production. This shows that farmers reduce or avoid the use of improved seed and
110
pesticides when their prices increase in the midland. It appears that with increasing prices of
improved seed, farmers tend to use local varieties and hence its effect on cost of production is
negative.
Table 38. Maximum likelihood estimates of allocative efficiency for each agro-ecology
ln (cost)
Lowland Midland Highland
Coef. Std. error Coef. Std. error Coef. Std. error
ln (land price) -0.238 0.726 1.295* 0.702 -1.515*** 0.581
ln(wage) -0.004 0.119 0.153 0.110 0.134* 0.074
ln (Fertilizer price) -1.160 1.209 2.337*** 0.452 0.665 0.442
ln (seed/pesticides) -0.110** 0.049 -0.111*** 0.039 0.041 0.042
ln (output) 0.137*** 0.047 0.124*** 0.042 0.056 0.036
Constant 13.201* 7.754 -9.139 6.150 18.48*** 4.383
Log likelihood 63.650
112.280
104.16
Wald chi-squared 20.99*** 46***
14.69**
Prob > chi-square 0.0008
0.0000
0.012
lnsig2v 6.018*** 0.566 -5.070*** 0.388 -4.60*** 0.302
lnsig2u -3.422*** 0.246 -4.356*** 0.568 -3.98*** 0.482
sigma_v 0.049 0.014 0.079 0.015 0.100 0.015
sigma_u 0.181 0.022 0.113 0.032 0.136 0.032
sigma2 0.035 0.007 0.019 0.005 0.028 0.006
lambda 3.660 0.032 1.429 0.046 1.359 0.046
Likelihod-ratio test of
sigma_u = 0 15.8***
1.29
2.15*
Prob >=chibar2 0.000 0.128 0.071
*p <0.10, ** p< 0.05, *** p< 0.01.
In the highland district, the effects of land price and labor wage were significant with negative
relation of land price with cost of wheat production (Table 38). In the highland, more land is
allotted to barley production. Higher rental value of land makes farmer allocate less land to
wheat production, and this also implies less allocation of other farm inputs for production of
111
wheat. Generally, wald test gave significant chi-square statistic showing that the effects of the
prices of inputs on cost of production were significant at one or five percent level of
significance in all agro-ecologies.
The variables indicated by sigma_v and sigima_u are the standard deviations of random
variation term and the inefficiency factors, respectively. Sigma2 is the total error variance and
lambda the ratio of the variance of random variation (v) to inefficiency term (u). The mean of
allocative efficiency scores shows that households in lowland, midland and highland districts
are allocatively efficient by about 89 percent, 88 percent and 87 percent respectively. One
way analysis of variance was used for testing the difference of the mean of allocative
efficiency scores among the districts. The test result shows significant F-value (2.73) at 10
percent level of significance (prob > F is equal to 0.095). Bartlett‘s test for equal variance also
shows significant chi-square statistic (10.24) at 1 percent level of significance (prob > chi2 =
0.006), indicating no equal variance of the means of scores among agro-ecologies.
Table 39. Summary of allocative efficiency scores of households in wheat production
Study area Mean Std. dev. Min Max
Lowland 0.895 0.063 0.549 0.968
Midland 0.885 0.052 0.723 0.968
Highland 0.877 0.068 0.675 0.977
Total 0.884 0.063 0.559 0.977
F-value 2.37; Prob > F = 0.095.
Chi-square value 10.24; Prob > chi2 = 0.006.
Source: Computation from own data
4.2.4. Economic efficiency
Economic efficiency is determined by multiplying the technical and allocative efficiencies of
a given farm or firm. The economic efficiency of farmers of the study area has been given in
112
Table 40. The mean technical, allocative and economic efficiencies of the study area were
about 75, 85 and 66 percents respectively. Economic efficiency ranges from minimum of
about 19 percent to maximum of 92 percent, indicating a wide range in economic efficiencies
of the farm households. It shows that there is a scope for increasing all efficiencies to improve
farm households‘ output and economic benefits in wheat production in the study area.
Summary of the mean economic efficiency of the three agro-ecologies show that lowland
district was the lowest with mean economic efficiency of 51 percent followed by highland
district with economic efficiency of 69 percent. Relatively, midland district had the highest
(73%) economic efficiency though there were chances of increasing the economic efficiencies
of farm households in wheat production in all agro-ecologies.
Table 40. Summary of mean technical, allocative and economic efficiencies
Efficiency Lowland Midland Highland Total
Technical 0.569 0.820 0.784
0.749
Allocative 0.896 0.888 0.878 0.854
Economic 0.512 0.729 0.690 0.663
Table 41 shows one way analysis of variance for testing mean economic efficiency of
households among districts. The test statistic (F value) shown as 102.52 with p-value (Prob >
F = 0.0000) is statistically significant at 1 percent level of significance implying that the mean
economic efficiency among districts is different.
Table 41. Analysis of variance of economic efficiency scores of districts
Source SS df MS F Prob> F
Between groups 2.568 2 1.284 102.52*** 0.0000
Within groups 4.734 378 0.013
Total 7.302 380 0.019
Bartlett's test for equal variances: chi2(2) = 12.15, Prob>chi2 = 0.002
113
The test of variance of means also shows significant chi-square statistic (chi-square = 12.15,
prob > chi2 = 0.002). This shows the variance of the mean economic efficiency of the agro-
ecologies were not equal (Table 41). Generally, the analyses of technical, allocative and
economic efficiencies in wheat production show considerable variations among agro-
ecologies. Households are inefficient technically, allocatively and economically. There is a
potential to improve households‘ wheat production efficiency with the current technology and
input levels.
4.3. Determinants and Impact of Farming Practices
This section presents factors influencing wheat row planting and choice of precursor crop for
rotation for wheat production, and the impact of wheat row planting on wheat yield of
smallholders in the study area.
4.3.1. Factors affecting adoption of wheat row planting
The explanatory variables that were assumed to affect adoption of planting wheat in row were
age and educational level of household head, farming experience of household head, total land
owned, household size, livestock holding size, total number of different types of crops
cultivated in 2012/13 cropping season, access to improved seed, chemical fertilizers and
agricultural extension services, off-farm income and agro-ecological orientation of
households (Table 5).
The dependent variable was adoption of wheat row planting. The variable is binary with two
outcomes. If a farmer practiced wheat row planting in 2012/13 cropping season, the variable
assumes value 1 or, 0 otherwise (did not practice planting in-row). STATA automatically
checks for collinear variables during logit estimation; and there were no collinear variables.
The result after the fourth iteration is shown in Table 42. The Wald test was also conducted
for testing the null hypothesis of logit model coefficients that all of the coefficients except the
intercept are simultaneously equal to zero. The resulting chi-square test with 14 degrees of
freedom (chi2 (14) = 40.70, prob > chi2 = 0.0002) is greater than the z- value in the output
114
from the estimation. This implies that the null hypothesis of all coefficients are
simultaneously equal to zero is rejected at 0.01 level (chi2= 40.7, df= 14, p< 0.01). The
likelihood-ratio test (LR chi2 (14) = 56.74, prob > chi2 = 0.0000) shows significant chi-
squared statistic. The measure of fittness indicated by Pseudo or McFadden‘s R2 is 0.11. The
goodness of fit is of secondary importance because what matters in logit model is the sign and
statistical significance of the coefficients of explanatory variables. The z-test which is equal to
the estimate divided by its standard error with two tailed significance level listed as p > z.
The result depicts that having different educational level and livestock holding sizes had
significant effects on the probability of practicing planting wheat in-row (z= 3.05, p< 0.01;
and z= 1.97, p< 0.05 respectively). Similarly, having access to improved seed and agricultural
extension services had significant effects on the probability of planting wheat in-row (z= 2.75,
p< 0.01; and z= 3.12, p< 0.01 respectively). Livestock is source of farm manures and traction
power as well as source of cash income for purchase of different farm inputs, and enhance
adoption of technology.
The coefficients of access to improved seed and extension service were tested whether they
were equal in effect using wald test. The test statistic of equality of coefficients (χ2 (1) = 3.06,
prob > χ2 = 0.08) was not significant at 0.05 level; and the null hypothesis of equal effect of
coefficients cannot be rejected. Therefore, the two variables had equal effect on the
probability of adoption of planting wheat in-row. Therefore, variables education, access to
improved seed and agricultural extension service and livestock holding size are statistically
different from zero at 0.05 level; and they are significant variables that affect adoption of
planting wheat in row. The estimated slope coefficient of the independent variables also
suggests that for a unit increase in the independent variable, the log of the odds in favor of
adopting row planting increases by the units equal to the value of the coefficient of the
variable.
The significant effects of the coefficient on probability of adoption of row planting have been
in line with the hypothesized effects specified under Sub-section 3.4.2.2 for each variable. As
specified there, educated farmers are better able to process information and test the row
planting method; livestock is a source of traction power in plowing and ease the labor
115
constraint in row planting; improved seed is basic input for practicing row planting since
farmers do not use local varieties; and agricultural extension service enables farmers to
understand the usefulness of the row planting and as a result these variables increase the
probability of household‘s participation in wheat row planting.
Table 42. Determinants of adoption of wheat row planting
Logistic regression Number of obs = 381
LR chi2(14) = 56.74***
Prob > chi2 = 0.0000
Log likelihood = -231.74
Pseudo R2 = 0.11
Practice of row
planting Coef. Std.error Z P > Z
Age -0.006 0.024 -0.24 0.813
Education 0.111*** 0.037 3.05 0.002
Experience 0.009 0.025 0.34 0.732
Land 0.005 0.108 0.04 0.966
Household size 0.086 0.078 1.1 0.271
Livestock 0.066** 0.034 1.97 0.049
Crops 0.049 0.092 0.53 0.597
Income -0.004 0.017 -0.24 0.809
Fertilizers -0.371 0.443 -0.84 0.403
Seed 0.747*** 0.272 2.74 0.006
Extension 1.969*** 0.631 3.12 0.002
Credit -0.049 0.262 -0.19 0.85
Lowland -0.115 0.371 -0.31 0.757
Midland 0.232 0.333 0.7 0.485
Constant -3.776*** 1.026 -3.68 0.000
*P < 0.10; ** p < 0.05; *** P < 0.01.
116
The sign of the coefficients are also the sign of the marginal effects for the logit model,
indicating that the marginal effects of these significant variables on the probability of
adopting planting wheat in row is positive.
Similarly, the effects of age, off-farm income, access to credit and chemical fertilizers, and
lowland and highland locations of household on the probability of adopting wheat row plating
were negative but non-significant. The constant (intercept) is the value of the log-odds in
favor of row planting if the values of independent variables are zero. However, this has no
physical meaning.
The linear combination of Xjβ was calculated to predict probability in the logit model, where
Xj are the independent variables in the jth
observation and β is the estimated parameter vector.
The prediction of probability of the positive outcome, that is, Pr (row planting = 1) in the logit
model and its summary statistics show that the predicted probabilities in the sample range
from 0.019 to 0.942, with average probability of 0.428. The mean probability of observing
adoption of planting wheat in-row was 0.428 (Table 43). That is, the computed mean
probability at the mean values of independent variables was 0.428, indicating that the
probability of a household adopting row planting was about 43 percent.
Table 43. Probability estimate in the logit model for the use of wheat row planting
Variable Obs Mean Std. Dev. Minimum Maximum
Pr (row planting = 1) 381 0.428 0.182 0.019 0.942
4.3.2. Factors affecting choice of precursor crop for rotation
The maximum likelihood estimation was used to estimate multinomial logit model with three
outcome categories of dependent variable and a set of explanatory variables. The dependent
variable is choice of precursor crop which has three categories (Pulse, vegetable and cereal).
Table 44 provides the maximum likelihood estimates of multinomial logit model for the
choice of precursor crop to wheat planting as dependent variable, and set of explanatory
variables assumed to affect the probability of the choice being one of the outcome categories.
117
Table 44. Determinants of the choice of precursor crop to wheat planting
Independent
variables
Choice Category
Pulses Vegetables
Coefficient Z-value Coefficient Z-value
Age -0.006 0.25 -0.132** 2.46
Education 0.056 1.46 0.114 1.61
Experience 0.012 0.46 0.127** 2.22
Land holding 0.022 0.2 0.036 0.17
Livestock holding 0.019 0.58 -0.062 0.86
Income 0.019 1.02 0.045* 1.65
Household size 0.060 0.75 -0.029 0.18
Access to pesticides 0.549** 2.04 1.121** 2.35
Access to credit -0.379 1.29 0.571 1.23
Access to fertilizers 0.032 0.08 0.167 0.19
Disease problem -0.623* 1.68 -0.349 0.42
Lowland location -1.179*** 2.83 -1.535* 1.71
Midland location 0.291 0.91 0.834 1.47
Constant -0.910 1.05 -0.460 0.25
Wald chi-squared (26) 77.350***
Prob > chi2 0.0000
Pseudo R2 0.115
Log likelihood -297.722
Base outcome Cereal
*p < 0.1, **p < 0.05, ***p < 0.01.
The null hypothesis is that all the coefficients associated with independent variables (except
the constant) are simultaneously equal to zero. The Wald test (Wald chi2 (26) equals to 77.35,
with p-value of Prob > chi2 = 0.0000 i.e. p < 0.0005) gives significant chi-square statistic,
implying rejection of null hypothesis. It means that the effect of independent variables on the
118
outcome variable was not zero. The measure of fit given by Pseudo R2 was also equal to
0.115. The effect of lowland location of household, access to pesticides, crop disease
problems age of household head, faming experience of household head and access to
pesticides, off-farm income were significant factors. The value of the constant or the
intercept is the value of the log-odds in favor of adopting a choice category if the values of the
independent variables are equal to zero. However, in most cases, the interpretation of the
intercept may not have any physical meaning.
It has to be noted that a positive or negative coefficient of multinomial logit need not mean
that an increase or decrease in the independent variable leads to an increase or decrease in the
probability of an outcome being selected. But the coefficients can be interpreted with
comparison to the base category (cereal choice category). So a positive coefficient means that
as independent variable increases, the households (farmers) are more likely to choose an
alternative with positive coefficient than the base alternative (cereal) whereas the negative
coefficients indicate that as the independent variables with negative coefficients increase, a
household is more likely to choose the base category (Cereal) than the other category with
negative signs of variables.
The significant negative coefficient of age (-0.132) for vegetable choice category (Table 44)
indicates that as age of household head increases from its mean, the farmer is more likely to
choose the base category (cereal) to vegetable as precursor crop to wheat planting. It appears
that since older people have more household size, and to feed the large household size, the
focus could be on cereal food crops. Similarly, as access and use of pesticides increases,
households are more likely to choose pulses and vegetable crops as precursors to wheat
planting. It seems that accessibility and utilization of pesticides encourage production of pulse
and vegetable crops. This could be due to the control of diseases and pests that affect the
production of these crops. If there is no access to pesticides, farmers may not produce pulses
and vegetables that are affected by various pests and diseases. Low land location of
households also encourages households to use cereals as precursor crops to wheat planting.
This could be due to the biophysical condition of lowland area. In lowland areas where
rainfall is scarce and farmers have no access to irrigation facilities, farmers are less
119
encouraged to adopt the production of pulse and vegetable crops. They rather go for early
maturing cereal food crops to meet their needs.
A likelihood ratio test for independent variables (Table 45) showed that the effects of age and
access to pesticides of household head on choice of precursor crop were significant at 0.05
level. The effect of farming experience of household was significant at 0.1 level. But the
effect of lowland compared to the highland was significant at p < 0.01. Therefore, the effects
of age, farming experience of household heads, household‘s access to pesticides and lowland
location were significant; and these variables could determine households‘ choices of
different precursor crops to wheat cultivation in the study area. Targeting these significant
variables in agricultural extension system could help to break cereal mono-cropping, and
encourage farmers to choose pulses and vegetables in their crop productions.
Table 45. Likelihood ratio tests for independent variables
Precursor crop Chi2 df P> Chi
2
Age 6.967** 2 0.031
Education 3.883 2 0.143
Experience 5.345* 2 0.069
Land holding 0.055 2 0.973
Livestock holding 1.472 2 0.479
Income 2.892 2 0.235
Household size 0.703 2 0.704
Access to pesticides 7.998** 2 0.018
Access to credit 4.287 2 0.117
Access to fertilizers 0.039 2 0.981
Disease problem 2.831 2 0.243
Lowland location 10.574*** 2 0.005
Midland location 2.513 2 0.280
Ho: All coefficients associated with given variable(s) are 0.
120
Hausman‘s test for independence of irrelevant alternatives (IIA) assumption has been given
for each of the alternatives (Table 46). The assumption implies that the choice between a
collection of alternatives is not affected if non chosen alternatives are made unavailable. That
is, adding or deleting outcomes does not affect the odds among the remaining outcomes.
Significant values of the test statistic (χ2) indicate violation of the IIA assumption. However,
the test statistic for the outcome categories (pulses and vegetables but not the base category)
found to be non significant, implying that IIA assumption has not been violated. Moreover,
according to Hausman and McFadden (1984), a negative value of the test statistic (χ2 = -0.127
and χ2 = -0.858) is evidence that IIA assumption has not been violated.
Table 46. Hausman tests of IIA assumption
Precursor χ2 df P> χ
2 Evidence
Pulses -0.127 14 1 for Ho
Vegetables -0.858 14 1 for Ho
Cereal 33.567 14 0.002 against Ho
Ho: Odds (Outcome-J vs Outcome-K) are independent of other alternatives.
Probability of positive outcome, that is, probabilities of choosing pulses, vegetables, and
cereal as precursor crop to wheat production were estimated from the multinomial logistic
regression. Table 47 shows the probabilities of the three outcome categories of the dependent
variable after multinomial logistic regression. The mean predicted probability was 0.562 for
the choice being cereal as precursor crop to wheat cultivation, and choice being pulse was
0.363, ranging from minimum probability 0.01 to maximum of 0.906. The mean predicted
probability for the choice being vegetable was 0.075.
Table 47. Probability estimates for choice categories after multinomial logistic regression
Variable Obs Mean Std. Dev. Min Max
Pr(precursor=pulses) 381 0.363 0.220 0.010 0.906
Pr(precursor=vegetable) 381 0.075 0.125 0.000 0.718
Pr(precursor=cereal) 381 0.562 0.243 0.074 0.990
121
This shows that the probability of choice being cereal crop as precursor to wheat cultivation
was very high compared to the other two choice categories. This indicates that planting of
wheat after cereal was a common farming practice in the study areas.
Since the parameter coefficient of multinomial logit is difficult to directly interpret, the
marginal effects could be used to measure the impact on the probability of observing each of
the choice outcomes. The marginal effects on choice probabilities evaluated at mean values of
the independent variables were computed and given in Table 48. As depicted in the table, for
each independent variable there are three marginal effects corresponding to the three
probabilities, and these three marginal effects sum to zero because probabilities of each
outcome sum to one. The marginal effect on pr(y = pulse) of a change in livestock holding
size estimated at the mean of variables indicates that a unit change in livestock size by one
TLU increases the probability of pulses being precursor crop to wheat planting by 0.005 while
the rest of the variables are held constant at their means. However, the marginal effect of
livestock holding size on probability of choosing pulses as precursor crop was statistically non
significant. Similarly, a year increase in age of household head from its mean decreases the
probability of vegetables being precursor to wheat planting by 0.006.
The marginal effects of age (at p < 0.01), farming experience of household head (at p < 0.05),
access to pesticides and lowland location of household (at p < 0.1) on the probability of
precursor to wheat planting being vegetable crops were significant. Similarly the marginal
effects of access to pesticides and lowland location of household on the probability of the
choice being pulse crop were significant at 0.1 and 0.01 levels respectively, with negative
marginal effect of the lowland location on the choice category. The marginal effects were
significant for educational level of household head, household access to pesticides, and
lowland location of household on the probability of choice being cereal crop at p < 0.1, 0.05
and 0.01levels, respectively. Household head‘s farming experience and access to pesticides
were significant variables that positively affect the probability of pulses or vegetables being
precursors to wheat production in the study area.
122
Table 48. Marginal effect after multinomial logistic regression
Variables
Choice Category
Pulses Vegetables Cereals
(dy/dx) Z-value (dy/dx) Z-value (dy/dx) Z- value
Age 0.001 0.15 -0.006*** -2.58 0.005 0.94
Education 0.011 1.26 0.004 1.38 -0.015* -1.73
Experience 0.001 0.1 0.005** 2.27 -0.006 -1.06
Land holding 0.004 0.18 0.002 0.14 -0.006 -0.22
Livestock
holding
0.005 0.74 -0.003 -0.98 -0.002 -0.27
Income 0.004 0.87 0.002 1.48 -0.006 -1.24
Household size 0.014 0.8 -0.002 -0.33 -0.012 -0.64
Access to
pesticides
0.102* 1.71 0.048* 1.76 -0.150** -2.46
Access to credit -0.093 -1.55 0.038 1.35 0.055 0.86
Access to
fertilizers
0.005 0.05 0.007 0.19 -0.012 -0.12
Disease problem -0.140 -1.59 -0.005 -0.11 0.145 1.63
Lowland location -0.214*** -3.2 -0.042* -1.77 0.257*** 3.71
Midland location 0.049 0.7 0.038 1.16 -0.088 -1.19
dy/dx is for discrete change of dummy variable from 0 to 1
Predicted probability of choice categories
Probability of pulses 0.338
Probability of vegetables 0.049
Probability of cereal 0.613
*p < 0.1; **p< 0.05; ***p< 0.01.
123
4.3.3. Impact of wheat row planting on yield
The use of propensity score matching model was to answer the question ―what would be
wheat output per hectare for households who planted wheat in-row had these households not
practiced planting wheat in-row?‖ That is, the impact of planting wheat in-row on wheat
yield was analyzed using propensity score matching.
Table 49. Logit estimates for propensity score for Lowland
Variables Coef. Std.error z P>z
Age 0.010 0.063 0.15 0.877
Education 0.290 0.108 2.69*** 0.007
Experience 0.069 0.062 1.11 0.266
Land holding 0.324 0.324 1.00 0.318
Crops 0.143 0.226 0.63 0.528
Rotation 0.806 0.683 1.18 0.238
Credit 0.543 0.620 0.88 0.381
Pesticides 1.571 0.902 1.74* 0.082
Seed 2.023 0.995 2.03** 0.042
Household size -0.520 0.278 -1.87* 0.062
Livestock holding 0.189 0.128 1.47 0.142
Income -0.012 0.067 -0.17 0.862
Constant -6.388 2.788 -2.29** 0.022
Number of observations = 83
LR χ2 (12) = 32.82***
Prob > χ2 = 0.001
Pseudo R2 = 0.292
Log likelihood = -39.756
*p < 0.1, **p < 0.05, ***p < 0.01.
The matching was carried out separately for each district so that participant and non-
participant farmers in row planting share similar observable characteristics. Because, the
124
midland location has positive effect whereas lowland and highland location of households has
negative effect on the probability of practicing wheat row planting (Table 42). The yield of
wheat i.e. the outcome variable due to participation and non-participation in wheat row
planting is statistically significantly different in midland and highland agro-ecologies, and
non-significant in lowland agro-ecology (Table 24). Therefore, to ensure similarity in
observable household characteristics, separate PSM analysis for each agro-ecology was
conducted. STATA version 11 and psmatch2 command was used to run the matching. The
matching estimator was selected based on the number of matched samples and insignificant
variables after matching as well as low Pseudo R2. Accordingly, the analyses of the impact of
row planting on wheat yield for lowland, midland, and highland districts have been given
below as in the order of the names of agro-ecologies specified.
Table 49 presents logit estimate for propensity scores for lowland district. Propensity score
regression suggests that most independent variables appear to raise the probability of
participation in row plating except household size and annual off-farm income of households.
Educational level of household head, household access to improved wheat seed and
pesticides, and household size were found significant factors affecting propensity of
participation in wheat row planting in lowland district.
Matching participant and nonparticipant households was carried out to determine the common
support region. The basic criterion for determining the common support region is to delete all
observations whose propensity score is smaller than the minimum propensity scores of
participants and larger than the maximum in the control group (Caliendo and Kopeining,
2008). That is, deleting all observations out of the overlapping region. The propensity score
(the conditional treatment probability) ranges from 0.186 to 0.973 for participant households,
and 0.008 to 0.893 for non participant households in lowland district. Therefore, the
propensity score for the common support region (overlapping region) ranges from 0.186 to
0.893 (Figure 6).
125
Figure 6. Graph of propensity score by treatment status for Lowland
Table 50. Performance of matching estimators for Lowland
Matching estimator
Performance criteria
Balancing test*
Pseudo R2
Matched
sample size
Nearest neighbor
Neighbor (1) 8 0.148 54
Neighbor (2) 12 0.055 54
Caliper matching
Caliper (0.01) 12 1.000 22
Caliper (0.25) 12 0.042 47
Kernel matching
Bandwidth (0.25) 12 0.021 54
Bandwidth (0.1) 12 0.036 54
Bandwidth (0.5) 12 0.032 54
*Number of independent variables with no statistically significant mean difference between
the matched groups of households.
0 .2 .4 .6 .8 1Propensity Score
Untreated: Off support Untreated: On support
Treated: On support Treated: Off support
126
Choice of matching algorithm was carried out from nearest neighbor, caliper and kernel
methods. The choice of estimator based on three criteria; namely, balancing test, pseudo R2
and matched sample size. The matching estimator which balances more independent
variables, has low pseudo R2 value and results in large matched sample was chosen as being
the best estimator of the data. Accordingly, Table 50 shows that kernel matching with
bandwidth 0.25 was found to be the best estimator for data of lowland district. It resulted in
lowest pseudo R2 value, well balanced covariates, and large sample size by discarding only 29
unmatched households from the sample of lowland district.
The impact of wheat row planting on household‘s wheat yield was based on sample of
matched treated and control groups. The average row planting effect (ATT) estimation result
showed that wheat row planting was not associated with significantly increased yield in
lowland district as indicated by statistically not significant t-value (Table 51). The average
wheat yield of participant households was higher than that of non participant households by
0.3 percent with t-value of 0.02, implying statistically non significant yield difference
between participant and non participant households in wheat row planting.
Table 51. Estimates of average treatment effects for Lowland
Outcome indicator Sample Treated Controls Difference S.E. T-stat
Wheat yield (q/ha)
Unmatched 17.37 15.75 1.62 1.22 1.34
ATT 16.64 16.59 0.05 1.66 0.02
ATU 16.64 15.66 -0.97 . .
ATE -0.47 . .
That is, the average impact of planting wheat in-row on wheat yield of participant farmers
was not significant in lowland district. Sensitivity analysis result showed that the inference for
the impact of wheat row planting on yield in lowland district is not changed though the
participant and control households have been allowed to differ in their odds of being treated in
terms of unobserved covariates (Appendix Table 12).
127
Similarly, for the midland agro-ecology, PSM analysis for the impact of row plating on wheat
yield was conducted. The same procedures and dependent and independent variables were
used as in the case of lowland district. The logit regression of propensity score depicts that the
conditional probability of participation in wheat row planting was affected by practice of crop
rotation, access to improved seed, and livestock holding size at 0.1, 0.05, and 0.01 levels,
respectively.
Table 52. Logit estimate for propensity score for Midland
Variables Coefficient Standard error z P>z
Age -0.037 0.062 -0.59 0.553
Education -0.052 0.078 -0.67 0.506
Experience -0.008 0.063 -0.13 0.895
Land holding -0.573 0.452 -1.27 0.205
Crops 0.104 0.232 0.45 0.653
Rotation 1.293 0.686 1.88* 0.060
Credit -1.860 1.185 -1.57 0.116
Seed -2.209 0.913 -2.42** 0.015
Household size 0.048 0.187 0.26 0.796
Livestock holding 0.374 0.129 2.9*** 0.004
Income 0.051 0.060 0.86 0.390
Constant -0.204 2.013 -0.1 0.919
Number of obs = 133
LR chi2(11) = 49.62***
Prob > chi2 = 0.0000
Pseudo R2 = 0.378
Log likelihood = -40.902
*p < 0.1, **p < 0.05, ***p < 0.01.
Access to improved seed negatively affected probability of participation in row planting in
midland agro-ecology. It seems that since midland district is well known for wheat
production, households with access to improved wheat seed are not willing to use time and
128
labor consuming farming practice. The overall effect of the independent variables on
probability of participation was different from zero as implied by significant chi-square
statistic of the likelihood ratio test.
The matching of participant and non-participant households to determine the common support
region indicate that the propensity score for the overlap region ranges from 0.0028 to 0.7212.
This is the region between the minimum propensity score of participants and the maximum
propensity score of non-participant households in wheat row planting (Table 53).
Table 53. Summary of propensity scores for participants and non-participants of Midland
Variable Obs Mean Std.dev Minimum Maximum
Participants 26 0.5167 0.2796 0.0028 0.9905
Non-participants 107 0.1174 0.1631 0.0007 0.7212
Total 133 0.1955 0.2478 0.0007 0.9905
Summary of propensity score has been depicted for participant, non-participants, common
support region and off-support regions from the two categories of households in Figure 7.
129
Figure 7. Graph of propensity score by treatment status for midland district
Table 54. Performance of matching estimators for Midland
Matching estimator
Performance criteria
Balancing test
Pseudo R2
Matched sample
size
Nearest neighbor
Neighbor (1) 10 0.443 120
Neighbor (2) 11 0.134 120
Caliper matching
Caliper (0.01) 11 1.000 20
Caliper (0.25) 11 0.226 38
Kernel matching
Bandwidth (0.25) 11 0.021 120
Bandwidth (0.1) 11 0.024 120
Bandwidth (0.5) 11 0.118 120
0 .2 .4 .6 .8 1Propensity Score
Untreated: Off support Untreated: On support
Treated: On support Treated: Off support
130
Kernel matching method with bandwidth of 0.25 was found to be the best estimator of the
data of Hetosa district (Table 54). As depicted in the table, relatively, this estimator resulted
in least pseudo R2 and large number of matched sample size by discarding only 13 unmatched
households from total of 133 households.
Estimation of average treatment effect (ATT) for midland district showed that wheat row
planting has significant effect on yield of participant farmers with significant t-statistic (2.46)
at one percent significance level (p < 0.01). The average wheat yield of participant households
in wheat row planting was higher by 14.6 percent when compared with the average yield of
non participant households in midland district.
Table 55. Estimates of average treatment effects for Midland
Outcome indicator Sample Treated Controls Difference S.E. T-stat
Wheat yield (q/ha)
Unmatched 41.231 30.135 11.096 1.771 6.27
ATT 39.583 33.819 5.765 2.345 2.46*
ATU 30.505 39.067 8.563 . .
ATE 8.096 . .
* Significant at 1% significance level.
In the highland district, the logit regression of propensity score depicts that the conditional
probability of participation in wheat row planting was affected by households‘ access to
improved wheat seed at 0.05 levels. In the high land barely is a major crop cultivated by
households. However, access to improved wheat seed might have influenced households to
practice wheat row planting. The overall effect of the independent variables on probability of
participation was different from zero as implied by significant chi-square statistic at 10
percent of the likelihood ratio test.
131
Table 56. Logit estimate for propensity score for Highland
Variables Coefficient Standard error z P>z
Age -0.061 0.046 -1.35 0.177
Education 0.072 0.060 1.2 0.229
Experience 0.054 0.048 1.13 0.259
Land holding 0.053 0.164 0.32 0.747
Crops 0.184 0.197 0.93 0.35
Rotation 0.782 0.640 1.22 0.222
Credit -0.480 0.489 -0.98 0.327
Pesticides -0.068 0.408 -0.17 0.867
Seed 0.840 0.404 2.08** 0.038
Household size 0.162 0.125 1.3 0.193
Livestock holding -0.008 0.042 -0.19 0.853
Income -0.003 0.021 -0.16 0.874
Constant -1.661 1.562 -1.06 0.288
Number of observation = 165
LR χ2 (12) = 20.98*
Prob > χ2 = 0.051
Pseudo R2 = 0.095
Log likelihood = -100.555
*p < 0.1, **p < 0.05.
Figure 8 shows that the distribution of propensity scores of both participant and non
participant households of highland district before common support condition is imposed. The
figure shows that there is wide area in which propensity score of participant and non
participant households are similar. Therefore, it is possible to match the two groups using the
common support region or overlapping area of propensity scores.
132
Figure 8. Kernel density of propensity scores of highland district
The summary statistics of propensity scores of households (Table 57) depicts that the
common support region lies within the range of 0.1651 to 0.7248 propensity scores for both
participant and non participant households.
Table 57. Summary of propensity scores for participants and nonparticipants of Highland
Variable Obs Mean Std.dev Minimum Maximum
Participants 66 0.4709 0.1579 0.1651 0.7604
Non-participants 99 0.3527 0.1629 0.0253 0.7248
Total 165 0.4000 0.1706 0.0253 0.7604
Kernel matching method with bandwidth of 0.1 was found to be the best estimator of the data
of highland district (Table 58). As depicted in the table, relatively, this estimator resulted in
least pseudo R2 (0.006) and large number of matched sample size (152) by discarding 13
unmatched households from total of 165 households.
0.5
11.5
22.5
De
nsity
0 .2 .4 .6 .8psmatch2: Propensity Score before matching
All households
Participant households
Non participant households
kernel = epanechnikov, bandwidth = 0.0553
Kernel density estimate
133
Table 58. Performance of matching estimators for Lemu-Bilbilo district
Matching estimator
Performance criteria
Balancing test Pseudo R2
Matched
sample size
Nearest neighbor
Neighbor (1) 11 0.064 152
Neighbor (2) 11 0.036 152
Caliper matching
Caliper (0.01) 11 0.077 92
Caliper (0.25) 12 0.087 121
Kernel matching
Bandwidth (0.25) 12 0.013 152
Bandwidth (0.1) 12 0.006 152
Bandwidth (0.5) 12 0.036 152
134
Figure 9. Graph of propensity score by treatment status for highland district
The estimated average treatment effect (ATT) for Lemu-Bilbilo district showed that wheat
row planting has significant effect on yield of participant farmers with significant t-statistic
(3.79) at 1 percent significance level (p < 0.01). The average wheat yield of participant
households in wheat row planting was higher by 13.9 percent when compared with the
average yield of non participant households in highland district.
Table 59. Estimates of average treatment effects for Highland
Outcome indicator Sample Treated Controls Difference S.E. T-stat
Wheat yield (q/ha)
Unmatched 27.947 23.347 4.600 0.969 4.75
ATT 27.690 23.816 3.874 1.023 3.79*
ATU 23.678 27.908 4.229 . .
ATE 4.082 . .
* Significant at 1% significance level.
0 .2 .4 .6 .8Propensity Score
Untreated: Off support Untreated: On support
Treated: On support Treated: Off support
135
All in all, the impact of wheat row planting was found to be significant on the wheat yield of
participant households both in midland and highland agro-ecologies. However, in lowland
district of Dodota, its impact on participants‘ wheat yield was non-significant when compared
with non participant households of similar characteristics. Therefore, to enhance wheat yield
in the lowland area, it is suggested that agricultural extension activities need to focus on the
biophysical nature and other agronomic practices before promoting and scaling up of wheat
row planting.
136
5. SUMMARY AND CONCLUSSIONS
5.1. Summary
The objectives of this study were to analyze wheat production efficiency in different agro-
ecologies, and identify the sources of inefficiencies in production and determinants for
practices of wheat row planting and crop rotation as well as impact of wheat row planting on
yield. Descriptive statistics, Cobb-Douglas stochastic frontier, logit, multinomial logit,
propensity score matching models, and cross-sectional survey data were used to achieve the
objectives of the study. Data on farm households‘ inputs and output, household
socioeconomic characteristics, access to and use of various farm inputs, use of improved
agronomic practices, etc. were collected from randomly selected 381 farm households for
2012/13 cropping season. The analyses of the data resulted in mean technical efficiency
estimates of 57, 82, and 78 percents for the lowland, midland, and highland agro-ecologies,
respectively. The allocative efficiency estimates for the lowland, midland and highland agro-
ecologies were 89, 88 and 87 percents, respectively; and the economic efficiency estimates for
the lowland, midland and highland agro-ecologies were about 51, 73, and 69 percents,
respectively. The technical efficiency ranges from about 24 to 94 percents among all sample
farm households.
Smallholders access to improved seed and credit, practice of crop rotation and wheat row
planting, availability of animal power and household labor, age of household head were
significant factors that influenced wheat production efficiency. In the lowland agro-ecology,
output elasticities associated with land, labor, chemical fertilizers and other inputs (seed and
pesticides) were positive and significant, with output elasticities of seed and pesticides are the
highest. Similarly, the responsiveness of wheat output to the changes in the size of land and
chemical fertilizers utilization was positive and significant in midland agro-ecology. In the
highland, the output elasticities of land and chemical fertilizers were significant with negative
elasticity of output due to land.
137
Wheat row planting is one of agronomic practices believed to increase wheat yield in
Ethiopia. The agricultural extension offices have been promoting and scaling-up wheat row
planting. As a result, wheat row planting has been practiced in various wheat producing
regions of the country. This study was carried out in major wheat producing zone of the
country to identify the factors affecting the adoption of wheat row planting. The study
identified that access to improved seed and agricultural extension services, educational level
of household head, and livestock holding size were the factors positively and significantly
influencing the adoption of wheat row planting. Households‘ agro-ecological setting was non-
significant in affecting households‘ probability of adopting wheat row planting. However,
though the effect was non-significant, lowland orientation of households had lower effect on
the probability of adopting wheat row plating relative to the midland and highland agro-
ecology. The midland had more effect on probability of adopting row planting relative to the
highland agro-ecology.
Crop rotation is also one of agronomic practices believed to increase wheat yield in Ethiopia.
However, most farmers practice cereal mono-cropping system in the study area. Crop
rotation, especially precursor crop planted preceding wheat planting, has the capacity to
improve wheat yield. The study identified that farming experience and skill of household
head, livestock holding sizes, and access to pesticides had positive and significant effects on
farmers‘ choices of pulse and vegetable crops being precursors to wheat planting.
The impact of wheat row planting was found to be significant on the wheat yield of
participant households both in midland and highland agro-ecologies. The average wheat yield
of participant households in wheat row planting was higher by 14.6 and 13.9 percents when
compared with the average yield of non participant households of similar observable
characteristics in midland and highland agro-ecologies, respectively. However, in lowland
district of Dodota, impact of row planting on participants wheat yield was not-significant
when compared with non participant households.
138
5.2. Conclusions and Recommendations
Efficiency analyses imply that there is more scope for increasing wheat output with the
current technology and input levels. Smallholders were inefficient in wheat production in the
study areas; and improvement in efficiencies in wheat production needs attention as it
provides significant source of enhancement in wheat output. The sources of inefficiencies
were mainly related to farming experiences and skills, availability of labor and livestock for
farm operations and other inputs, access to improved seed and credit as well as adoption of
row planting and crop rotation farming practices. Though production efficiency was relatively
higher in the midland agro-ecology, there was disparity of production efficiency among agro-
ecologies and within agro-ecology. The elasticity of output with respect to inputs implies that
increased utilization or full exploitation of the potentials of the farm inputs can enhance wheat
yield in all agro-ecologies. The highland is more conducive to barley production. Farmers
relatively prefer barley to wheat, and allocate more farm inputs to barley production. The
results suggest that improving wheat yield requires improving the technical efficiency and
farm inputs utilizations in all agro-ecologies as well as farm households‘ socioeconomic
characteristics that affect technical efficiency in wheat production.
Low use or limited access to improved farm inputs is also a major challenge facing wheat
production in Ethiopia. Low utilization of farm inputs and adoption of improved farming
techniques made yield of wheat low. To feed the rapidly growing population and meet the
high demand, smallholders need to increase wheat yield through adoption of yield enhancing
farming practices that include row planting and crop rotation. Limited use of chemical
fertilizers and crop rotation are the characteristics of households in wheat production in the
study area. Though, great efforts of agricultural extension service to promote crop rotations,
most farmers are practicing planting of cereal after cereal. This has negative impact on crops
yield since most farmers are unable to afford the high prices of farm inputs specially prices of
chemical fertilizers and pesticides. Crop rotation improves soil fertility, reduces plant diseases
and weeds problems, and thereby increases crops yield. Improving households farming
experience and skills, livestock production and management, and access to and use of
pesticides could increase households‘ adoption of crop rotation. Targeting these variables in
139
the agricultural extension services could help to break the cereal mono-cropping system in the
study area. Wheat row planting has the potential of improving wheat yield. However, the row
plating needs to be associated with other agronomic practices in order to have increased wheat
yield.
The study recommends that improving production efficiency and encouraging adoption of
wheat row planting and crop rotation can enhance wheat yield in the study area. Increased use
or full exploitation of the potentials of farm inputs especially chemical fertilizers, improved
seed and pesticides improves wheat yield in all agro-ecologies. Socioeconomic characteristics
such as age of household, livestock holding size, crop rotation and wheat row planting reduce
the inefficiency of wheat production by lowering the deviation of observed output from the
frontier. Therefore, faming experiences and skills gained through age, better livestock
production and management, promotion and scaling up of wheat row planting and crop
rotation by agricultural extension need to be targeted for improving wheat production
efficiency. Households‘ socioeconomic characteristics such as education, livestock holding,
access to improved wheat seed and agricultural extension services increase households‘
probability of participation in wheat row planting. Promotion of education among farm
households, livestock production and management, accessibility and use of improved seed
through agricultural extension and other stakeholders, and provision of agricultural extension
services on wheat row planting increase probability of participation in wheat row planting. To
break cereal mono-cropping, improving farming skill of farmers through extension services,
promoting livestock management practices, and increasing accessibility and use of pesticides
will help farmers to adopt pulse and vegetable crops for rotation. Though wheat row planting
had non-significant impact on yield in selected lowland agro-ecology, further agronomic
studies are required on seed and fertilizer rates, plant spacing, early hand weeding and hoeing
as well as on other biophysical conditions of lowland that affect wheat yield. Generally,
agricultural research and extension activities need to target agro-ecology based levels of farm
inputs in production for full exploitation of the potentials of the inputs. Successful promotion
and scaling-up of wheat row planting and crop rotation need due consideration of farm
households‘ educational level, livestock holding, access to and use of improved wheat seed,
agricultural extension and pesticides.
140
6. REFERENCES
Ahmed, M. H., Teferra A. T., Yuya B. A. and Melese K.A., 2014. Impact of Soil
Conservation on Farm Efficiency of Maize Growers in Arsi Negelle, Central Rift Valley of
Ethiopia. International Journal of Economics and Empirical Research 2(2): 36-43.
Aigner, D., and S. Chu, 1968. On Estimating the Industry Production Function. American
Economic Review 58: 826–839.
Aigner, D., T. Amemiya and D. Poirier, 1976. On the Estimation of Production Frontiers.
International Economic Review 17: 377–396.
Aigner, D.J., C.A.K. Lovell and P. Schmidt, 1977. Formulation and estimation of stochastic
frontier production function models. Journal of Econometrics 6: 21-37.
Alemayehu Seyoum, P. Dorosh and Sinafikeh Asrat, 2011. Crop production in Ethiopia:
Regional patterns and trends, ESSP II Working Paper No. 0016. Development Strategy and
Governance Division, International Food Policy Research Institute.
Arega D., 2010. Productivity growth and effects of R & D in African agriculture. Agricultural
Economics 41: 223-238.
Arega D. and H.M. Rashid, 2006. The efficiency of traditional and hybrid maize production in
eastern Ethiopia: An extended efficiency decomposition approach. Journal of African
Economies 15(1): 91-116.
Arega Alene and Zeller, M., 2005. Technology adoption and efficiency in multiple crops
production in eastern Ethiopia: A comparison of parametric and non-parametric distance
functions. International Institute of Tropical Agriculture.
Asfaw Negassa, J. Hellin and Bekele Shiferaw, 2012. Determinants of adoption and spatial
diversity of wheat varieties on household farms in Turkey. CIMMYT Socio-economic
working Paper 2. Mexico, D.F.: CIMMYT.
Asres E., M. Nohmi, K. Yasunobu, A. Ishida and Arega D., 2014. The effect of agricultural
extension service on technical efficiency of teff producers in Ethiopia. American Journal of
Applied Sciences 11(2): 223 – 239.
Battese, G.E., 1992. Frontier production functions and technical efficiency: A Survey of
empirical applications in Agricultural Economics. Agricultural Economics 7: 185- 208.
Battese, G.E., and T.J. Coelli, 1995. A model for technical inefficiency effects in a stochastic
frontier production function for pane data. Empirical Economics 20: 325 – 332.
141
Becker, S.O. and A. Ichino, 2002. Estimation of average treatment effects based on propensity
scores. The Stata Journal 2(4): 358 – 377.
Bekele, H., H. Verkuijl, W. Mwangi and D. Tanner, 2000. Adoption of improved wheat
technologies in Adaba and Dodola Woredas of the Bale highlands, Ethiopia. Mexico, D.F.:
International Maize and Wheat Improvement Center (CIMMYT) and Ethiopian Agricultural
Research Organization (EARO).
Beltran, J. C., B. White, M. Burton, G. J. Doole and D.J. Pannell, 2013. Determinants of
herbicide use in rice production in the Philippines. Agricultural Economics 44: 45-55.
Berhanu Gebremedhin, Moti Jaleta and D. Hoekstra, 2009. Smallholders, institutional
services, and commercial transformation in Ethiopia. Journal of Agricultural Economics 40:
773-787.
Braun, H.J., G, Atlinand Payne, T., 2010. Multi-location testing as a tool to identify plant
response to global climate change. In: Reynolds, CRP.(ed.). Climate change and crop
production, CABI, London, UK.
Byerlee, D., Spielman, D.J., Alemu, D., M. Gautam, 2007. Policies to promote cereal
intensification in Ethiopia: A Review of evidence and experience. IFPRI discussion paper no.
707. Washington, DC: IFPRI.
Caliendo, M. and Kopeinig, S., 2005. Some practical guidance for the implementation of
propensity score matching, Discussion paper No. 1588, University of Cologne.
Castellazzi, M.S., Wood, G.A., Burgess P.J., Morris J., K.F. Conrad and Perry J.N., 2008. A
systematic representation of crop rotations. Agricultural Systems 97: 26-33.
Cavatassi, R., Lipper, L. and U. Narlochc, 2011. Modern variety adoption and risk
management in drought prone areas: insights from the sorghum farmers of eastern Ethiopia.
Agricultural Economics 42: 279–292.
CSA (Central Statistical Agency), 2005-2010. Annual Agricultural Sample Survey Report.
Addis Ababa, Ethiopia.
CSA (Central Statistical Agency), 2011. Report on area and production of major crops,
private peasant holdings, meher season, volume 1, statistical bulletin. Addis Ababa.
CSA (Central Statistical Agency), 2013. Report on area and production of major crops,
private peasant holdings, meher season, Addis Ababa.
CIMMYT (International Maize and Wheat Improvement Center) Economics Program, 1993.
The adoption of agricultural technology: A guide for survey design. Mexico, D.F.: CIMMYT.
142
CIMMYT (International Maize and Wheat Improvement Center), 2011. Wheat global alliance
for improving food security and the livelihoods of resource poor in developing world. Report
submitted by CIMMYT and ICARDA to CGIAR consortium board, 30 August
2011.(http://www.cimmyt.org/wheat) Accessed on November 4, 2012
CIMMYT (International Maize and Wheat Improvement Center), 2012. Wheat for food
security in Africa. Africa‘s food security and the changing demand for wheat over the coming
decades.
Cobb, C. W. and P. H. Douglas, 1928. A Theory of production. American Economic Review
18: 139-165.
Coelli, T. J., D. S Prasada Rao, C. J. O‘Donnell and G. E. Battese, 2005. An introduction to
efficiency and productivity analysis 2nd
ed. Springer, New York. NY.
Coelli, T. J. and S. Perelman, 1999. A comparison of parametric and non-parametric distance
functions: With application to European railways. European Journal of Operational Research
117: 326-339.
Conradie, B., Piesse, J., Thirle, C., 2009. District-level total factor productivity in agriculture:
Western Cape Province, South Africa, 1952-2002. Agricultural Economics 40(3): 265-280.
Debreu, G., 1951. The coefficient of resource utilization. Econometrica 19: 273–292.
Dehejia, R. H. and Wahba, S., 2002. Propensity score matching methods for non experimental
causal studies. The Review of Economics and Statistics, 84(1):151–161.
DSA (Development Studies Associates), 2006. Study on improving the efficiency of input
markets. Addis Ababa: Ministry of Agriculture and Rural Development, Federal Democratic
Republic of Ethiopia.
Diao, X., D. Headey and Johnson, M., 2008. Toward a green revolution in Africa: What
would it achieve, and what would it require? Agricultural Economics 39: 539–550.
Diao, X., P. Hazell, D. Rensnick and J. Thurlow, 2006. The role of agriculture in Sub-Saharan
Africa, DSGD paper No.29, IFPRI.
Dixon, J., H. J. Braun, P. Kosina, and J. Crouch (eds.), 2009. Wheat facts and futures 2009.
Mexico, D.F.: CIMMYT.
Doss, C. R., W. Mwangi, H. Verkuijl and H. de Groote, 2003. Adoption of Maize and Wheat
Technologies in Eastern Africa: A Synthesis of the Findings of 22 Case Studies. CIMMYT
Economics Working Paper 03-06. Mexico, D.F.: CIMMYT.
EEA/EEPRI (Ethiopian Economic Association/Ethiopian Economic Policy Research
Institute), 2006. Evaluation of the Ethiopian Agricultural Extension with Particular Emphasis
143
on the Participatory Demonstration and Training Extension System (PADETES). Addis
Ababa: EEA/EEPRI.
EIAR (Ethiopian Institute of Agricultural Research), 2011. Coordination of national
agricultural research system, English and Amharic versions, EIAR, Addis Ababa.
Eshetu T., Ketema M. and Kassa K., 2013. Economic impact of support service program on
micro and small enterprises: The case of Dire Dawa Administration, Ethiopia. Agris On-line
Papers in Economics and Informatics 5(1): 21-29.
FAOSTAT, 2008 and 2010.(http://faostat.fao.org) Accessed on November 5, 2012
FAO (Food and Agriculture Organization of the United Nations), 2009. How to feed the
world in 2050.(www.fao.org/fileadmin/templates/wsfs/docs/expertpaper/Howto Feed the
World in 2050) Accessed November 12, 2012.
Farrell, M.J., 1957. The measurement of productive efficiency. Journal of Royal Statistical
Society 120 (3): 253 – 290.
Feder, G., R. E. Just and D. Ziberman, 1982. Adoption of agricultural innovations in
developing countries: A survey. World Bank staff working paper no. 542, The World Bank,
Washington, D.C., USA.
Fernandez, M.D. and P.L. Nuthall, 2009. Technical efficiency in the production of sugarcane
in Central negros area, Philippines: an application of data envelopment analysis. Journal of
International Society for Southeast Asian Agricultural Sciences 15(1): 77-90.
Fischer, R.A., D. Byerlee and G.O Edmeades, 2009. Can technology deliver on the yield
challenge to 2050? FAO, economic and social development department, Rome.
Forsund, F.R., Lovell, C.A.K. and Schmidt, P., 1980. A Survey of frontier production
functions and of their relationship to efficiency measurement. Journal of Econometrics 13: 5-
25.
Gadd, H., G. Hansson and J. Mansson, 2008. Evaluating the impact of firm subsidy using a
multilevel propensity score approach. Working paper series, Center of Labor Market Police
Research Nr 3, 2009. Swedish Institute for Growth Policy Studies.
Gashaw Tadese, G.N. Francesconi and Kinde Getnet, 2014. Impact of agricultural
cooperatives on smallholders‘ technical efficiency: Empirical evidence from Ethiopia. Annals
of Public and Cooperative Economics 85(2): 257- 286.
Greene, W. H., 1993. ―The Econometric approach to efficiency analysis,‖ in The
measurement of productive efficiency, H. Fried, K. Lovell, and S. Schmidt, eds., Oxford
University Press, Oxford.
144
Greene, W.H., 1997. Frontier production functions. In Pesaran, M.H. and P. Shmidt editors,
Handbook of applied econometrics, volume II: Micro-econometrics. Blackwell Publishers
Ltd.
Greene, W.H., 2012. Econometric analysis 7th
edition. USA, Pearson education, Inc.
Gujarati, D.N., 2004. Basic econometrics 4th
edition. India, New Delhi, Tata McGraw-Hill
publishing company
Guush B., Zelekawork P., T. Kibrom., and Seneshaw T., 2011. Food grain consumption and
calorie intake patterns in Ethiopia. Ethiopia strategy support program II (ESSP II) ESSP II
working paper No. 23, International Food Policy Research Institute.
Hailu Beyene, 2008. Adoption of improved tef and wheat production technologies in crop-
livestock mixed systems in northern and western Shewa zones of Ethiopia. PhD thesis,
University of Pretoria, Faculty of natural and agricultural sciences, Pretoria.
Harris, H., Masri, S., Pala, M., Raya, J and Singh M., 2007. Rain fed wheat based rotation
under Mediterranean condition: Crop sequences, nitrogen fertilization and stubble grazing in
relation to grain and straw quality. European Journal of Agronomy 28(2):112-118.
Hassan Bashir, Bezabih Emana, Belay Kassa and Jema Haji, 2012. Determinants of chemical
fertilizer technology adoption in North Eastern highlands of Ethiopia: the double hurdle
approach. Journal of Research in Economics and International Finance 1(2): 39-49.
Hausman, J. and D. McFadden, 1984. Specification test for the multinomial logit model.
Econometrica 52 (5): 1219 – 1240.
Heckman, J., H. Ichimura, J. Smith and P. Todd, 1998. Characterizing selection bias using
experimental data. Econometrica 66: 1017–1098.
Headey, D.D., Alauddin, M. and Rao, D.S.P., 2010. Explaining agricultural productivity
growth: An international perspective. Agricultural Economics 41: 1-14.
IFAD (International Fund for Agricultural Development), 2011. Enabling rural people to
overcome poverty. (http://www. ruralpovertyportal.org) Accessed July 30, 2011.
IFPRI (International Food Policy Research Institute), 2012. Global food policy report of 2011,
IFPRI. 2033 K Street, NW Washington, DC 20006-1002, USA.
Jahnke, H.E.,1982. Livestock Production Systems and Livestock Development in Tropical
Africa. Keil Wisenschaftsverlag Vauk, Keil.
Jalan, J. and Ravallion, M., 2003. Estimating the benefit incidence of an antipoverty program
by propensity-score matching. Journal of Business and Economics Statistics, 21(1): 19-30.
145
Kaleab Kebede and Berhanu Adenew, 2011. Analysis of technical efficiency: lessons and
implications for wheat producing commercial farms in Ethiopia. Journal of Economics and
Sustainable Development: 8(2).(http://www.iiste.org) Accessed on November 05, 2012.
Kansas State University wheat research and extension. (http:// www.ksu.edu/wheatpage)
Accessed on November 5, 2012.
Kirsten, A., R. Dorward, Colin, Poulton and Nick Vink, 2009. Institutional Economics
Perspectives on African Agricultural Development, Edited by Johann F. International Food
Policy Research Institute 2033 K Street, NW Washington, D.C. 20006-1002, U.S.A.
Kislev, Y. and N.S. Bachach, 1973. The process of innovation cycle. American Journal of
Agricultural Economics 55: 28 – 37.
Kopp, R. J. and V. K. Smith, 1980. Frontier production function estimates for steam electric
generation: A comparative analysis. Southern Economic Journal 47: 1049-59.
Kothari, C.R., 2004. Research Methodology: Methods and Techniques 2nd
revised edition.
New Dheli, New Age International Publishers.
Krasuaythong, T., 2008. Technologies adoption in vegetable production in Northern
Thailand. Doctoral Dissertation, Leibniz Universitat Hannover, Germany.
Krejcie, R.V. and D.W. Morgan, 1970. Determing sample size for research activities.
Educational and Psychological Measurement 30: 607-610.
Kumbhakar, S.C., C.F. Parmeter and E.G. Tsionas, 2013. A zero inefficiency stochastic
frontier model. Journal of Econometrics 172: 66 – 76.
Langyintuo, A.S., 2008. Modelling agricultural technology adoption: Principles and concepts.
Harare, Zimbabwe, CIMMYT.
Lauren, N.K., Gallandt, E.R. and Mallory, E.B., 2012. Impact of spring wheat planting
density, row spacing, and mechanical weed control on yield, grain protein, and economic
return in Maine. Weed Science 60: 244 – 253.
Leeuw, F., 2003. Reconstructing program theories: Methods available and problems to be
solved. American Journal of Evaluation 24(1): 5–20.
Leeuw, F. and J. Vaessen, 2009. Impact evaluations and development. NONIE guidance on
impact evaluation, Washington, DC 20433. (www.worldbank.org/ieg/nonie) Accessed on
May 29, 2014.
Leigh Tesfatsion, 1980. Global and approximate global optimality of myopic economics
decisions. Journal of Economic Dynamics and Control 2: 1 – 26.
146
Maddala, G.S., 1992. Introduction to Econometrics 2nd
edition. NewYork, Macmillan
publishing company.
Majumder, M.K., L. Mozumdar and P.C. Roy, 2009. Productivity and resource use efficiency
of Boro rice production. Journal of Bangladesh Agricultural University 7(2): 247-252.
Meeusen, W. and J. van den Broeck, 1977. Efficient Estimation from Cobb-Douglas
Production Functions with Composed Error. International Economic Review 18: 435-444.
Mendola, M., 2007. Agricultural technology adoption and poverty reduction: A propensity
matching analysis for rural Bangladesh. Food Policy 32: 372 – 393.
Mesay Yami, Tesfaye Solomon, Bedada Begna, Fekadu Fufa, Tolesa Alemu and Dawit
Alemu, 2013. Source of technical inefficiency of smallholder wheat farmers in selected
watwerlogged areas of Ethiopia: A translog production function approach. African Journal of
Agricultural Research 8(29): 3930 – 3940.
Mesfin Astatikie, Belayheh Legesse and Elias Zerfu, 2012. Analysis of economic impacts of
Triticale(X- Triticosecale wittmack) adoption: The case of Farta Wereda, Ethiopia. Journal
of Scientific Research and Reviews 1(2): 020-025.
MoFED (Ministry of Finance and Economic Development), 2006. Ethiopia: Building on
Progress; a Plan for Accelerated and Sustained Development to End Poverty (PASDEP).
(2005/06-2009/10), Volume I, Addis Ababa.
MoFED (Ministry of Finance and Economic Development), 2010. Five years Growth and
Transformation Plan (GTP): The Federal Democratic Republic of Ethiopia.
Moghaddam, A., Ramroudi H., Koohkan Sh. A. and Fanaei, H.R., 2011. Effects of crop
rotation systems and nitrogen levels on wheat yield, some soil properties and weed
population. International Journal of AgriScience 1(3): 156 – 163.
Negassa, A., 2012. Determinants of adoption and spatial diversity of wheat varieities on
household farms in Turkey. International maize and wheat improvement center,
socioeconomics program working paper, CIMMYT.
Nyangena, W. and O.M. Juma, 2014. Impact of farm technology on yields: The case of
improved maize varieties and inorganic fertilizers in Kenya. Environment for development,
Discussion paper series 14-02, EfD, Kenya.
Ogisi Oraye D., Chukwuji, C.O. and Okeke D.C., 2012. Efficiency of resource use by rice
farmers in Ebonyi state, southeastern Nigeria: A data envelopment analysis. Asian Journal of
Agricultural and Rural Development 2(2): 149- 154.
Okpe, I.J., C.C. Abur and S.O. Omniyi, 2012. Resource use efficiency and rice production in
Guma local government area of Benue state: An application of stochastic frontier production
147
function. International Review of Social Sciences and Humanities 1(3): 108- 116.
(http://www.irssh.com) Accessed on July 30, 2012.
Oromiya Regional State, 2011. Arsi zone profile: physical and socioeconomic profile of Arsi
zone and districts. (http://www.oromiyaa.com) Accessed on November 5, 2012.
Perloff, J.M., 2008. Microeconomics: Theory and applications with calculus. Pearson
Education, Inc., New York.
Rada, N.E. and Buccola, S.T., 2011. Agricultural policy and productivity: evidence from
Brazilian censuses. Agricultural Economics 43: 355-367.
Raitzer, D.A. and T.G. Kelly, 2008. Assessing the contribution of impact assessment to donor
decisions for international agricultural research. Research Evaluation 17(3): 187-199.
Rashid, S., 2010. Staple Food Prices in Ethiopia. A paper prepared for the COMESA policy
seminar on ―Variation in staple food prices: Causes, consequence, and policy options‖,
Maputo, Mozambique, 25-26 January 2010, under the African Agricultural Marketing Project
(AAMP).
Rogers, E.M., 1983. Diffusion of innovation, New York, The Free Press.
Rogers, P.J., T.A. Hacsi, A. Petrosino, and T.A. Huebner (eds.), 2000. Program theory in
evaluation: Challenges and opportunities, new directions for evaluation 87, Jossey-Bass, San
Francisco.
Rogers, P.J., 2012. Introduction to impact evaluation. Impact evaluation notes no. 1, RMIT
University, Australia. (www.interaction.org/impact-evaluation notes) Accessed on May 29,
2014.
Rosegrant, M.W. and Agcaoili, M., 2010. Global food demand, supply, and price prospects to
2010. International Food Policy Research Institute, Washington, D.C. USA.
Rosembaum, P.R. and Rubin, D.B., 1983. The central role of the propensity score in
observational studies for causal effects. Biometrika, 70(1): 41–55.
Smith, J.A. and Todd, P.E., 2005. ―Does matching overcome Lalonde‘s critique of
nonexpermental estimator?‖ Journal of Econometrics 125: 305 – 353
Solomon Asfaw, Bekele Shiferaw, Simtow, F. and Mekbib Gebretsadik, 2011. Agricultural
technology adoption, seed access constraints and commercialization in Ethiopia. Journal of
Development and Agricultural Economics 3(9), 436 – 447.
Storck, H., Brehanu Adnew, Bezabih Emana, A. Borowiecki and Shemelis Weldehawariat,
1991. Farming system and farm management practices of smallholders in the Hararge
148
highlands. Farming system and resource economics in the topics, vol 11. Kiel:
Wirtschaftsverlag Vauk.
Sunding, D. and D. Zilberman, 2000. The agricultural innovation process: Research and
technology adoption in a changing agricultural sector. Handbook of Agricultural Economics,
University of California, Berkeley, USA.
Taylor, T.G., H.E. Drummond and A.T. Gomes, 1986. Agricultural credit programs and
production efficiency: an analysis of traditional farming in southeastern Minas Gerais, Brazil.
American Journal of Agricultural Economics 68:110-119.
Teame Hailemariam, 2011. Factors determining fertilizer adoption of peasant farm sector in
Northern Ethiopia, Tigray. MSc. Thesis, Norwegian Unversity of Life Sciences, department
of economics and resource management.
Timmer, P., 1971. Using a probabilistic frontier production function to measure technical
efficiency. Journal of Political Economy 79: 776–794.
Todo, Y. and R. Takahashi, 2011. Impact of farmers field schools on agricultural income and
skills. Evidence from an aid-funded project in rural Ethiopia. Japan international cooperation
agency research institute, Tokyo, Japan.
Trochim, W.M.K., 1989.An introduction to concept mapping for planning and evaluation.
Evaluation and Program Planning 12: 1–16.
Uaiene, R.N and C. Arndt, 2009. Farm household efficiency in Mozambique: A stochastic
frontier analysis, national directorate of studies and policy analysis, ministry of planning and
development, Mozambique.
Udry, C., 1996. Gender, agricultural production, and the theory of the household. Journal of
Political Economy 104 (5): 1010-1046
Udry, C., J. Haddinott , H. Alderman and L. Haddad .1995. Gender differentials in farm
productivity: Implications for household efficiency and agricultural policy. Food Policy 20
(5): 407-423
Vanclay, F. and G. Lawrence, 1994. Farmer rationality and the adoption of environmentally
sound practices: A critique of the assumptions of traditional agricultural extension. European
Journal for Agricultural Education and Extension 1 (1): 59–90
Vandercasteenlen, J., Mekdim D., Mintlen B. and Alemayehu S., 2013. The impact of the
promotion of row planting on farmers‘ teff yields in Ethiopia. Ethiopia Strategy Support
Program II, ESSP research note 27. International Food Policy Research Institute.
Varian, H. R., 1984. The non parametric approach to production analysis. Econometrica
52(3): 579 – 598.
149
Varian, H.R., 2005. Intermediate microeconomics: A modern approach 7th
edition, Pearson
Education, Inc., New York.
Yenealem Kassa, Fekadu Beyene, Jema Haji and Belaineh Legesse, 2013. Impact of
integrated soil and water conservation program on crop production and income in west
Harerghe Zone, Ethiopia. International Journal of Environmental Monitoring and Analysis.
1(4): 111-120
Yin, R., 2000. Alternative measurements of productive efficiency in the global bleached
softwood pulp sector. Forest Science 46: 558-569.
Yohannis Mengesha and Jema Haji, 2003. Socioeconomic impact of HIV/AIDS on Ethiopian
households: The Case of Dire Dawa Administration. International Researcher 2 (1): 117 –
124.
Yu, B., A. Nin-Pratt, J. Funes and Sinafikeh Asrat, 2011. Cereal production and technology
adoption in Ethiopia. Ethiopia Strategy Support Program II (ESSP II), ESSP II Working
Paper 31, International Food Policy Research Institute.
WB (World Bank), 2009. World development report 2009.
Wooldridge, J.M., 2010. Econometric Analysis of Cross Section and Panel Data, 2nd
edition.
USA, The MIT press.
150
APPENDICES
151
Appendix A. Descriptive statistics of some socioeconomic variables of households
Appendix Table 1. Total area cultivated, production and yield of grain crops for private holdings in Meher season of 2012/13
Grain Crops
Ethiopia Oromia Arsi
Area in
hectare
Production,
ql
Yield
(ql/ha)
Area in
hectare
Production,
ql
Yield
(ql/ha)
Area in
hectare
Production,
ql
Yield
(ql/ha)
Cereals 9,601,035 196,511,515 20 4,486,163 99,568,008 22 496,224 11,350,632 23
Pulses 1,863,445 27,510,312 15 734,045 11,683,119 16 74,958 1,355,730 18
Oil crop 818,449 7,266,644 9 378,563 3,430,075 9 24,058 315,088 13
Total 12,282,90 231,288,472 19 5,598,772 114,681,202 20 595,239 13,021,451 22
Cereal crops
Teff 2,730,273 37,652,412 14 1,256,565 17,535,597 14 81,785 1,153,111 14
Barley 1,018,753 17,816,522 17 448,545 8,977,418 20 93,757 2,270,313 24
Wheat 1,627,647 34,347,061 21 872,972 20,262,900 23 209,219 5,115,027 24
Maize 2,013,045 61,583,176 31 1,115,957 35,908,457 32 77,682 1,987,977 26
Sorghum 1,711,485 36,042,620 21 675,657 14,898,157 22 31,816 798,179 25
Finger millet 431,507 7,422,971 17 92,307 1,532,060 17 0 0
Oats/'Aja' 26,514 436,338 16 21,889 378,062 17 1,943 25,370 13
Rice. 41,811 1,210,416 29 2,270 75,358.24 33 0 0
Total 9,601,035 196,511,515 20 4,486,163 99,568,008 22 496,224 11,350,632 23
Source: CSA, 2013
152
Appendix Table 2. Age, educational status and farming experience of household heads
Household head
characteristics
Dodota Hetosa Lemu-Bilbilo
F-value P-value Mean
Std.
Deviation Mean
Std.
Deviation Mean
Std.
Deviation
Age ( years) 43.34 10.46 48.8 10.63 46.52 11.8 6.19*** 0.0023
Educational
status (grades)
5.41 3.36 4.56 4.16 4.69 3.39
1.51 0.2223
Farming ( years) 22.47 10.4 27.87 10.64 25.58 11.44
6.25*** 0.0021
***p < 0.01
153
Appendix Table 3. Distribution of marital status of sample household heads by district
Marital status
District
Total Dodota Hetosa Lemu Bilbilo
Married
Number 78 114 153 345
District (%) 94.0 85.7 92.7 90.6
Total (%) 20.5 29.9 40.2 90.6
Single
Number 0 2 2 4
District (%) 0.0 1.5 1.2 1.0
Total (%) 0.0 .5 .5 1.0
Divorced
Number 3 4 4 11
District (%) 3.6 3.0 2.4 2.9
Total (%) .8 1.0 1.0 2.9
Widowed
Number 2 13 6 21
District (%) 2.4 9.8 3.6 5.5
Total (%) .5 3.4 1.6 5.5
Total Number 83 133 165 381
154
Appendix Table 4. Descriptive statistics of household size by district
Household
family
size
Dodota Hetosa Lemu-Bilbilo
Minimum Maximum Mean
Std.
Dev Minimum Maximum Mean
Std.
Dev Minimum Maximum Mean
Std.
Dev
Male 0.00 9.00 3.43 1.78 1.00 10.00 3.35 1.58 0.00 13.00 3.52 1.72
Female 0.00 7.00 3.18 1.54 0.00 8.00 2.73 1.29 1.00 12.00 3.03 1.61
Total 1.00 14.00 6.63 2.47 1.00 15.00 6.08 2.32 2.00 20.00 6.53 2.48
155
Appendix Table 5. Land use by district (hectares)
Land use
type
Dodota Hetosa Lemu-Bilbilo
Min Max Mean
Std.
Dev Min Max Mean
Std.
Dev Min Max Mean
Std.
Dev
Total 0.50 6.00 2.29 1.48 0.25 6.75 2.19 1.10 0.35 12.00 3.21 1.96
Agricultural 0.25 5.50 2.02 1.37 0.15 6.25 1.87 0.95 0.25 8.00 2.23 1.35
Grazing 0.00 0.50 0.05 0.14 0.00 0.75 0.13 0.14 0.00 6.00 0.73 0.72
Woodlot 0.00 0.25 0.01 0.04 0.00 0.25 0.00 0.02 0.00 0.50 0.02 0.08
Homestead 0.00 0.50 0.23 0.14 0.00 0.50 0.16 0.09 0.00 1.00 0.25 0.18
156
Appendix Table 6. Number and proportion of sample households planted wheat in row and
broadcast
Method of planting
Name of district
Total Dodota Hetosa Lemu-Bilbilo
Broadcast
Number
49
107
99
255
Percent within
district 59.0 80.5 60.0 66.9
Percent of total 12.9 28.1 26.0 66.9
In row
Number 34 26 66 126
Percent within
district 41.0 19.5 40.0 33.1
Percent of total 8.9 6.8 17.3 33.1
Appendix Table 7. Average productivity per hectare for wheat planting methods, in quintals
Method of planting Dodota Hetosa Lemu-Bilbilo Total
Broadcast 15.74 30.14 23.35 24.73
In row 17.37 41.23 27.95 27.83
Average yield gain (%) 10.31 36.82 19.70 12.53
157
Appendix Table 8. Livestock population conversion factor into TLU
Livestock Type TLU Conversion Factors
Cattle 0.7
Sheep 0.1
Goat 0.1
Mule 0.7
Horse 0.8
Camel 1
Chicken 0.01
Pig 0.2
Donkey 0.5
Source: Jahnke, 1982
Appendix Table 9. Conversion factors for adult equivalents
Age group (years) Male Female
< 10 0.6 0.6
10 - 13 0.9 0.8
14 -16 1.0 0.75
17 -50 1.0 0.75
> 50 1.0 0.75
Source: Storck et al., 1991.
158
Appendix Table 10. Thermal zone classification of Arsi zone
Altitude (meters
above sea level)
Annual mean
temperature
Descriptions Area (%)
< 500 >250C Warm
500 – 1,500 20 – 250C Moderately warm 20
1,500 – 2,300 15 – 200C Moderately cool 40
2,300 – 3,200 10 – 150C Cool 34
> 3,200 <100C Cold 6
Source: www.oromiya.com, 25 January 2013.
159
Appendix Table 11. Sensitivity analysis for the impact of row planting in Dodota district
Gamma sig+ sig- t-hat+ t-hat- CI+ CI-
1 2.80E-09 2.80E-09 0.5 0.5 0.5 0.5
1.25 9.20E-08 3.50E-11 0.5 0.5 -3.90E-07 0.5
1.5 9.60E-07 4.60E-13 0.5 0.5 -3.90E-07 0.5
1.75 5.20E-06 6.10E-15 0.5 0.5 -3.90E-07 0.5
2 0.000019 1.10E-16 -3.90E-07 0.5 -3.90E-07 0.5
* gamma - log odds of differential assignment due to unobserved factors
sig+ - upper bound significance level
sig- - lower bound significance level
t-hat+ - upper bound Hodges-Lehmann point estimate
t-hat- - lower bound Hodges-Lehmann point estimate
CI+ - upper bound confidence interval (a= .95)
CI- - lower bound confidence interval (a= .95)
160
Appendix Table 12. Sensitivity analysis for the impact of row planting in Hetosa district
Gamma sig+ sig- t-hat+ t-hat- CI+ CI-
1 1.70E-07 1.70E-07 -3.70E-07 -3.70E-07 -3.70E-07 -3.70E-07
1.25 2.50E-06 6.00E-09 -3.70E-07 -3.70E-07 -3.70E-07 0.5
1.5 1.60E-05 2.10E-10 -3.70E-07 -3.70E-07 -3.70E-07 0.5
1.75 5.80E-05 7.60E-12 -3.70E-07 -3.70E-07 -3.70E-07 0.5
2 0.000156 2.80E-13 -3.70E-07 0.5 -3.70E-07 0.5
* gamma - log odds of differential assignment due to unobserved factors
sig+ - upper bound significance level
sig- - lower bound significance level
t-hat+ - upper bound Hodges-Lehmann point estimate
t-hat- - lower bound Hodges-Lehmann point estimate
CI+ - upper bound confidence interval (a= .95)
CI- - lower bound confidence interval (a= .95)
161
Appendix Table 13. Sensitivity analysis for the impact of row planting in Lemu-Bilbilo district
Gamma sig+ sig- t-hat+ t-hat- CI+ CI-
1 2.20E-16 2.20E-16 0.5 0.5 0.5 0.5
1.25 1.80E-13 0 0.5 0.5 -3.80E-07 0.5
1.5 1.60E-11 0 0.5 0.5 -3.80E-07 0.5
1.75 4.10E-10 0 0.5 0.5 -3.80E-07 0.5
2 4.60E-09 0 -3.80E-07 0.5 -3.80E-07 0.5
* gamma - log odds of differential assignment due to unobserved factors
sig+ - upper bound significance level
sig- - lower bound significance level
t-hat+ - upper bound Hodges-Lehmann point estimate
t-hat- - lower bound Hodges-Lehmann point estimate
CI+ - upper bound confidence interval (a= .95)
CI- - lower bound confidence interval (a= .95)
162
Appendix Table 14. Tests of mean and variance differences among districts for some
variables
Household characteristics Total Dodota Hetosa
Lemu-
Bilbilo
F- value or
chi2-value
Prob >
F or chi2
Age of Household head 46.62 43.34 48.8 46.52 6.19*** 0.0023
Educational level of
household head
4.8 5.41 4.56 4.69 7.76** 0.021
Farming experience of
household head
25.7 22.47 27.87 25.58 6.25*** 0.0021
Total household size 6.4 6.63 6.08 6.53 1.72 0.1807
Household size in adult
equivalent 4.16 4.18 4.23 4.09 0.26 0.7737
Land holding size (ha) 2.65 2.29 2.19 3.21 17.35*** 0.0000
Man-days per hectare 70.58 41.2 54.89 98 94.97*** 0.0000
Wheat area cultivated (ha) 1.1 1.6 1.6 0.5 60.31*** 0.0000
Total output of wheat (ql) 28.44 25.45 50.19 12.43 59.74*** 0.0000
Wheat yield (ql/ha) 24.95 15.63 30.89 24.85 152.62*** 0.0000
Output value (birr/ha) 15419 10234 19251 14938 137.61*** 0.0000
Wheat sold(ql) 13.78 12.12 26.06 4.71 37.31*** 0.0000
Livestock in TLU 7.28 4.18 5.13 10.58 67.54*** 0.0000
Oxen owned 2.87 2.33 2.71 3.27 11.33*** 0.0000
Chemical fertilizers
(ql/ha) 0.98 0.74 0.90 1.16 51.95*** 0.0000
Seed and pesticide (kg/ha) 734 766 938 555 46.92*** 0.0000
Cost per hectare (ln) 9.24 9.11 9.32 9.24 63.85*** 0.0000
Fertilizer price (ln) 2.69 2.68 2.69 2.69 12.04*** 0.0000
Seed/pesticide price (ln) 1.48 1.67 1.55 1.32 53.56*** 0.0000
Land price per hectare (ln) 8.09 8.04 8.23 7.98 6339*** 0.0000
Wage rate/day (ln) 3.52 3.79 3.61 3.32 354.77*** 0.0000
Output in kg/ha (ln) 7.76 7.31 8.01 7.78 175.87*** 0.0000
Number of wheat plots 1.97 1.91 2.80 1.33 58.31*** 0.0000
163
Crop types 4.34 4.24 4.36 4.37 30.12*** 0.0000
Off-farm income (000
birr) 7.64 6.08 4.64 10.85 21.79*** 0.0000
Profit from wheat (birr/ha) 5012 1083.91 8039.89 4547.29 81.96*** 0.0000
Technical efficiency 0.749 0.569 0.82 0.784 159.72*** 0.0000
Allocative efficiency 0.855 0.896 0.888 0.878 2.5* 0.0838
Economic efficiency 0.665 0.512 0.729 0.69 105.09*** 0.0000
*p< 0.1, **p < 0.05, ***p < 0.01.
164
Appendix B. Survey questionnaire for field data collection
1. General Information
1.1. Household ID _____________
1.2. Name of data enumerator _____________________________
1.3. Date of interview ____________________________________
1.4. Name of district (wereda) ____________________________
1.5. Name of PA(kebele) _________________________________
1.6. Agro-ecological setting of district(wereda):
1) High land 2) Mid land 3) Low land
1.7. Mean annual rainfall of 2012 of the district in mm ____________
1.8. Average rainfall of 2012 cropping season (months) in mm _______________
2. Household Information:
2.1. Name of household head ________________________________
2.2. Age of household head (HHH) ___________________
2.3. Sex of HHH: 0) female 1) male
2.4. Marital status of HHH: 1) married 2) Single 3) divorced 4) Widowed
2.5. Educational status of HHH:
1) Illiterate=0 2) read and write =1 3) number of grades completed ___________
2.6 Main occupation of HHH:
1) Crop production 2) Crop production and animal rearing 3) government job
4) Off-farm activities
2.7. Farming experience of household head: _______________ years
2.8. Educational status of family members(number of family members):
Number of illiterate ___________ High school:__________
Primary school: _____________ Tertiary education:_______
Junior school:______________
2.9 Family size in number:
Total family size: ________ Male: _________ Female: _____________
165
2.10 Age structure of household:
Age class Male Female Total
Less than 5 years
5 to 14 years
15 to 64 years
Above 64 years
Total
2.11. Family members‘ involvement in crop production activities:
Age class Number of Males Number of Females
Full time Part-time Full time Part-time
Less than 5 years
5 to 14 years
15 to 64 years
Above 64 years
Total
3. Land holding size (in 2012):
3.1 Total privately owned land in hectares: ___________
3.2 Land use type (in 2012):
Privately owned land use Area in hectares
1 Agricultural land
2 Grazing land
3 Woodlot(forest land)
4 Homestead area (residential area)
5 Other land, specify_________
Total
3.3 Rented out agricultural land in 2012__________ ha. Rented in agri. land ________ ha.
4. Cropping pattern and crop outputs in 2012
166
4.1 Types, area cultivated, output and price of outputs of crops (in main season of 2012)
Crop type
Area
cultivated
in hectare
Total
output in
quintal
Average price
per quintal in
March (birr)
Total
revenue
1 Wheat variety
1.1 Hawii (HAR 2501)
1.2 Pavon-76
1.3 Kubsa (HAR 1685)
1.4 Kekaba
1.5 Danda‘d
1.6 Hoggana
1.7 Shorima
1.8 Meda-Walabu (HAR 1480)
1.9 Galama (HAR 604)
1.18
1.19
1.20
Total
2 Barley
2.1 Malt barley
2.2 Food barley
Total
3 Faba beans (bakela)
4 Field peas (ater)
5 Chick pea (shimbra)
6 Lentil (msir)
7 Linseed (telba)
8 Rapeseed (gomen)
9 Tef
10 Maize
167
11 Sorghum
12 Potato
13 Tomato
14 Carrot
15 Beetroot(key sir)
16 Cabbage
17 Onion
18 Garlic (nech shinkurt)
19
20
4.2. Total number of different crops (wheat, barley, maize, peas, etc. not variety of specific
crop) cultivated in 2012? __________________
4.3. Do you rotate planting of cereal crops on same land with pulses or other crops every
year?
0) No 1) Yes
4.4. How many times did you rotate pulses with cereal crops in the past 5 years (2007-2011)
on the largest plot of land planted with wheat in 2012? ________________ times
4.5. What are the reasons for not using crop rotation? 1) Lack of seed 2) Weed
problem 3) Disease problem 4) Land shortage 5) Price advantage 6) harvesting
and threshing advantage 7) _____________________________
4.6. Total wheat area cultivated in 2012 ____________________ ha
4.7. How many wheat plots did you have in 2012? _______________ plots.
4.8. What is the area of your largest wheat plot in 2012? ______________ hectares.
4.9. Which wheat variety did you plant on this largest wheat plot in 2012? _______________
4.10. How many quintals of wheat did you get from this largest plot? ____________ ql
4.11. Let‘s take the largest wheat plot of 2012 and record the type of crops planted on it over
the past 5 years
168
Year
2012
(2005 E.C)
2011
(2004 E.C)
2010
(2003 E.C)
2009
(2002 E.C)
2008
(2001 E.C)
2007
(2000 E.C)
Crop
planted on
same land
Wheat
Cross check with (4.4)
4.12. What was the precursor crop to wheat on this largest wheat farm (crop planted in 2011)?
1) Pulses 2) Oil crops 3) Vegetables (horticulture) 4) Maize 5) other cereal
4.13. What was the nature of wheat planting on this largest plot in 2012?
0) broadcast 1) in row
4. 14. Do you have awareness on the usefulness of wheat row planting? 0) No 1) Yes
4. 15. Did you practice row planting of wheat in 2012? 0) No 1) Yes
If yes, did you weed by hand the row planted wheat? 0) No 1) Yes
Did you hoe the row planted wheat? 0) No 1) Yes
4. 16. Did you make early hand weeding of row planted wheat? 0) No 1) Yes
4. 17. How do you manage weed problem? 1) Hand weeding 2) by herbicides 3)
both hand weeding and use of herbicides
169
5. Farm inputs utilization in wheat production
5.1 Quantity of seed, fertilizers and chemicals used in wheat production in 2012
N0 Inputs used for wheat
cultivation
Amount used
on all wheat
plots
Amount used
on the largest
wheat plot
Price in birr per
quintal, or liter for
chemicals
1 Wheat seed in Kg
2 Fertilizers:
DAP in Kg
UREA in Kg
DAP + UREA in Kg
Manures in quintals
3 Chemicals:
Herbicides in local ‗melekiya’
Fungicides in local ‗melekiya’
5.2. Labor utilization on the largest wheat plot in 2012
Activity Family labor Hired labor
Children Adults No Hours
worked
Total
Cost
paid
Plowing by
oxen
Number
involved
Hours
worked
Male Female
No hours No hours
1st plowing
2nd
plowing
3rd
plowing
4th
plowing
5th
plowing
Planting
Total
Hand
Weeding
170
1st weeding
2nd
weeding
3rd
weeding
Total
Hoeing
1st hoeing
2nd
hoeing
3rd
hoeing
Total
Harvesting
by hand &
transporting
Threshing
5.3. What is the average wage of a daily laborer for a full day working hours in birr? During:
Plowing ____________ birr; Planting ____________ birr; weeding ____________ birr;
Harvesting ___________ birr; threshing ____________ birr
5.4. Do you have your own oxen for plowing? 0) No 1) Yes
5.5. How many oxen do you have? ______________
5.6 If you do not have oxen, what is the cost of hiring a pair of oxen only for plowing one
timad (0.25 hectares) of land? ____________ Birr
5.7. Did you use tractor for plowing your wheat land in 2012? 0) No 1) Yes
If yes, what was the cost for plowing one timad by tractor? ___________ Birr
5.8. Did you use combine harvester for harvesting and threshing of wheat in 2012?
0) No 1) Yes
If yes, what was the cost of combine harvester per one quintal of wheat output? _______ Birr
5.9. Do you have cash constraint for purchasing farm inputs and for farming activities?
0) No 1) Yes
1)
171
5.10. Did you get credit services for agricultural activities in 2012?
0) No 1) No
5.11. Do you need agricultural extension services? 0) No 1) Yes
5.12. Did you get extension services in wheat production in 2012?
0) No 1) Yes
5.13. Do you have access to chemical fertilizers? 0) No 1) Yes
If yes, do you get the required amount? 0) No 1) Yes
5.14. Do you have access to herbicides and fungicides? 0) No 1) Yes
If yes, do you get the required amount with affordable prices? 0) No 1) Yes
5.15. Do you have access to improved wheat seed? 0) No 1) Yes
5.16. Have you been faced weeds and diseases problems on your wheat farms?
0) No 1) Yes
5.17.What was the scale of weeds and diseases problems on your wheat farm in the past five
years? _________
Use scale: 1 for no weeds and diseases problems; 2 for less problem; 3 for medium problem; 4
for higher problem; 5 for severe or very high weeds and disease problems
172
6. Livestock holding size
No Type of animal Number of local
breed
Number of improved
breed
Total
1 Cows
2 Oxen
3 Bull (weyfen)
4 Heifer (gider)
5 Calf (tija)
6 Goats
7 Sheep
8 Horses
9 Donkeys
10 Mules
11 Camels
12 Chicken
7. Income of household in the last one year (2012)
NO Source Annual income in birr
1 Off-farm activities
2 Farm activity for other individual
3 Sale of livestock
4 Sale of livestock products
5 Sale of firewood
6 Remittance
7 Others________________