THREE ESSAYS ON LAND AND AN INTENSIVE FARMING SYSTEM IN
SUB-SAHARAN AFRICA: EVIDENCE FROM KENYA
By
Rie Muraoka

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
Agricultural, Food, and Resource Economics – Doctor of Philosophy
2015

ABSTRACT
THREE ESSAYS ON LAND AND AN INTENSIVE FARMING SYSTEM IN SUBSAHARAN AFRICA: EVIDENCE FROM KENYA
By
Rie Muraoka
In light of the high rate of population growth and rapid urbanization, the arable land
frontier in Sub-Saharan Africa countries has been exhausted and the land-labor ratio has been
shrinking. Under this setting, access to land for smallholders and the landless and improvement
of land productivity have critical implication for poverty reduction and food insecurity. This
dissertation is composed of three essays that study the determinants and impacts of land access,
land distribution, and an intensified farming system in the context of Sub-Saharan Africa using
household survey data from rural Kenya.
The first essay utilizes household and parcel-level data from rural Kenya to explore the
linkage between land access and food security. We find that a 10% increase in operated land size
would increase household total food consumption per capita, cereal consumption per capita, noncereal consumption, and home produced food consumption by 2.6%, 2.1%, 2.7% and 5.4%,
respectively. We also find that land rental is the dominant mechanism that poor rural farmers use
to access additional land for cultivation. However, the levels of long-term land investment
(measured by application of organic fertilizer) and land productivity are significantly lower for
rented parcels than for own parcels even after household fixed effect and parcel-level observed
characteristics are controlled for. Furthermore, the amount of land actually rented in or out is
found to be significantly below the amount of land desired if the land rental market is
functioning perfectly. These findings point to the existence of problems with land rental markets
that impede their ability to fully contribute to national food security and poverty reduction goals.

The second essay aims to explore the determinants of the new maize farming system,
which is characterized by adoption of high-yielding maize varieties, application of chemical
fertilizer and manure produced by stall-fed improved dairy cows, and intercropping, especially
the combination of maize and legumes, and its impact on land productivity and household
income. We examine not only the impacts of new technologies and production practices but also
the impacts of the entire new maize farming system by generating an agricultural intensification
index based on a principal component analysis. Our estimation results show that an increase in
sub-location level population density and a decrease in the land-labor ratio of an individual
household accelerate farming intensification, and that adoption of each new technology and
production practice has positive and significant impacts on land productivity. These findings are
further supported by the significantly positive impacts of the agriculture intensification index on
land productivity.
The third essay attempts to assess the determinants of agricultural land distribution and
the effects of land distribution on agricultural productivity, income and poverty based on microlevel long panel data in Kenya. The estimation results show that the village level population
density is negatively correlated with the village level Gini coefficient of own farmland. The Gini
coefficient of agricultural land is found to have significant and negative impacts on agricultural
productivity, income and poverty. Specifically, an increase in the Gini coefficient by 0.1 would
reduce the value of crop production per acre, net crop income per acre, net crop income per
capita, the net livestock income per capita, the net non-farm income per capita, the net total
income per capita, and chance of being out of poverty by 4.3%, 5.1%, 4.1%, 6.6%, 5.0%, 3.8%
and 2.8%, respectively.

ACKNOWLEDGEMENTS

This work would not have been possible without the assistance of my major advisor,
Songqing Jin. I am sincerely grateful for all the guidance and support Songqing has provided
over my years as a graduate student. I am also grateful to my dissertation committee members,
Eric Crawford, Thomas Jayne and Jeffrey Wooldridge, whose insightful comments and
suggestions have greatly improved my dissertation, as well as my approach to research. This
dissertation would not have been possible without the support of Keijiro Otsuka and Tomoya
Matsumoto, my mentors and collaborators from the National Graduate Institute for Policy
Studies, who offered feedback on my research ideas and empirical results and also provided
research funding.
The data come from the Tegemeo Agricultural Monitoring and Policy Analysis
(TAMPA) project, administered by the Tegemeo Institute of Agricultural Policy and
Development at Egerton University and Michigan State University, and from the Research on
Poverty, Environment, and Agricultural Technology (RePEAT) project, administered by the
National Graduate Institute for Policy Studies and the Tegemeo Institute at Egerton University.
A special thanks to Jordan Chamberlin, Thomas Jayne, Tomoya Mastumoto, Milu Muyanga,
Kiprono Philemon, and Takashi Yamano for making the data available for this dissertation.
I would like to thank my fellow graduate students, faculty, and staff in the Department of
Agricultural, Food, and Resource Economics for their company and help throughout my time at
Michigan State University. Finally, I reserve my special appreciation to my family. I always
appreciate their unwavering love and support.

iv

TABLE OF CONTENTS

LIST OF TABLES ........................................................................................................................ vii
LIST OF FIGURES ....................................................................................................................... ix
INTRODUCTION .......................................................................................................................... 1
REFERENCES ......................................................................................................................... 7
CHAPTER 1: LAND ACCESS, LAND RENTAL AND FOOD SECURITY: EVIDENCE
FROM KENYA .............................................................................................................................. 9
1.1 Introduction ......................................................................................................................... 9
1.2 Background ....................................................................................................................... 11
1.2.1 Land Rental and Sales Markets in the Presence of Market Failures ........................ 11
1.2.2 Empirical Evidence on Land Rental Markets .......................................................... 13
1.2.3 Land Rental Markets in Kenya ................................................................................ 14
1.3 Estimation Strategy ........................................................................................................... 15
1.4 Data Source and Descriptive Evidence ............................................................................. 19
1.4.1 Data .......................................................................................................................... 19
1.4.2 Modes of Land Transfer........................................................................................... 21
1.4.3 Land Access and Household Characteristics ........................................................... 21
1.4.4 Land Access, Production, Income and Food Security ............................................. 23
1.4.5 Land Tenure Status, Value of Crop Production, Crop Income, and Input Use ..........
Intensity............................................................................................................................. 25
1.5 Econometric Results ......................................................................................................... 26
1.5.1 Land Access and Food Security............................................................................... 27
1.5.2 Land Access, Production, and Income ..................................................................... 30
1.5.3 Land Rental and Value of Crop Output, Crop Income per Acre, and Input Use ........
Intensity............................................................................................................................. 31
1.5.4 Determinants of Net Land Leasing In and Out ........................................................ 36
1.6 Conclusions and Policy Implication ................................................................................. 39
APPENDIX ............................................................................................................................. 41
REFERENCES ....................................................................................................................... 44
CHAPTER 2: THE POSSIBILITY OF A MAIZE GREEN REVOLUSION IN KENYA: AN
ASSESSMENT OF AN EMERGING INTENSIVE FARMING SYSTEM ................................ 49
2.1 Introduction ....................................................................................................................... 49
2.2 Background Information ................................................................................................... 51
2.3 Descriptive Analysis ......................................................................................................... 53
2.3.1 Data .......................................................................................................................... 53
2.3.2 Maize and Milk Production in Kenya ...................................................................... 55
2.4 The Agriculture Intensification Index ............................................................................... 59
2.5 Estimation Strategy ........................................................................................................... 64
v

2.5.1 Determinants of the New Maize Farming System Adoption ................................... 64
2.5.2 Impact of the New Maize Farming System on Agricultural Production ................. 66
2.6 Estimation Results ............................................................................................................ 68
2.6.1 Determinants of Adoption of the New Maize Farming System .............................. 68
2.6.2 Impact of the New Maize Farming System on Agricultural Production ................. 70
2.6.3 Impact of the New Maize Farming System on Profit .............................................. 77
2.7 Conclusion and Policy Implication ................................................................................... 80
APPENDIX ............................................................................................................................. 82
REFERENCES ....................................................................................................................... 85
CHAPTER 3: THE EFFECT OF LAND DISTRIBUTION ON INCOME AND POVERTY
REDUCTION: EVIDENCE FROM KENYA .............................................................................. 88
3.1 Introduction ....................................................................................................................... 88
3.2 Background ....................................................................................................................... 91
3.2.1 Why Does Land Inequality Affect Economic Growth and Poverty? ...................... 91
3.2.2 Land Distribution in Kenya ..................................................................................... 94
3.3 Estimation Strategy ........................................................................................................... 95
3.4 Data and Sample Characteristics ...................................................................................... 98
3.4.1 Household Characteristics ..................................................................................... 101
3.4.2 Population Density and Land Distribution in Kenya ............................................. 103
3.5 Estimation Results .......................................................................................................... 108
3.5.1 Determinants of Land Distribution ........................................................................ 108
3.5.2 Effects of Land Distribution on Agricultural Land Productivity ........................... 108
3.5.3 Effects of Land Distribution on Income and Poverty ............................................ 109
3.5.4 Heterogeneous Effects of Land Distribution on Income and Poverty ................... 113
3.6 Conclusion and Policy Recommendation ....................................................................... 114
REFERENCES ..................................................................................................................... 117

vi

LIST OF TABLES

Table 1.1 Frequency of households’ acquisition of land parcels, by mode of acquisition ........... 21
Table 1.2 Household characteristics by land rental market status in 2007 ................................... 22
Table 1.3 Household incomes by operated farm size category in 2007 ....................................... 24
Table 1.4 Household food security status by operated land size category in 2007 ...................... 24
Table 1.5 Tenure statuses, value of output per acre, and input use intensity by parcel type in 2007
....................................................................................................................................................... 26
Table 1.6 Impact of and access on per capita food consumption (household fixed effect model) 28
Table 1.7 Impact of and access on per capita food consumption (household IV model) ............. 29
Table 1.8 Impact of land access on production and agriculture income (household fixed effect
and IV model) ............................................................................................................................... 30
Table 1.9 Impact of land tenure on value of output per acre (household fixed effect model, parcel
level data in 2007) ......................................................................................................................... 33
Table 1.10 Impact of land tenure on fertilizer use (household fixed effect Tobit model, parcel
level in 2007) ................................................................................................................................ 34
Table 1.11 Determinants of net land leased-in (maximum likelihood estimates, 2007 data) ....... 35
Table A1.1 The first stage regression of Table 1.7 and 8 ............................................................. 42
Table A1.2 Determinants of land leased-in and leased-out (probit model-marginal effect at mean,
2007 data)...................................................................................................................................... 43
Table 2.1 Sample household and sub-location characteristics...................................................... 54
Table 2.2 Crop production of the maize plots per cropping season.............................................. 57
Table2.3 Means of yield and fertilizer application by seed type in the maize plots per cropping
season in 2012 ............................................................................................................................... 58
Table 2.4 Milk production per household in a year ...................................................................... 59
Table 2.5 Factor loading for maize production intensification index in maize plots ................... 61
vii

Table 2.6 Mean of crop production by quartile of the agriculture intensification index in maize
plots in 2012.................................................................................................................................. 63
Table 2.7 Estimation results of the determinants of input intensification per cropping season
(parcel fixed effects model, plot level data) .................................................................................. 69
Table 2.8 Estimation results of the effects of input intensification on crop production per
cropping season (parcel and parcel-year fixed effects models, plot level data) ........................... 72
Table 2.9 Estimation results of the effects of the intensification index on crop production per
cropping season (parcel and parcel-year fixed effects models, plot level data) ........................... 73
Table 2.10 Estimation results of the effects of the intensification index on agriculture production
per year (household fixed effects model, household level data) ................................................... 75
Table 2.11 Estimation results of the effects of the intensification index on non-farm and total
household income per year (household fixed effects model, household level data) .................... 76
Table 2.12 Crop production of the largest pure-stand maize plot or the largest intercropped maize
plot in the main cropping season .................................................................................................. 78
Table 2.13 Estimation results of the effects of the intensification index on crop production in the
main cropping season (parcel fixed effects models, the largest maize plot level data) ................ 79
Table A2.1 Pairwise correlation coefficients matrix of input use................................................. 83
Table A2.2 Estimation results of the effects of input intensification on crop production per
cropping season (parcel and parcel-year fixed effects models, plot level data) ........................... 84
Table 3.1 Household characteristics ........................................................................................... 100
Table 3.2 Village characteristics ................................................................................................. 104
Table 3.3 Determinants of the Gini coefficient of owned farm lands (village first-differencing
and fixed effects model, village level data) ................................................................................ 106
Table 3.4 Effect of the village Gini coefficient of owned farm land on agricultural land
productivity (household first differenced model, household level data) .................................... 107
Table 3.5 Effect of village Gini coefficient of owned farm land on income and poverty
(household first differenced model, household level data) ......................................................... 110
Table 3.6 Heterogeneous effect of village Gini coefficient of owned farm land on income and
poverty (household first differenced model, household level data) ............................................ 112
viii

LIST OF FIGURES

Figure 2.1 Organic green revolution in East Africa ...................................................................... 52
Figure 3.1 Lorenz curve of national own agricultural landholding distribution over time (19972010) ........................................................................................................................................... 104
Figure 3.2 Correlation between log of income per capita and the Gini coefficient of own farm
land.............................................................................................................................................. 105

ix

INTRODUCTION

Eradication of poverty and hunger is one of the Millennium Development Goals (UN,
2014). Even though considerable policies and aids have been devoted to achieve this goal, levels
of poverty and hunger still remain high in Sub-Saharan Africa. For instance, there are 26
countries where the number of people living in extreme poverty is equal to or more than 40
percent of the population in 2011 in the world and 24 out of 26 countries are from Sub-Saharan
Africa (World Bank, 2015). Additionally, Sub-Saharan Africa has 239 million people who suffer
from hunger/undernourished in 2010, which is 30% of its total population (FAO, 2010).
Due to the fact that the majority of rural households rely on agriculture to earn a
livelihood in Sub-Saharan Africa (Quan, 2000), growth in agricultural sector is imperative for
Sub-Saharan Africa countries to reduce hunger and poverty. When we consider strategies for
agricultural development in Sub-Saharan Africa, a special attention needs to be paid to land by
researchers and policy makers because land is one of the most important productive assets for
agriculture. It has been considered for a long time that Africa is a land abundant continent and
has the high land-labor ratio (Bilsborrow, 1987; Wood; 2003; Fenske, 2010). However, recent
research shows that in fact, due to the high rate of population growth and the rapid urbanization,
the arable land frontier has been exhausted in Sub-Saharan Africa and the land-labor ratio has
shrunk in recent years (Jayne, Headey, and Chamberlin, 2014). Under this setting, access to land
for smallholders and the landless and improvements of land productivity have critical implication
for poverty reduction and food security. Therefore, this dissertation entitled “Three Essays on
Land and an Intensive Farming System in Sub-Saharan Africa: Evidence from Kenya” studies

1

the determinants and effects of land access, land distribution and an intensified farming system
in the context of Sub-Saharan Africa using household survey data form rural Kenya.
The first essay (chapter 1) is entitled “Land Access, Land Rental, and Food Security:
Evidence from Kenya.” Though quite a few studies examine the roles of land access with focus
on the land rental markets, the relationship between land access and food security is not well
assessed. Therefore, this essay attempts to assess the relationship between land access and
household income and food consumption, to compare land productivity and investment
differences between owned and rented-in parcels, and to investigate the extent which households
are able to access the optimal amount of operational land size through rental markets.
This essay is based on the combination of parcel level data and household level panel
data that covers 713 households in rural Kenya from 2004 and 2007. First, in order to quantify
the relationship between land access and income and food consumption, we rely on the fixed
effects estimator and the instrumental variable estimator to overcome the endogeneity issue of
land access. Second, within-household estimation based on parcel level data in 2007, which
could eliminate unobservable of a household, is conducted to estimate the difference in land
productivity and investment between owned and rented-in parcels. Finally, we adopt the
switching regression in order to assess the extent to which farmers could adjust their operational
farm size to the optimal level through land rental markets if the land markets were to function
perfectly.
The estimation results show several interesting findings. First, land access is significantly
and positively related to food security. An increase in operational land size of 10% would raise
household total food consumption per capita, cereal consumption per capita, and non-cereal

2

consumption per capita, and self-produced food consumption by 2.6%, 2.1%, 2.7%, and 5.4%
respectively. Second, though land rental is found to be the single most important mechanism
through which households access an additional land, it is found that land productivity and
investment are significantly lower for rented-in parcels than owned parcels. Lastly, our
estimation results indicate that land rental markets in rural Kenya do not allow farmers to adjust
their farm size up to the optimal level. Tenants (landlords) rented in (out) 67% (50%) of the size
of land they would like to rent in (out) if the land markets are working perfectly.
The second essay (chapter 2) is entitled “The Possibility of a Maize Green Revolution in
the Highland of Kenya: An Assessment of an Emerging Intensive Farming System.” As
population pressure on land grows rapidly in Kenya, rural farmers have started to intensify land
use, which has led to the emergence of a new maize farming system. The new system is
characterized by adoption of high-yielding maize varieties, application of chemical fertilizer and
manure produced by stall-fed improved dairy cows, and intercropping, especially the
combination of maize and legumes. However, very few studies assess the determinants and
productivity impacts of this new farming system in Sub-Saharan Africa. Therefore, the objective
of this study is to explore the determinants and impacts of the new maize farming system.
This essay is based on the combination of plot level data and household level panel data
that covers 622 households, who grew maize on at least 20% of their farms, in rural Kenya in
2004 and 2012. The advantage of this dataset is the parcel panel data in which we could trace the
same parcel from 2004 and 2012. This makes us possible to compare the change in input
intensification and land productivity for the same parcel over time, controlling for time-invariant
household and parcel level unobservable. Furthermore, a parcel may have several plots in a year,

3

which makes it able to compare the differences in input intensification and land productivity for
the same parcel in the same year, controlling for time variant parcel level unobservable. Taking
advantage of this parcel panel data, we estimate the parcel level fix effect models to assess the
determinants and impacts of the new maize farming system. Additionally, we adopt the
household level fix effect model in order to quantify the effect of it on household agriculture,
non-farm, and total income.
Our results show that an increase in sub-location level population density raises the rate
of hybrid maize seed adoption and the extent of agricultural intensification, meanwhile a
decrease in the land-labor ratio increases chemical fertilizer application and the degree of
agricultural intensification. These findings suggest that population pressure accelerates farming
intensification. Furthermore, it is found that the adoption of hybrid maize seed, intercropping
legumes with maize, manure application, and chemical fertilizer application have positive and
significant effects on land productivity. These effects are confirmed by the consistent and the
significantly positive impacts of the agriculture intensification index not only on land
productivity in terms of value of production and net income per hectare but also on the
household total income per capita.
The third essay (chapter 3) is entitled “The Effect of Land Distribution on Income and
Poverty Reduction: Evidence from Kenya.” Unequal distribution of agricultural lands could
negatively affect agricultural productivity, income and poverty through imperfect credit and
factor markets, political economy, and the inverse land size-productivity relationship. Though
quite a few studies have explored the relationship between asset distribution and agricultural
productivity, income growth, or poverty, most of them are based on country level data. While

4

country level analysis is informative, it is not possible to provide an answer for how asset
inequality affects household income and poverty. Therefore, this essay attempts to explore the
relationship between land inequality and household income and poverty based on household
survey data.
We will estimate the impact of land distribution on agricultural productivity, income and
poverty of rural households in Kenya based on household level panel data from 5 rounds of
repeated household survey from 1997 to 2010. The long-panel data allows us to estimate the
first-differencing model that controls for household level time-invariant unobservable. We
further explore whether effects of land inequality differ across households of different socioeconomic backgrounds by adding interaction terms between the inequality variable and
household wealth and land endowment. Additionally, the determinants of land distribution are
assessed by panel regression models based on village panel data.
The regression analyses yield several interesting results. First, the community level
population density is negatively related to the community level land distribution. As the village
population density increases by 100 persons per square kilometer, the Gini coefficient of own
farmland would decrease by 0.001. Second, unequal land distribution is found to have negative
impact on agricultural land productivity. An increase in the Gini coefficient of own farmland of
0.1 would reduce the value of crop production per acre and net crop income per acre by 4.3%
and 5.1% respectively. Third, land inequality has significant and negative effects on household
income and poverty. An increase in the Gini coefficient of 0.1 would decrease the net crop
income per capita, net livestock income per capita, net non-farm income per capita, net total
income per capita, and the probability of being out of poverty by 4.1%, 6.6%, 5.0%, 3.8% and

5

2.8%, respectively. Lastly, the negative effects of land inequality are larger and more significant
for the poorer households.

6

REFERENCES

7

REFERENCES

Fenske, J. (2013). Does Land Abundance Explain African Institutions? Economic Letter,
123(573), 1316-1390.
Food and Agriculture Organization (FAO). (2010). The State of Food Insecurity in the World.
Retrieved from http://www.fao.org/docrep/013/i1683e/i1683e.pdf .
Jayne, T.S., D. Headey, and J. Chamberlin (2014). Land pressures, the evolution of farming
systems, and development strategies in Africa: A synthesis. Food Policy, 48 1-17.
Quan, J., (2000). Land tenure, economic growth and poverty in sub-Saharan Africa. In C.
Toulmin, & J. Quan (Eds.), Evolving Land Rights, Policy and Tenure in Africa (pp.31-50).
London: IIED and NRI.
UN. (2014). The Millennium Development Goals Report. Retrieved from
http://www.un.org/millenniumgoals/2014%20MDG%20report/MDG%202014%20English%
20web.pdf
World Bank. (2015) Global Monitoring Report 2014/2015: Ending Poverty and Sharing
Prosperity. Retrieved from
http://www.worldbank.org/content/dam/Worldbank/gmr/gmr2014/GMR_2014_Full_Report.
pdf

8

CHAPTER 1: LAND ACCESS, LAND RENTAL AND FOOD SECURITY: EVIDENCE
FROM KENYA

1.1 Introduction
Despite the considerable efforts by national governments and the international
community to reduce food insecurity and improve nutrition over the years, food insecurity and
malnutrition still persist worldwide. For example, 1.4 billion people lived on less than $1.25 a
day, the international poverty line in 2005 (Chen and Ravallion, 2008). And according to FAO
(2010), 925 million people suffered from food insecurity in 2010.1 While the largest number of
under-nourished people is in Asia and Pacific (578 million), Sub-Saharan Africa has the highest
incidence of under-nourishment where 30% of its total population (roughly 239 million people)
suffers from chronic hunger, compared to 16% in Asia and Pacific. Furthermore, Sub-Saharan
Africa is the only region where the number of malnourished children has increased in the past 10
years (Ezzati et al., 2002). The number of underweight children is very large and malnutrition is
the major cause of child death in Sub-Saharan Africa (UN SCN, 2004; Black et al., 2003). The
situation of African women and children is particularly serious, as well as the situation among
female teenagers who receive less food than their male counterparts in the same households
(Albert, 2012).
These problems of food insecurity are likely to be exacerbated in densely populated and
poverty-stricken areas of Africa where the arable land frontier has been exhausted, and where
farm sizes are small and declining due to increased population pressures and sluggish structural

1

According to FAO, food security is defined as having physical and economic access to sufficient safe and
nutritious food for people to meet their dietary needs and food preferences for a healthy and active life (PinstrupAnderson 2009).

9

transformation processes (Jayne et al., 2014). This situation characterizes many areas of rural
Kenya (Muyanga and Jayne, 2014). In such settings, many land-constrained rural households
rely on land markets as an important means for increasing their access to land (Holden et al.,
2009; Yamano et al., 2009; Jin and Jayne, 2013). However, the potential of land rental markets
to support poor households’ ability to improve their access to food is poorly understood. This
study is motivated by the need to more accurately understand the potential of land rental markets
to improve rural households’ access to land and their food security status.
While there are many studies of land access with focus on the determinants of land rental
market participation in numerous countries, the relationship between land access and food
security has not been well explored in the literature. This study relies on the combination of
parcel level data and panel household data covering 713 rural Kenya households from 2004 and
2007 to explore the relationship between land access and food security. Specifically, our analysis
aims to achieve the following three objectives: (1) to assess and quantify the relationship
between operated land size, household income and food consumption (a proxy for food security);
(2) to use the parcel level data from households that cultivate both owned parcels and rented
parcels to compare land productivity and investment differences between these two types of
parcels; and (3) to investigate the extent to which households are able to access the optimal
amount of operational land size2 through land rental markets.
Our descriptive and econometric analyses yield several important findings. First, we find
a strong positive relationship between land access and food security. A 10% increase in operated

2

The optimal amount of operational land size is defined as the land size which would maximize value of output per
unit of land given the level of the household’s labor and other resources under the assumption of perfect land
markets.

10

land size would increase household total food consumption per capita, cereal consumption per
capita, non-cereal consumption, and home produced food consumption by 2.6%, 2.1%, 2.7% and
5.4%, respectively. Second, we find that land rental is the single most important mechanism that
land-poor households use to access additional land for cultivation. However, our analysis also
highlights considerable concerns with the performance of rental markets in Kenya. We find that
land productivity of rented parcels is significantly lower than owned parcels and farmers tend to
apply less organic fertilizer to rented land than to own land. Furthermore, land rental markets are
not able to allow farmers to achieve the optimal amount of operated land size, as suggested by
the fact that tenants rented in 67% of the amount of land they would like to rent in and landlords
rented out in 50% of the amount of land they would like to rent out.
The organization of this paper is as follows. Section 1.2 reviews the literature on the
performance of land rental markets in developing areas and in Kenya. Section 1.3 presents our
estimation strategy. The data used in this research is discussed in Section 1.4, followed by
descriptive statistics. Section 1.5 discusses estimation results. Finally, Section 1.6 summarizes
the major findings and draws policy implications.

1.2 Background
1.2.1 Land Rental and Sales Markets in the Presence of Market Failures
If all markets function perfectly and farming technology exhibits constant return to scale,
the initial endowment of land would not matter in terms of production efficiency because landlabor ratio would be equalized across all households through market equilibration (Feder, 1985;
Bardhan and Udry, 1999). Even if there is no land market, efficient outcomes could be achieved

11

as long as other factor markets function perfectly. However, there is ample evidence pointing
toward imperfection of rural factor markets in developing areas (De Janvry et al, 1991;
Binswanger and Rosenzweig, 1986). One of the reasons why factor markets do not function well
in developing areas is the presence of high monitoring costs of hired labor, which makes farmers
prefer to use family labor rather than hired labor on their farms. When labor and other factor
markets do not function well, households with surplus labor (or other excess assets relative to
land) can benefit from acquiring additional land. Land sales and land rental markets are therefore
potentially important means for enabling land-poor households to improve agricultural
production efficiency when labor market fails to function perfectly (Deininger, 2003).
There are several reasons why land rental markets may achieve these gains for poor rural
households more effectively than land sales market. First, land purchases require a much greater
up-front payment than renting land. Hence, land rental markets are more accessible for farmers,
especially poor farmers facing credits constraints (Hayami and Otsuka, 1993). Second, rental
payment sometime can be paid after harvest, which makes renting land by poor farmers possible
(Jin and Deininger, 2009). Third, rental markets are more flexible in terms of duration. Finally,
rental markets are less risky than sales markets. Distress sale is an example that farmers sold land
at a very cheap price to cope with emergency conditions and they ended up losing the land
forever (Rawal, 2001; Gaspart et al., 2007). These considerations partially explain why land
sales markets are generally much less active than rental markets in Africa (Holden et al., 2009).
For the same reasons, rental markets are widely promoted by the Government of Kenya
(Government of Kenya, 2007, paras 162 and 163) and many other developing countries
(Deininger, 2003).

12

1.2.2 Empirical Evidence on Land Rental Markets
For the past decade or so, land rental markets have been actively studied although there is
considerably more evidence to draw upon from South and East Asia than from Africa (Holden et
al., 2009). A few highly consistent and important findings have emerged from the large number
of studies covering a large number of countries. First, with few exceptions, land rental markets
have been found to be a major way -- if not the only way -- for enabling land-poor households to
access land (Jin and Deininger 2009 and Kimura et al. 2011 on China, Deininger and Jin 2008 on
Vietnam; Deininger et al. 2008 on India, Pender and Fafchamp 2001, Deininger et al. 2008, and
Gebregziabher and Holden 2011 on Ethiopia, Migot-Adholla et al. 1994 on Ghana; Holden et al.
2006 on Malawi; Yamano et al. 2009, and Jin and Jayne 2013 on Kenya; Andre and Platteau
1998 on Rwanda; Deininger and Mpuga 2009 on Uganda). Second, land rental markets are
generally found to enhance farm productivity. (Jin and Deininger, 2009; Deininger and Jin, 2008,
Deininger et al., 2007; Deininger and Mpuga, 2009; Jin and Jayne, 2013). Third, many studies
identified the presence of significant transaction costs associated with participating in land rental
markets. When uncertainty and imperfection information prevails, transaction costs such as
search costs to identify suitable tenants and agency costs to avoid moral hazard problem in land
use by tenants could lead imperfections in land rental market (Binswanger and Rosenzweing,
1986; Bardhan, 1989). In this regard, past studies found that the existence of the transaction costs
typically does not allow farmers to fully adjust their operated land size to the optimal level
(Skoufias, 1995; Deininger, Ali and Alemu, 2008; Yamano et al., 2009; Kimura et al., 2011).

13

1.2.3 Land Rental Markets in Kenya
Unlike many countries in Africa, Holden et al. (2009) point out that both land sales and
rental markets are allowed in Kenya. Rural households’ participation in rental markets appears to
be rising. Less than 10% of rural households rented land in several districts in Kenya in the late
1990s (Wangila, 1999). However, Yamano et al. (2009) find that 17.9% households rented land
in 15 districts in Kenya in 2004. Jin and Jayne (2013) show that the proportion of households
renting in land increased from 18% to 20% from 1997 to 2007 in 24 districts in Kenya. The data
used in this study (which is a panel of the 2004 data used by Yamano et al. (2009) in their
analysis) showed that 22.3% of households rented in land in 2007, suggesting that the proportion
of households renting in land increased by 5% in 3 years.
The Government of Kenya’s National Land Policy (2007) states that “the potential to
provide access to land to those who are productive but own little or no land” and also says that
government should “encourage the development of land rental markets while protecting the
rights of smallholders by providing better information about transactions to enhance their
bargaining power” (Government of Kenya, 2007, paras 162 and 163). Given the fact that the
Kenyan government takes a positive stance to promote land rental markets and that a significant
proportion of Kenya farmers are participating in land rental markets, it is important to understand
how well the current land rental markets are functioning in terms of allowing farmers to access
additional land for agriculture and the ensuing effects on household income and food security.

14

1.3 Estimation Strategy
To estimate the effect of operated land size on crop production, income, and consumption
based on household level panel data, we specify the following reduced form
𝑌𝑖𝑙𝑡 = 𝛼𝑖 + 𝛽1 𝐿𝑎𝑛𝑑𝑖𝑙𝑡 + 𝛽2 𝑋𝑖𝑙𝑡 + 𝛽3 𝐿𝑜𝑐𝑙 + 𝛽4 𝐷𝑡 + 𝛽5 𝐿𝑜𝑐𝑙 ∗ 𝐷𝑡 + 𝜀𝑖𝑙𝑡

(1)

where 𝑌𝑖𝑡 is one of eight output variables of interest (total food consumption per capita, total
cereal consumption per capita, total non-cereal consumption per capita, amounts of purchased
food per capita, amounts of home produced food per capita, value of crop production per capita,
net crop income per capita which is the difference between value of crop production and all paid
costs associated with crop production, and net total income per capita which is the sum of net
crop income, net livestock income, wage income, net income from self-owned business and
transfer income such as remittances and pensions) of household i in location l in time t. 3 𝐿𝑎𝑛𝑑𝑖𝑡
is operated land size. 𝑋𝑖𝑡 is a vector of househol control variables including household size, total
value of assets, household head’s age, a dummy for female head, and a dummy for head with
primary education, 𝐷𝑡 is a year dummy variable which seizes time trend, 𝐿𝑜𝑐𝑙 is location
dummies which capture the time-invariant regional characteristics, and 𝐿𝑜𝑐𝑙 ∗ 𝐷𝑡 is a vector of
interaction terms between a year dummy and location dummies that are expected to capture timespecific regional shocks. 𝛼𝑖𝑙 is a household fixed effect that captures household farmer
management ability, household risk preferences, unmeasured household wealth, and so on, that
are correlated with operated land size and production/food consumption. The existence of 𝛼𝑖𝑙
would cause OLS estimates to be biased and inconsistent. To purge 𝛼𝑖𝑙 , we take advantage of

3

Location is the administrative unit in Kenya. There are 86 locations in the sample data

15

household panel data and estimate equation (1) using a panel fixed effect estimation approach (or
first-differenced estimation approach).4 This is equivalent to estimating the following equation:
̅)
𝑌𝑖𝑙𝑡 − ̅̅̅
𝑌𝑖𝑙 = 𝛽1 (𝐿𝑎𝑛𝑑𝑖𝑡 − ̅̅̅̅̅̅̅̅̅
𝐿𝑎𝑛𝑑𝑖𝑙 ) + 𝛽2 (𝑋𝑖𝑙𝑡 − ̅̅̅̅
𝑋𝑖𝑙 ) + 𝛽4 (𝐷𝑡 − 𝐷
+𝛽5 (𝐿𝑜𝑐𝑙 ∗ 𝐷𝑡 − ̅̅̅̅̅̅̅̅̅̅̅
𝐿𝑜𝑐𝑙 ∗ 𝐷) + (𝜀𝑖𝑙𝑡 − ̅̅̅)
𝜀𝑖𝑙

(2)

̅𝑖 , ̅̅̅̅̅̅̅̅
where 𝑌
𝐿𝑎𝑛𝑑𝑖 , and 𝑋̅𝑖 are the mean values over the two time periods for the corresponding
variable. In equation (2), household fixed effect (αil) has been dropped. 𝛽1 is the key parameter
of interest to be estimated.
However, even after 𝛼𝑖𝑙 is purged, the existence of possible time varying unobservables
and the reverse causality issues (i.e., income level could also affect a rural household’s ability to
access to operational land size) could still lead to biased and inconsistent OLS estimates of
equation (2). To obtain consistent estimates, we also estimate equation (1) by an instrumental
variable (IV) approach based on pooled cross sectional household data. Operated land area in
year t is instrumented by inherited land area (accumulative prior to time t) because land
inheritance is likely to be correlated with operated land size but unlikely to affect the outcome
variables of interest directly except through its effect on operated land size. The reason why IV
estimation of equation (1) instead of IV estimation of equation (2) is adopted is that there are
very few households who inherited land between the two rounds of survey (only 20 of
households did so).
To examine the yield and input use intensity differences between own and rented parcels
based on the parcel level data in 2007,5 we use the following reduced form:
4

Given that the panel data covers two time periods, the fixed effect and first-differenced estimation approaches give
the same estimation results.
5

We only use 2007 data because 2004 data does not have the parcel characteristics variables.

16

𝑌𝑖𝑒𝑙𝑑𝑖𝑗 𝑜𝑟 𝐼𝑛𝑝𝑢𝑡𝑖𝑗 = 𝛾𝑖 + 𝛿1 𝑅𝑒𝑛𝑡𝑖𝑗 + 𝛿2 𝑍𝑖𝑗 + 𝜖𝑖𝑗

(3)

where 𝑌𝑖𝑒𝑙𝑑𝑖𝑗 is either value of crop production per acre of land or net crop income per acre of
land of parcel j belonging to household i, Inputij is input use intensity variable (either the amount
of organic fertilizer or that of chemical fertilizer per acre of land), 𝑅𝑒𝑛𝑡𝑖𝑗 is a dummy variable
for land ownership (equal to 1 for rented parcels, and 0 for owned parcels), 𝑍𝑖𝑗 is a vector of
parcel characteristics including steepness, irrigation condition, and distance to homestead, and 𝛾𝑖
is a household level fixed effect capturing unobserved household level factors that
simultaneously affecting farmer’s productivity/input use intensity and household’s tendency to
rent land (e.g., farming skills, access to technology, and wealth). We note that OLS estimation of
𝛿1 will be biased because 𝛾𝑖 is correlated with Rentit (or E(ci|Rentit=1)≠0). To deal with this, we
take advantage of the fact that all the households who rented in land also happen to own land, so
we can use the owner-cum-tenants subsample to perform within-household estimation to
eliminate 𝛾𝑖 , an approach that is widely adopted in parcel-level analysis (Shaban, 1987; Jacoby
and Mansuri, 2009; Deininger et al. 2013). The fixed effects model of (2) could be written as
follows:
(𝑌𝑖𝑒𝑙𝑑𝑖𝑗 − ̅̅̅̅̅̅̅̅
𝑌𝑖𝑒𝑙𝑑𝑖 ) 𝑜𝑟 (𝐼𝑛𝑝𝑢𝑡𝑖𝑗 − ̅̅̅̅̅̅̅̅
𝐼𝑛𝑝𝑢𝑡𝑖 ) = 𝛿1 (𝑅𝑒𝑛𝑡𝑖𝑗 − ̅̅̅̅̅̅̅
𝑅𝑒𝑛𝑡𝑖 ) + 𝛿2 (𝑍𝑖𝑗 − 𝑍̅𝑖 ) + (𝜖𝑖𝑗 − 𝜖̅)
𝑖
(4).
A second source of bias is that E(𝜖𝑖𝑗 |Rentij=1) ≠0); in fact, it is argued in the literature
that the rented parcels are generally of lower quality than owned ones (Jacoby and Mansuri
2009). We include all the main parcel-level variables collected in the survey (e.g., irrigation,
distance, and steepness) to control parcel heterogeneity. Without being able to fully control for
the unobserved factors, our land productivity estimate can be regarded as an upper bound (i.e., if
17

the bias could be eliminated the coefficient estimate would be smaller). A third estimation
challenge stems from the fact that a large number of households that do not apply organic or
chemical fertilizer. To account for the zero value of fertilizer or organic manure and the time
invariant heterogeneity, we adopt the semi-parametric trimmed LAD approach (Honore, 1992) to
estimate a fixed effect Tobit model for the input use regressions.
Finally, we adopt a switching regression to estimate the extent to which farmers are able
to adjust their operated land size to the optimal level through participating in land rental markets
based on cross sectional household level data in 2007. Following Skoufias (1995) and Deininger
et al. (2008), the switching regression with three rental participation regimes can be specified as
the following:
−𝛼𝑜𝑢𝑡 + 𝛽𝑜𝑢𝑡 𝐿𝑖 + 𝛾𝑜𝑢𝑡 𝑍𝑖 + 𝜀𝑖
𝑖𝑓 𝜀𝑖 < 𝛼𝑜𝑢𝑡 − 𝛽𝑜𝑢𝑡 𝐿𝑖 − 𝛾𝑜𝑢𝑡 𝑍𝑖
if 𝛼𝑜𝑢𝑡 − 𝛽𝑜𝑢𝑡 𝐿𝑖 − 𝛾𝑜𝑢𝑡 𝑍𝑖 ≤ 𝜀𝑖 ≤ 𝛼𝑖𝑛 − 𝛽𝑖𝑛 𝐿𝑖 − 𝛾𝑖𝑛 𝑍𝑖 } (5)
𝑦𝑖 = { 0
−𝛼𝑖𝑛 + 𝛽𝑖𝑛 𝐿𝑖 + 𝛾𝑖𝑛 𝑍𝑖 + 𝜀𝑖
𝑖𝑓 𝜀𝑖 > 𝛼𝑖𝑛 − 𝛽𝑖𝑛 𝐿𝑖 − 𝛾𝑖𝑛 𝑍𝑖
where 𝑦𝑖 is the amount of net land area leased-in; subscript out and in denote the rental market
participation status of household i with negative net area leased-in and positive area leased-in,
respectively. Households who rented-in lands have positive y and those who rented-out lands
have negative y. In our sample, there is one household who rented-in and rented-out lands at the
same time and this household was omitted from this estimation. Li is household’s land
endowment (the key variable of interest); 𝑍𝑖 is a vector of household characteristics including
land endowment (e.g., ownership of bullocks, labor endowment, value of total assets, household
head age, number of dependents, average number of owned steep land parcels, average number
of owned irrigated land parcels, average distance from homestead to owned land parcels); 𝛼𝑖𝑛
and 𝛼𝑜𝑢𝑡 are the constant terms and 𝛽𝑖𝑛 and 𝛽𝑜𝑢𝑡 are the coefficients on land endowment, γin and

18

γout are vectors of other coefficients to be estimated. The coefficients of land endowment (𝛽𝑖𝑛 and
𝛽𝑜𝑢𝑡 ) are the key coefficients of interest. The magnitude of these coefficients allows us to test
whether and to what extent land rental allows households to optimally adjust operated land size.
Specifically, a fully functioning rental market without transaction costs would imply 𝛽𝑖𝑛 equals
to -1 and 𝛽𝑜𝑢𝑡 equals to 1. So to test whether 𝛽𝑖𝑛 = −1 (or 𝛽𝑜𝑢𝑡 =1) is to test whether the rental
market allows tenants (landlords) to rent-in (rent-out) the amount of land they would like to rent
in absence of transaction costs. A detailed description on the derivation of the hypothesis can be
found in Skoufias (1995). Equation (4) is estimated by maximum likelihood estimation (MLE).
In the programing of MLE, first, it divides samples in two groups, one with 0 or positive values
of net land leasing-in and another with 0 or negative values of it and then starting values for the
iterations are provided. Then, single censored Tobit regressions in two sub-samples are applied.

1.4 Data Source and Descriptive Evidence
1.4.1 Data
The household- and parcel-level data used in our analysis are from a survey called
RePEAT.6 The data is jointly collected by the National Graduate Institute for Policy Studies
(GRIPS), the World Agroforestry Center, and Tegemeo Institute of Agricultural Policy and
Development. The RePEAT survey is based on the survey conducted by the Smallholder Diary
Project (SDP)7 that collected data from more than 3,300 households randomly from communities

6

RePEAT is Research on Poverty, Environment, and Agricultural technology and survey projects in Ethiopia,
Kenya, and Uganda founded by GRIPS and Foundation for Advanced Studies on International Development
(FASID).
7

The SDP is a project jointly by the Ministry of Livestock Development and Fisheries, the Kenya Agricultural
Research Institute, and the International Livestock Research Institute (ILRI).

19

in the Central, Rift Valley, Nyanza, and Western, and Eastern provinces in Kenya (Yamano,
Otsuka, and Place, 2011). In 2004, the RePEAT survey randomly selected 99 sub-locations and
10 households from each of the selected sub-locations, which results in a sample of 934
households. The second round of the RePEAT survey was conducted in 2007. Due to budget
constraints, 23 sub-locations in Eastern province were dropped in 2007. The survey targeted 773
households but interviewed 718 households in 76 sub-locations (the attrition rate is 7.1%). Since
5 households were not engaged in Agriculture, the panel sample used in this study has 713
households. The RePEAT survey includes detailed household information on agricultural
activities (cropping, raising livestock and growing trees), land (land tenure, land acquisition,
parcel characteristics), demographics, education, assets, salary, expenditure, consumption and so
on. Household and parcel level data are used for both descriptive and econometric analysis. Land
parcels used for analysis are agricultural farm land grown with grains, vegetables, fruits,
commercial, and all other crops. Net total income is computed as the sum of net crop income, net
livestock income, wage income, net income from self-owned business and transfer income such
as remittances and pensions. Each net income is computed as gross income minus costs. The
household food consumption is measured by total food availability which is composed of homeproduced food consumption (the difference between value of total food production and value of
total sale of food production), home-produced dairy consumption (difference between value of
total production and value of total sale of dairy products), and purchased food consumption (total
expenditure on all food items). The food consumption is classified into total food consumption,
cereal consumption, non-cereal consumption (meat, vegetables, fruits, dairy products, fish, etc.).
Each category is further differentiated by sources – either from home production or from market

20

purchase. All consumptions are measured in value term. To eliminate outlier issues in outcome
variables, values were winsorized at the 99th percentile.

1.4.2 Modes of Land Transfer
Table 1.1 shows the relative importance of various means by which Kenyan households
acquire additional lands in 2003 and 2006. While purchased parcels accounted for only 7 of
households’ operated parcels in 2003 and 2006, the number of rented parcels was 63 in 2003 and
50 in 2006, respectively. In addition, 8 parcels in 2003 and 7 parcels in 2006 were inherited or
gained through other channels. This indicates that land rental markets were a much more
important source of land acquisition than any other channels. The average size of purchased
parcels during the two periods was 1.28 acre, larger than the average of rented parcels (0.81
acre). Overall, land rental is the most important way by which farmers access additional land for
cultivation on a year-to-year basis.
Table 1.1 Frequency of households’ acquisition of land parcels, by mode of acquisition
Mode of acquisition
Purchased
Number of parcels
Mean area (acres) per parcel
Rented-in
Number of parcels
Mean area (acre)
Inherited or other
Number of parcels
Mean area (acre)

Average

2003

2006

7
1.28

7
0.84

7
1.73

57
0.81

63
0.79

50
0.84

7.5
1.34

8
0.95

7
1.74

1.4.3 Land Access and Household Characteristics
Table 1.2 describes household characteristics in 2007 for four groups according to their
land access status: those who rented-in land, those who rented-out land, those who were autarkic
21

and those who purchased land between 2004 and 2006. The simple tabulation reveals a number
of interesting insights with regard to land access. Households who purchased land during 2004
and 2006 were most resource-endowed households among groups. They have the largest number
of household size and working age household members. The percentage of households with
heads having completed primary education and who owned bullocks are also the highest. And
most of all, the average total value of assets of this group is the highest in all groups. And as
noted before, the number of households purchasing land is very small (only 20 out of 713
households). Hence, access to land through land purchase is only used by very few wealthy
households.
Table 1.2 Household characteristics by land rental market status in 2007
Rental status
Land owned (acres)
Household size (# of people)
Number of working age members (15-64)
Number of dependents*
Household head's age
% of female-headed HHs
% of heads completing primary education
% of HHs with bullocks
Total value of asset (KSh)
Total value of livestock (KSh)
Number of Observations

All

Rent-in

Rent-out

Autarkic

4.2
6.1
3.3
2.6
58
0.23
0.39
0.20
83,628
47,639
713

3.3
6.4
3.5
2.6
56
0.16
0.50
0.25
89,304
59,171
158

9.2
6.3
3.4
2.8
56
0.31
0.27
0.17
110,482
43,706
52

4.0
6.0
3.2
2.6
59
0.25
0.35
0.18
71,485
44,073
483

Purchased land
during 2004-06
4.2
6.9
4.3
2.5
47
0.05
0.80
0.40
262,230
52,860
20

Table 1.2 also shows that land rental markets also tend to transfer land from households
with higher land-labor ratio (on average, 2.7 acre per working age member) to those with smaller
land-labor ratio (0.94 acre per working age member). Land rental markets also tend to transfer
land from female-headed households to male-headed households and also from households
without a bullock to those with at least one bullock. The share of heads having completed

22

primary educations is higher for those renting in land than those renting out land. This may
suggest that, for individuals with at least a primary education, the marginal return to labor may
exceed that in non-farm sectors where casual labor and low-skill jobs predominate. Farming may
also be a source of income diversification for educated Kenyans.

1.4.4 Land Access, Production, Income and Food Security
Table 1.3 and 1.4 presents descriptive findings on a relationship between land access
measured as operational farm size and households’ agriculture income, and food security by
dividing households into 4 quartiles based on land size in 2007. Table 1.3 shows a clear and
consistent positive relationship between operational land size and value of crop production per
capita, net crop income per capita, and net total income per capita. Net crop income is defined as
the value of all crop production minus costs associated with crop production, and net total
income is a summation of net crop income, net livestock income, non-farm income and transfer
income such as remittance and pension. Each net income is computed as gross income minus
costs. Total value of crop production and net crop income for the top quartile of households are
more than double those of households in the bottom quartile. These differences in farm
production are consistent with differences in net total income by farm size, even though a
significant share of household income is derived from non-farm activities.

23

Table 1.3 Household incomes by operated farm size category in 2007
1st (smallest)
2nd
3rd
4th (largest) Average
Operational farm size quartilea
7,779
9,696
15,428
24,319
14,275
Value of crop production (KSh/capita)
b
7,036
8,602
13,491
20,840
12,467
Net crop income (KSh/capita)
Net total income (KSh/capita)c
31,547
40,195
42,361
66,170
44,952
a
Operational farm size includes rented land.
b
Net crop income is defined as the value of all crop production minus all paid costs associated with crop
production.
c
Net total income is computed as the sum of net crop income, net livestock income, wage income, net
income from self-owned business and transfer.

Table 1.4 Household food security status by operated land size category in 2007
Operational farm size quartilea
Total food consumption(KSh/capita)

b

1st (smallest)

2nd

3rd

4th (largest)

Average

14,081

14,916

16,013

20,978

16,484

c

Total cereal consumption (KSh/capita)
4,103
4,478
4,764
6,383
4,927
Total non-cereal consumption(KSh/capita)
11,916
12,910 13,757
17,473
14,000
Total food consumption from own
7,159
8,409
9,586
13,905
9,747
production (KSh/capita)
Value of food purchased (KSh/capita)
6,912
6,488
6,283
6,646
6,588
a
Operational farm size includes rented land.
b
Total food consumption includes cereal consumption and non-cereal consumption (meat, vegetables,
fruits, dairy products, fish, etc.).
c
Total cereal consumption includes maize, millet, rice, wheat, sorghum, oats, barely, and simsim.

The data also show a robust and positive relationship between land access and food
consumption. As Table 1.4 shows, households who belong to the highest operational land
quartile consumed the highest value of total food, cereal, non-cereal and self-produced food.
Among all the food categories, the largest source of the difference is food consumption from
own production. For example, food consumption from own production for the top quartile
household is almost double of that at the bottom quartile. On the other hand, there is very little
variation in value of purchased food across quartiles. The results suggest that the main
contribution of land access to rural households’ food security status is through their own farm
production.

24

1.4.5 Land Tenure Status, Value of Crop Production, Crop Income, and Input Use
Intensity
Table 1.5 compares net crop income and input use intensity between rented parcels and
owned parcels using parcel level data in 2007.8 The data indicate marked differences in value of
crop production and net crop income per acre cultivated area and an amount of organic fertilizer
use between the two types of parcels. Value of crop output and net crop income per acre on
rented parcels is significantly lower than owned parcels by 28% for value of crop output and by
34% for net income (Kenyan Shieling (KSh) 16,551 vs KSh 23,017 and KSh 13,779 vs. KSh
20,781). Data on organic fertilizer per acre also pointed toward remarkable correlation between
tenure type and incentives to apply organic fertilizer. We find that organic fertilizer use in the
rented parcels is less than half the level of that in own parcels (307 kg vs. 687 kg). This is not the
case for chemical fertilizer. In fact, the amount of chemical fertilizer per acre is slightly higher in
rented parcels than owned parcels (21 kg vs. 18 kg). Agronomy literatures discuss an importance
of organic fertilizer application as a strategy used by households to improve the long-run fertility
and productivity of their soils (Chikowo et al., 2004; Tittonell and Giller, 2013). It takes a few
years of continued organic fertilizer application to reap the full benefits of organic fertilizer use.
In contrast, application of chemical fertilizer will reap immediate payoff. It is therefore not
surprising that application of organic fertilizer is very different between owned and rented
parcels, whereas application of chemical fertilizer is not, because farmers are not incentivized to

8

The results are similar to the 2004 sample. We only report the 2007 data to make it consistent with the econometric
analysis. Due to the fact several of the key parcel level characteristics were not collected in the 2004 survey, we
excluded 2004 from the regression analysis.

25

make the long-term investment when they only use the parcels temporarily and cannot fully
recoup the investment.
Table 1.5 Tenure statuses, value of output per acre, and input use intensity by parcel type in 2007
Tenure status
Own parcels
Rented parcels
All parcels
23,017
16,551
Value of crop production (KSh/acre)
21,727
a
20,781
13,779
Net crop income per acre (KSh/acre)
19,395
687
307
614
Quantity of organic fertilizer (kg/acre)
b
18
21
18
Quantity of chemical fertilizer (kg/acre)
2.1
1.1
1.9
Farm size (acre)
984
221
1,241
Number of parcels
a
Net crop income is defined as the value of all crop production minus all paid costs associated with crop
production. 0 or negative crop income dummy are included.
b
Quantity of chemical fertilizer is measured in NPK equivalent.

While the simple tabulations presented in the descriptive analysis provide preliminary
insights about the relationship between operated land size and household production and food
security, and differences in input use intensity and crop income per unit land between rented and
owned parcels, we will need to rely on rigorous multivariate econometric analyses to draw
inferences about the causal relationships between land access and food security, etc.

1.5 Econometric Results
This section presents the main econometric results based on equations (2), (3), and (4).
We find the econometric results are mostly consistent with the descriptive results presented in
the previous section. For example, the econometric results confirm the positive and significant
relationship between land access, crop productivity, and food security. Additionally, value of
land productivity and an amount of organic fertilizer per acre are significantly lower for the
rented parcels than for the owned parcels after the household fixed effect and important parcel
level characteristics are controlled for. In addition, the switching regression allows us to gain
26

additional insights about the extent to which households are able to adjust their operated land
size relative to the optimal size.

1.5.1 Land Access and Food Security
Table 1.6 reports the effect of land access on food consumption that was estimated by
panel fixed effect estimation using a panel household data set from 2004 and 2007. The model is
in log-log specification, so the coefficients are elasticities. The results are highly consistent with
the descriptive findings as the coefficient on operated land size is positive and significant at the 1%
level in the case of household total food consumption, total cereal consumption, non-cereal
consumption, and self-produced food consumption. The magnitudes of the coefficients on
operated land size suggest that a 10% increase in operated land size would increase household
total food consumption per capita, cereal consumption per capita, non-cereal consumption, and
self-produced food consumption by 2.0%, 3.2%, 1.3%, and 4.1%, respectively. The negative
coefficients on household size for all the consumption categories are consistent with the
literature that food consumption is associated with considerable economies of scale which
describes that expenditure of food consumption per capita of a household with few members is
generally more than that of a household with more members (Nelson, 1988; Pradhan and
Ravallion, 2000). Total value of livestock is important for total food consumption, non-cereal
consumption and own produced consumption, but not for cereal consumption and food purchase,
suggesting that wealthier households have more diversified and nutritious dietary patterns.

27

Table 1.6 Impact of and access on per capita food consumption (household fixed effect model)a
Log of total
food
consumption
b
(KSh/capita)

Log of total
cereal
consumption
(KSh/capita)c

Log of total
non-cereal
consumption
(KSh/capita)

Log of total food
consumption from
own production
(KSh/capita)

Log of value of
food purchased
(KSh/capita)

Explanatory variables
Log of operational farm
size (acre)

(1)
(2)
(3)
(4)
(5)
0.203***
0.315***
0.129***
0.406***
0.0566
(0.0359)
(0.0530)
(0.0387)
(0.0529)
(0.0411)
Log of HH size (residents)
-0.764***
-0.790***
-0.785***
-0.846***
-0.680***
(0.0609)
(0.0933)
(0.0679)
(0.0680)
(0.0986)
Female-headed (=1)
0.0145
0.0239
-0.0434
-0.149
0.0893
(0.0783)
(0.130)
(0.0916)
(0.0983)
(0.112)
Log of head's age
-0.285*
-0.368*
-0.178
-0.340*
-0.103
(0.150)
(0.211)
(0.173)
(0.197)
(0.216)
Head completed primary
-0.0835
0.0199
-0.105
-0.173**
0.00160
education (=1)
(0.0571)
(0.0869)
(0.0667)
(0.0829)
(0.0770)
Log of value of livestock
0.0422***
-0.00654
0.0757***
0.0862***
0.0137
(KSh)
(0.0125)
(0.0185)
(0.0159)
(0.0185)
(0.0158)
Constant
11.75***
11.13***
10.91***
10.72***
10.47***
(0.626)
(0.901)
(0.720)
(0.821)
(0.918)
1,426
1,426
1,426
1,426
1,426
Observations
0.583
0.405
0.513
0.454
0.515
R-squared
713
713
713
713
713
Number of HH
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2007 and locations, and year 2012 dummy are included in all regressions.
b
Total food consumption includes cereal consumption and non-cereal consumption (meat, vegetables, fruits,
dairy products, fish, etc.).
c
Total cereal consumption includes maize, millet, rice, wheat, sorghum, oats, barely, and simsim.

Table 1.7 reports the estimated effects of land access on food consumption using the IV
method. The result of the first stage regression is shown in Table A1.1, which shows that
inherited land size is positively and significantly related to operated land size at 1% level. The
coefficients on operated land size are consistent with those of the fixed effect results, as those
remains positive and statistically significant for food consumption except for food purchase. The
magnitude of coefficients on operated land size suggests that a 10% increase in operated land
size would increase household total food consumption, cereal consumption, non-cereal

28

consumption, and home produced food consumption by 2.6%, 2.1%, 2.7% and 5.4%,
respectively, again pointing toward the fact that land access helps improve food security through
food availability.
Table 1.7 Impact of and access on per capita food consumption (household IV model)a
Log of total
food
consumption
(KSh/capita)b

Log of total
cereal
consumption
(KSh/capita)c

Explanatory variables
Log of operational farm
size (acre)

Log of total
non-cereal
consumption
(KSh/capita)

Log of total food
consumption from
own production
(KSh/capita)

Log of value of
food purchased
(KSh/capita)

(1)
(2)
(3)
(4)
(5)
0.261***
0.210***
0.267***
0.544***
0.00569
(0.0435)
(0.0594)
(0.0537)
(0.0631)
(0.0480)
Log of HH size
-0.771***
-0.715***
-0.777***
-0.839***
-0.644***
(0.0248)
(0.0352)
(0.0299)
(0.0371)
(0.0324)
Female-headed (=1)
-0.0685**
-0.0410
-0.0583
-0.0641
-0.0786*
(0.0315)
(0.0434)
(0.0362)
(0.0459)
(0.0404)
Log of head's age
-0.0814
-0.124*
-0.0155
0.0960
-0.221***
(0.0547)
(0.0724)
(0.0665)
(0.0815)
(0.0676)
Head completed primary
0.111***
0.127***
0.118***
0.0719*
0.163***
education (=1)
(0.0270)
(0.0373)
(0.0315)
(0.0387)
(0.0332)
Log of value of livestock
0.0666***
0.0298***
0.115***
0.156***
0.00304
(KSh)
(0.00841)
(0.00979)
(0.0104)
(0.0121)
(0.00912)
Constant
10.54***
9.686***
9.641***
8.013***
10.94***
(0.271)
(0.342)
(0.318)
(0.376)
(0.306)
Observations
1,426
1,426
1,426
1,426
1,426
R-squared
0.660
0.463
0.637
0.655
0.520
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2007 and locations, and year 2012 and no inheritance dummies are included
in all regressions.
b
Total food consumption includes cereal consumption and non-cereal consumption (meat, vegetables, fruits,
dairy products, fish, etc.).
c
Total cereal consumption includes maize, millet, rice, wheat, sorghum, oats, barely, and simsim.

29

Table 1.8 Impact of land access on production and agriculture income (household fixed effect
and IV model)a
Log of value of crop
production (KSh/capita)
FE

Log of net crop income
(KSh/capita)b

IV

FE

IV

Log of net total income
(KSh/capita)c
FE

IV

Explanatory variables
Log of operational farm
size (acre)

(1)
(2)
(3)
(4)
(5)
(6)
0.708***
0.829***
0.682***
0.800***
0.334***
0.442***
(0.0639)
(0.0751)
(0.0734)
(0.0882)
(0.0567)
(0.0841)
Log of HH size
-0.870*** -0.915***
-0.889***
-0.877***
-0.796***
-0.787***
(0.0721)
(0.0426)
(0.0921)
(0.0512)
(0.0982)
(0.0481)
Female-headed (=1)
-0.207*
-0.102*
-0.230
-0.0907
-0.249*
-0.200***
(0.121)
(0.0536)
(0.164)
(0.0628)
(0.135)
(0.0591)
Log of head's age
-0.202
-0.0230
-0.495**
-0.0242
-0.193
-0.424***
(0.209)
(0.0928)
(0.247)
(0.110)
(0.221)
(0.103)
Head completed
-0.0651
0.0975**
-0.0105
0.0600
-0.0821
0.208***
primary education (=1)
(0.0855)
(0.0445)
(0.106)
(0.0536)
(0.109)
(0.0508)
Log of value of
0.0179
0.0513***
0.0127
0.0448***
0.0397*
0.0887***
livestock (KSh)
(0.0183)
(0.0139)
(0.0227)
(0.0162)
(0.0205)
(0.0151)
Constant
10.34***
9.054***
11.24***
8.777***
11.48***
11.39***
(0.873)
(0.430)
(1.021)
(0.512)
(0.902)
(0.462)
Observations
1,425
1,425
1,426
1,426
1,426
1,426
R-squared
0.464
0.651
0.595
0.658
0.364
0.498
Number of HH
713
713
713
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2007 and locations and year 2012 dummy are included in all regressions and
no inheritance dummy is included in all IV regressions.
b
Net crop income is defined as the value of all crop production minus all paid costs associated with crop
production.
c
Net total income is computed as the sum of net crop income, net livestock income, wage income, net income
from self-owned business and transfer income.

1.5.2 Land Access, Production, and Income
Table 1.8 shows the impacts of land access on crop production and household income
using household level panel data in 2004 and 2007. As in the case of food consumption, the
econometrics results on the impact of operated land size on crop production and income are
generally consistent with the descriptive evidence. Like in Table 1.7, we report results based on

30

both the fixed effect estimation (columns 1, 3 and 5) and on IV estimation (columns 2, 4, 6).
Results from fixed effect model estimation indicate that doubling the operated land size would
increase value of crop production per capita by 71%, net crop income per capita by 68 % and net
total income per capita by 33%. The IV results also suggest significant and large effect of land
access on crop production, net crop income and net total income; doubling the operated land size
would lead to 83%, 80% and 44% increase respectively. Results on other variables are also
interesting and mostly as expected. For example, female-headed households have significantly
lower crop production and net total income. Total value of livestock has consistently positive and
significant impact on production and net income in all IV estimations and on net total income in
the fixed effect model. This is not surprising as high value of livestock may help household buy
more inputs and also provide more organic fertilizer for crop production.

1.5.3 Land Rental and Value of Crop Output, Crop Income per Acre, and Input Use
Intensity
Table 1.9 reports the estimation results on the impact of land rental on value of crop
production and net crop income using parcel level data in 2007. The base models include only
land area and dummy for a rented parcel (columns 1 and 3). The base model is expanded by
including irrigation dummy, steepness dummy, and distance from home to the parcel (columns 2
and 4). The base model indicates that land productivity is 52% (in the case of the value of crop
production per acre) or 50% (in the case of net crop income per acre) less for the rented parcels
than for the owned parcels. Adding the parcel characteristics (irrigation, distance to home and
steepness) only slightly reduced the coefficients (from 52% to 51%, and from 51% to 49%,
respectively, for the case of value of crop production per acre, and for the case of net crop
31

income). Two key explanations for lower land productivity of rented parcels compared to owned
parcels. One explanation is related to tenure security. Because rental is informal and more
temporary, tenants do not have incentive to invest on the rented land as compared to own land.
The other explanation is that parcel quality including soil fertility may be lower for rented parcel
than for owned parcels. Lower land productivity on rented parcels in our case is likely to be
outcome of the two combined effects. Given the magnitude of differences in value of crop output
and net crop income between own and rented parcels and the fact that the differences are only
trivially affected by the addition of the observed parcel characteristics, it is suggested that the
large productivity differences between the two types of parcels are unlikely to be solely caused
by the unobserved quality difference of parcels.
Table 1.10 reports the fixed effect Tobit results on input use intensity (i.e., the amount of
organic and chemical fertilizer use per acre). Consistent with the descriptive evidence, the
coefficient on rented parcels is negative and statistically significant for organic fertilizer but
insignificant for chemical fertilizer. Lower level of organic fertilizer on the rented parcels is
consistent with the argument that farmers have less incentive to make investment including
organic fertilizer on parcels that are less secure. If the unobserved land quality is taken into
account in the investment decision, then our results on rented parcels are the lower bound
estimate of disincentive effect of tenure insecurity. The negative and significant coefficient on a
dummy variable for rented parcels in the organic fertilizer regression (columns 1 and 2, Table
1.10) further increases our confidence that tenure insecurity is likely to have contributed to lower
land productivity of rented parcels as compared to owned parcels.

32

Table 1.9 Impact of land tenure on value of output per acre (household fixed effect model, parcel
level data in 2007)
Log of value of crop
Log of net crop income
production (KSh/acre)
(KSh/acre)a
(1)
(2)
(3)
(4)
Explanatory variables
Log of operational farm size (acre)
-0.0788
-0.103
-0.175***
-0.196***
(0.0653)
(0.0670)
(0.0483)
(0.0482)
Rented in parcel (=1)
-0.518***
-0.511***
-0.495***
-0.485***
(0.0901)
(0.0921)
(0.0846)
(0.0842)
Steep parcel (=1)
0.00304
-0.0118
(0.142)
(0.132)
Irrigated parcel (=1)
0.897
0.770**
(0.552)
(0.385)
Distance to parcel (km)
-0.0258**
-0.0314***
(0.0110)
(0.0106)
Log of paid input costs (KSh/acre)
0.0901***
0.101***
(0.0222)
(0.0217)
Constant
9.055***
9.004***
9.578***
9.595***
(0.133)
(0.134)
(0.0227)
(0.0340)
Observations
1,241
1,226
1,241
1,226
R-squared
0.103
0.139
0.625
0.636
Number of HH
713
712
713
712
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Net crop income is defined as the value of all crop production minus all paid costs associated with crop
production.

33

Table 1.10 Impact of land tenure on fertilizer use (household fixed effect Tobit model, parcel
level in 2007)a
Organic fertilizer (kg/acre)

Explanatory variables
Log of operational farm size (acre)
Rented in parcel (=1)

(1)
61.93
(107.3)
-2,294***
(614.1)

Steep parcel (=1)
Irrigated parcel (=1)
Distance to parcel (km)

(2)
-71.85
(179.4)
-1,990***
(598.8)
195.7
(568.4)
-1,580**
(752.1)
-383.0
(233.4)
1,226

1,241
Observations
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
All regression is estimated by the semi-parametric trimmed LAD approach.
b
Quantity of chemical fertilizer is measured in NPK equivalent.

34

Chemical fertilizer
(kg/acre)b
(3)
(4)
-0.447
-1.022
(0.997)
(1.129)
-0.249
-0.297
(2.413)
(2.640)
4.282
(3.288)
7.313
(17.24)
-0.249
(0.350)
1,241
1,226

Table 1.11 Determinants of net land leased-in (maximum likelihood estimates, 2007 data)
(1)
(2)
Explanatory variables
Leasing-in equation
Area owned (acre)
-0.723***
-0.673***
(0.115)
(0.138)
Female-headed (=1)
-0.275
-0.320
(1.352)
(1.372)
Own bulls (=1)
6.744**
7.674*
(3.181)
(3.971)
Log of value of assets (KSh)
-0.0961
0.102
(0.581)
(0.625)
Number of working members (15-64 years of age)
0.317
0.483**
(0.247)
(0.247)
Number of dependents (<15 & >64 years of age)
-0.246
-0.125
(0.253)
(0.246)
Head completed primary education (=1)
6.530**
6.957**
(2.860)
(2.891)
Head's age
0.0963**
0.0939**
(0.0459)
(0.0442)
Average number of own steeped land parcels
1.763
0.775
(1.396)
(1.368)
Average number of own irrigated land parcels
-0.505
-0.0593
(3.088)
(3.298)
Average distance to own land parcels (km)
3.957***
2.902**
(1.429)
(1.171)
Constant
9.291
-0.278
(12.77)
(14.26)
Leasing-out equation
Area owned (acre)
0.444***
0.495***
(0.0771)
(0.107)
Female-headed (=1)
1.091
1.234
(0.786)
(0.769)
Own bulls (=1)
-2.393***
-3.266***
(0.890)
(0.996)
Log of value of assets (KSh)
-0.633**
-0.336
(0.308)
(0.327)
Number of working members (15-64 years of age)
-0.314**
-0.315**
(0.142)
(0.144)
Number of dependents (<15 & >64 years of age)
-0.175
-0.227
(0.141)
(0.148)

35

Table 1.11 (cont’d)
Explanatory variables
Leasing-out equation
Head completed primary education (=1)

(1)

(2)

-1.478**
-1.881**
(0.727)
(0.768)
Head's age
0.0294
0.0236
(0.0228)
(0.0245)
Average number of own steeped land parcels
0.410
0.549
(0.621)
(0.600)
Average number of own irrigated land parcels
-2.240
-3.693**
(1.664)
(1.679)
Average distance to own land parcels (km)
-0.00719
-0.00958
(0.132)
(0.115)
Constant
47.84***
29.07
(12.20)
(33.78)
No
Yes
District dummies included
7.016***
6.803***
Σ
(1.25)
(1.18)
-1003.55
-978.94
Log likelihood
712
712
Observations
The numbers in parentheses are standard errors adjusted for clustering effect at the village level.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
1.5.4 Determinants of Net Land Leasing In and Out
Table 1.11 reports the results for the switching regression of land rental (equation 5)
using data from 2007 to show the extent of land rental market imperfection. The base model
(column 1) only includes all the relevant household characteristics and the augmented model
(column 2) includes both the household characteristics and the district dummies.
First, we look at the lease-in side. If the rental land market functions perfectly, the
coefficients of land endowment will be -1 (i.e., βin = -1 in equation (5)) in columns (1) and (2).
However, the coefficients of area owned in renting-in equation are -0.73 for the base model and
-0.67 for the augmented model and both coefficients are significantly different from -1 at the 1%
significance level. This indicates that land rental markets do not perform perfectly. Tenants who
36

rented in land only rented in 73% to 67% of the amount of land they would like to rent in in the
case of no transaction costs of renting land.9
Next, we turn to the lease-out side. If the land rental market functions perfectly, the
coefficient of land endowment would be 1 (i.e., βout = 1 in equation (5)) in columns (2) and (3).
However, the coefficients of area owned in lease-out equation are 0.44 for the base model and
0.48 for the augmented model and both coefficients are statistically different from 1 at 1% level
of significance. In other words, households who rented out land were only able to rent out 44%
to 48% of the amount of land they would want to rent out in the case of no transaction costs of
renting land.10 To put our estimates into a context, farmers in India were found to be able to
rented-in (or out) 78% (or 68%) of the amount of land they would desire to rent-in (or out) if
there is free of transaction costs of renting while the situation is much more dire in Ethiopia as
the corresponding figures are 30% and 21% (Deininger et al., 2008). Thus, the land rental market
in Kenya functions at a level that is comparable to that in India but much higher than that in
Ethiopia.
The coefficients on other variables provide further insights on the performance of land
rental markets in Kenya. First of all, land rental does allow land-poor households to rent in land
from land-rich households as indicated by the negative coefficients of land endowment in the
rent-in and positive coefficients of land endowment in the rent-out equations. Similarly, rental
markets tend to transfer land from households with less labor to households with more labor, as
9

The average land area of households who rented-in lands is 1.2 hectare. If tenants could adjust rented-in land size
fully, which means if βin = -1, the computed average rented-in land area would become 1.9 hectare for column (2)
case.
10

The average land area of households who rented-in lands is 8.4 hectare. If landholders could adjust rented-out land
size fully, which means if βout = 1, the calculated average rented-in land area would become 18.1 hectare for column
(2) case.

37

the coefficients of number of household members whose ages are between 14 years old and 65
years old is significant and negative in the rent-out equation and positive and significant (though
only in the expanded model) in the rent-in equation. Consistent with the literature, having a
bullock or not is very significant in household’s renting decisions, as land rental tends to transfer
land from households without a bullock to households with at least one bullock. Additionally,
households with heads having completed primary educations are more likely rented-in and
rented-out, which is consistent with finding in the descriptive result. This may show that the
marginal return to labor for individuals who completed primary education may be higher for
farm sector than non-farm sector where low-skilled jobs prevail. Average distance to owned land
parcels is positive and significant in the leasing-in model, suggesting households with land that is
far away from their homes tend to rent-in land. The negative coefficients for the average number
of own irrigated parcels (though significant only in the specification with district dummies) in
the leased-out models tend to suggest that households with good quality land are less likely to
rent out land.
For robustness check, I also estimated simple probit models to identify determinants of
land rental market participation. Results from probit models are reported in Table A1.2. Except
for the working age members in the leased-out model, results on other key variables such as area
owned, own bullock dummy, and whether a head completed primary education dummy or not
are largely consistent with those from the Tobit switching regression models.

38

1.6 Conclusions and Policy Implication
While African governments have devoted enormous efforts to promoting food security,
the prevalence of malnutrition and food insecurity is still quite high. Raising farm production and
productivity is a top food security strategy for rural households who remain largely dependent on
agriculture for their livelihoods. Considerable evidence shows a strong correlation between
operated farm size and food production in rural Africa. However, there is no rigorous empirical
evidence to shed light on the linkage between land access and food security. It is quite possible
that households with relatively small landholding sizes have diversified to a greater extent into
non-farm activities and are able to fully offset their lower own farm production with food
purchased through their non-farm incomes. We attempt to fill in these knowledge gaps by
exploring the relationship between land access and food security using data from rural Kenya in
this study.
Our analyses yield three salient findings. First of all, we establish a strong linkage
between land access and food security. In general, households with small farms are not able to
procure sufficient food through non-farm jobs to achieve comparable levels of food consumption
per capita as their relatively land-abundant neighbors. Second, we find that land rental markets
are the most important means available to land-constrained rural households to access additional
land for cultivation. Third, regression results show that rental markets perform below their
potential. The value of crop production and net crop income are significantly lower on rented
parcels than on owned parcels even after parcel characteristics and household fixed effects are
controlled for. Consistently, farmers also apply less organic fertilizer on rented parcels than on
owned parcels. In addition, farmers are not able to use land rental markets to attain their optimal

39

operated land size. Tenants (landlords) were only able to rent in (out) from 67 to 73% (from 44
to 48%) of the optimal amount of land they would like to rent in absence of transaction costs of
renting land.
Therefore, while land rental markets currently play a positive role in promoting
household food security in rural Kenya, there appears to be untapped potential for them to play a
more important role than they currently do. Policy efforts to improve the functioning of land
rental markets may be an under-recognized yet potentially important component of food security
and nutrition strategies in rural Kenya, and most likely in other parts of the region. More detailed
research on the organization and behavior of land rental markets is needed to identify the
specific causes of the apparently considerable underperformance of Kenya’s rural land rental
markets.

40

APPENDIX

41

Table A1.1 The first stage regression of Table 1.7 and 8a
Explanatory variables
Log of operational farm size (acre)
Log of HH size
Female-headed (=1)
Log of head's age
Head completed primary education (=1)
Log of value of livestock (KSh)
Constant

Log of operational farm size (acre)
0.378***
(0.0222)
0.101***
(0.0334)
-0.145***
(0.0419)
0.0784
(0.0737)
0.113***
(0.0365)
0.0811***
(0.00886)
-0.0431
(0.372)
1,426
0.480

Observations
R-squared
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2007 and locations, and year 2012 and no inheritance dummies
are included in the regression.

42

Table A1.2 Determinants of land leased-in and leased-out (probit model-marginal effect at mean,
2007 data)
Leasing-in
(1)
Area owned (acre)

Leasing-out
(2)

-0.0130**
-0.0160***
(0.00565)
(0.00568)
Female-headed (=1)
-0.0558
-0.0629*
(0.0369)
(0.0358)
Own bulls (=1)
0.103**
0.156***
(0.0509)
(0.0578)
Log of value of assets (KSh)
0.0196
0.00651
(0.0147)
(0.0158)
Number of working members (15-64)
0.0139**
0.0148**
(0.00603)
(0.00611)
Number of dependents (<15 & >64)
0.00555
0.00855
(0.00699)
(0.00737)
Head completed primary education (=1)
0.0855**
0.0985**
(0.0395)
(0.0401)
Head's age
-0.00145
-0.00129
(0.00124)
(0.00128)
Average number of own steeped land parcels
-0.0250
-0.0311
(0.0288)
(0.0292)
Average number of own irrigated land parcels
0.125
0.199**
(0.0977)
(0.0986)
Average distance to own land parcels (km)
-0.00356
-0.00284
(0.00624)
(0.00584)
No
Yes
District dummies included
38.83
60.83
Chi-squared
2
0.06
0.09
Pseudo R
711
711
Observations
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.

43

(3)

(4)

0.00343***
(0.00119)
0.0119
(0.0149)
-0.0299**
(0.0142)
0.00292
(0.00602)
-0.00241
(0.00246)
0.00193
(0.00233)
-0.0418***
(0.0131)
-0.00112**
(0.000503)
-0.0169
(0.0128)
0.00741
(0.0365)
-0.0378***
(0.00979)
No
26.89
0.11
711

0.00275**
(0.00128)
0.0123
(0.0143)
-0.0279**
(0.0120)
-0.000108
(0.00603)
-0.00337
(0.00225)
0.000581
(0.00215)
-0.0383***
(0.0122)
-0.00104**
(0.000485)
-0.00744
(0.0122)
-0.00152
(0.0325)
-0.0268***
(0.00834)
Yes
65.85
0.16
711

REFERENCES

44

REFERENCES

Albert, S. (2012). Food Security for Africa: an urgent global challenge, Agriculture & Food
Security, 1(1), 1-2.
André, C., & Platteau, J. (1998). Land Relations Under Unbearable Stress: Rwanda Caught in the
Malthusian Trap. Journal of Economic Behavior & Organization, 34 (1), 1–47.
Baland, J. M., Gaspart, F., Platteau, J. P., & Place, F. (2007). The distributive impact of land
markets in Uganda, Economic Development and Cultural Change, 55(2), 283-311.
Bardhan, P., & Udry, C. (1999). Development Microeconomics. New York: Oxford University
Press.
Binswanger, H. & Rosenzweig, M. (1986). Behavioral and Material Determinants of Production
Relations in Agriculture. Journal of Development Studies, 22(3), 503-539.
Black, R. E., Morris, S. S. & Bryce, J. (2003). Where and Why Are 10 Million Children Dying
Every Year?, The Lancet, 361 (9376), 2226–2234.
Boon, E. K., (2002). Food Security in Africa: Challenges and Prospects. Regional Sustainable
Development Review: Africa, 1.
Chen, S., & Ravallion, M. (2008). The Developing World Is Poorer Than We Thought, But No
Less Successful in The Fight Against Poverty. (World Bank Policy Research Working Paper
Series 4703).Washington, DC: The World Bank
Chikowo, R., Mapfumo, P. Nyamugafata,P & Giller, K. E. (2004). Maize productivity and
mineral N dynamics following different soil fertility management practices on a depleted
sandy soil in Zimbabwe. Agriclture, Ecosystems and the Environment, 102(2), 119-131.
De Janvry, A., Fafchamps, M. & Sadoulet, E. (1991). Peasant household behavior with missing
markets: some paradoxes explained. The Economic Journal, 101, 1400–1417.
Deininger, K. (2003). Land policies for growth and poverty reduction. (World Bank Policy
Research Report 26384). Washington, DC: The World Bank
Deininger, K., Ali, D. A. & Alemu, T. (2011). Impacts of Land Certification on Tenure Security,
Investment, and Land Markets: Evidence from Ethiopia. Land Economics, 87(2), 312-334
Deininger, K., & Jin, S. (2008). Land Sales and Rental Markets in Transition: Evidence from
Rural Vietnam. Oxford Bulletin of Economics and Statistics, 70(1), 67–101.
45

Deininger, K., Jin, S,. & Nagarajan ,H.(2008). Efficiency and equity impacts of rural land rental
restriction: Evidence from India. European Economic Review, 52(2), 892-918
Deininger, K., Jin, S., & V. Yadav. (2013). Does Sharecropping Affect Long-term Investment?
Evidence from West Bengal’s Tenancy Reforsm. American Journal of Agricultural
Economics, 95(3), 772-790.
Deininger, K., & Mpuga, D. (2009). Land market in Uganda: What is their impact and who
benefit?. In S. T. Holden, K. Otsuka & F. M. Place (Eds.), The Emergence of Land Markets
in Africa (pp.131-155). Washington, DC; RFF Press
European Union. (2012). The EU’s Long-term Action to Tackle Global Hunger and Undernutrition. Retrieved from http://ec.europa.eu/europeaid/what/food-security/documents/foodsecurity_fact_sheet_2012_en.pdf.
Ezzati, M., Lopez, A. D., Rodgers, A., Vander Hoorn, S., & Murray, C. J. (2002). Selected major
risk factors and global and regional burden of disease. Lancet, 360(9343), 1347.
FAO. (2010). The State of Food Insecurity in the World. Retrieved from
http://www.fao.org/docrep/013/i1683e/i1683e.pdf.
Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of Agricultural Innovations in
Developing Countries: A Survey. Economic Development and Cultural Change, 33(2), 255–
298.
Gebregziabher, G, & Holden, S. T. (2011). Distress Rentals and the Land Rental Market as a
Safety Net: Contract Choice Evidence from Tigray, Ethiopia. Agricultural Economics, 42,
45–60.
Government of Kenya. (2007). National Land Policy. Nairobi: National Land Policy Secretariat.
Hayami, Y., & Otsuka, K. (1993). The Economics of Contract Choice: An Agrarian Perspective.
New York: Oxford University Press.
Holden, S., Kaarhus, R., & Lunduka, R. (2006). Land Policy Reform: The Role of Land Markets
and Women’s Land Rights in Malawi (Noragric Report 36). Aas, Norway: Norweigian
University of Life Sciences
Holden, S. T, Otsuka, K. & Place, F. M. (2009). The Emergence of Land Markets in
Africa:Impacts on Poverty, Equity, and Efficiency. Washington, DC: RFF Press.
Honore, B. E. (1992). Trimmed Lad and Least-Squares Estimation of Truncated and Censored
Regression-Models with Fixed Effects. Econometrica, 60(3): 533-65.
46

Jacoby, H. G., & Mansuri, G .(2009). Incentives, Supervision, and Sharecropper Productivity.
Journal of Development Economics, 88(2), 232–241.
Jayne, T.S., Chamberlin, J., & Headey, D. (2014). Land pressures, the evolution of farming
systems, and development strategies in Africa: A synthesis. Food Policy, 48, 1-17.
Jin, S, & Deininger, K. (2009). Land Rental Markets in the Process of Rural Structural
Transformation: Productivity and Equity Impacts from China. Journal of Comparative
Economics, 37(4), 629–646.
Jin, S, & Jayne, T. S. (2013) Land Rental Markets in Kenya: Implications for Efficiency, Equity,
Household Income, and Poverty. Land Economics. 89 (2), 246-271.
Kimura, S., Otsuka, K., Sonobe, T. & Rozelle, S. (2011). Efficiency of Land Allocation through
Tenancy Markets: Evidence from China, Economic Development and Cultural Change,
59(3), 485 - 510.
Migot-Adholla, S. E., Benneh, G., Place, F., Atsu, S. & Bruce, J. W. (1994). Land, Security of
Tenure, and Productivity in Ghana. Searching for Land Tenure Security in Africa. 97–118.
MSU Food Security Group (FSG). (2009). Lessons Learned from 25 Years of Food Security
Research, Capacity-Building, and Outreach. (MSU International Development Working
Paper 101). East Lansing: Michigan State University
Muyanga, M. & Jayne, T. S. (2014). Effects of Rising Rural Population Density on Smallholder
Agriculture in Kenya. Food Policy, 48, 98-113.
Nelson, J. A. (1988). Household economies of scale in consumption: theory and evidence.
Econometrica, 56(6):1301-1314.
Owusu, V., Abdulai, A., & Abdul-Rahman, S. (2011). Non-farm Work and Food Security Among
Farm Households in Northern Ghana. Food Policy, 36(2), 108–118.
Pender, J., & Fafchamps, M. (2001). Land Lease Markets and Agricultural Efficiency: Theory
and Evidence from Ethiopia. (EPTD Discussion Paper 81). Washington, DC: Environment
and Production Technology Division, International Food Policy Research Institute.
Pinstrup-Andersen, P. (2009). Food Security: Definition and Measurement. Food Security, 1(1):
5–7.
Pradhan, M., & Ravallion, M. (2000). Measuring poverty using qualitative perceptions of
consumption adequacy. Review of Economics and Statistics, 82(3):462-471.

47

Rawal, V. (2001). Agrarian Reform and Land Markets: A Study of Land Transactions in Two
Villages of West Bengal, 1977–1995, Economic Development and Cultural Change,
49(3):611-62.
Shaban, R. A. (1987). Testing Between Competing Models of Sharecropping. Journal of Political
Economy, 95(5), 893–920.
Skoufias, E. (1995). Household Resources, Transaction Costs, and Adjustment Through Land
Tenancy. Land Economics, 71(1), 42–56.
Tittonell, P. & Giller, K.E., (2013). When yield gaps are poverty traps: The paradigm of
ecological intensification in African smallholder agriculture. Field Crops Research, 143, 7690.
UNDP, (2012). Africa Human Development Report 2012: Retrieved from
http://www.afhdr.org/AfHDR/documents/HDR.pdf
UN Millennium Project (2005). Halving Hunger: It Can Be Done. Retrieved from
http://www.unmillenniumproject.org/documents/HTF-SumVers_FINAL.pdf
UN System Standing Committee on Nutrition (SCN). (2004). 5th Report on the World Nutrition
Situation. Retrieved from
http://www.unsystem.org/scn/Publications/AnnualMeeting/SCN31/SCN5Report.pdf.
USAID (2007). Mozambique Food Security Programming Framework 2008-2012. Food and
Nutrition Technical Assistance (FANTA) Project. Washington D.C: Academy for
Educational Development (AED).
USDA, (2009). Food Security Analysis: Kenya. Retrieved from
http://www.fas.usda.gov/itp/Kenya/KenyaFoodSecurity09.pdf.
Wangila, J. (1999). Socio-Economic Characteristics of Households in the Soil Fertility
Replenishment Pilot Project Area in Vihiga and Siaya Districts. Nairobi: ICRAE.
Yamano, T., Place, F. M., Nyangena, W., Wanjiku, J. & Otsuka, K. (2009). Efficiency and Equity
Impacts of Land Markets in Kenya. In S. T. Holden, K. Otsuka & F. M. Place (Eds.), The
Emergence of Land Markets in Africa (pp.93-111). Washington, DC; RFF Press
Yamano, T., Otsuka, K., & Place, F. (2011). Emerging Development of Agriculture in East Africa.
New York: Springer Netherlands

48

CHAPTER 2: THE POSSIBILITY OF A MAIZE GREEN REVOLUSION IN KENYA:
AN ASSESSMENT OF AN EMERGING INTENSIVE FARMING SYSTEM

2.1 Introduction
The improvement of agricultural productivity is imperative for poverty reduction in
developing countries in general, and in sub-Saharan Africa, in particular, considering its high
rate of population growth, increasingly limited availability of cultivatable lands, and the rise
of food prices in the international market (David and Otsuka, 1994; Otsuka, Estudillo, and
Sawada, 2008; Barret, Carter and Timmer, 2010). Asia experienced a rapid rise of agricultural
productivity, known as the “Green Revolution,” characterized by adoption of chemical
fertilizer and fertilizer-responsive high-yielding varieties in the 1970s and 1980s, along with
expansion of irrigation infrastructure (Hayami and Kikuchi, 1978; David and Otsuka, 1994;
Evenson and Gollin, 2003c; Hayami and Godo, 2005; Otsuka and Larson, 2013b). In
contrast, Africa is the only continent experiencing the stagnation of agricultural productivity.
Researchers, therefore, continue to look for ways to enhance agricultural productivity in
Africa. However, it is widely believed that underdeveloped infrastructure and markets lead to
high transaction costs for purchase of chemical fertilizer and seeds of high-yielding varieties
and to poor access to irrigation, and, hence, it is not possible for small farmers to achieve
rapid growth in agricultural productivity (Jayne et al., 2003; Kydd et al., 2004; Reardon et al.,
1999; Gregory and Bumb, 2006).
Yet, under these circumstances, some farmers have begun adopting a new farming
system of maize production in the highlands of Kenya characterized by application of organic
fertilizer, i.e., manure produced from improved dairy cattle in addition to use of hybrid seeds,
chemical fertilizer, intercropping with legumes, and crop rotation (Otsuka and Yamano,
2005). A typical farmer in this system grows Napier grass, which is a common feed crop for

49

cattle that can also repel pests, feeds it to improved cattle that are raised in stalls, collects
manure from the stalls, and applies it on maize fields, where intercropping hybrid maize with
nitrogen-fixing legumes is practiced. This farming system is similar not only to the Green
Revolution in Asia in the 1970s and 1980s whose essence is application of high-yielding
varieties and chemical fertilizer, but also to the agricultural revolution in U.K. in the 18th
century, which is based on application of manure produced from stall-fed cattle as well as
production of feeds on crop fields. It may not be unrealistic to assume that this new farming
system, which embodies the essence of the two preceding revolutions in agricultural history,
will bring about “revolutionary” changes in farm productivity in Sub-Saharan Africa.
To our knowledge, however, no study has statistically examined the determinants of
adoption and productivity impacts of this emerging farming system in Sub-Saharan Africa.
Therefore, this study aims to identify the determinants of adoption of this new farming
system and to estimate its impact on productivity of maize, the major staple crop in Kenya,
through regression analyses based on the unique parcel level panel dataset. In addition to
estimating the effects of each element of the new farming system on productivity, this study
attempts to measure the impact of the entire system by creating a single agriculture
intensification index that captures this multidimensional input intensification. Our approach
will provide insights into the effects of the new farming system on the land productivity of
maize farming, which should assist policy makers in constructing new, effective strategies for
agricultural productivity improvement in Sub-Saharan Africa.
The remainder of the chapter is structured as follows. Section 2.2 outlines the
background of this study, while Section 2.3 describes the data collection method and provides
descriptive statistics. Section 2.4 explains how the maize farming system index is
constructed, Section 2.5 describes our identification strategies, and Section 2.6 presents

50

estimation results. Finally, Section 2.7 discusses conclusions and policy implications of this
study.

2.2 Background Information
In the 18th century, the agricultural revolution was realized by introduction of the
turnip as a feed crop, the stall-feeding of cattle, and ample application of manure to crop
fields (Timmer, 1969). This new farming system was based on crop rotation, feed production,
stall-fed cattle, and application of manure, which enhanced crop yields. In contrast to cattle
grazing under a three-field system which requires large areas of land but does not require
intensive labor use, stall-feeding of cattle is labor intensive as it requires feed crops or
feeding grass. The collection of manure from stalls and its application to crop fields is also
labor intensive. In addition, stall-feeding of cattle makes it possible to fully collect manure.
Therefore, a farming system based on stall-feeding of cattle is more labor-using and yieldenhancing technology than the traditional three-field farming system based on grazing. This
method seems to fit with densely populated areas in Sub-Saharan Africa, which have been
experiencing rapid population growth, shrinkage of cultivatable lands per capita, and
declining soil fertility.
Asia has experienced rapid productivity growth mainly in rice and wheat since the
late 1960s (David and Otsuka, 1994; Hayami and Godo, 2005), which is called the Green
Revolution. This high growth in agricultural productivity was realized by application of
chemical fertilizer, adoption of high-yielding modern rice varieties, and development of
irrigation. Farmers used modern varieties and chemical fertilizer simultaneously because
provision of soil nutrients is necessary to realize high yield potential of the modern varieties.
Therefore, the important lesson from the Green Revolution in Asia is that both adoption of

51

high-yielding varieties and application of chemical fertilizer are necessary to increase crop
yields significantly (Hayami and Ruttan, 1985; David and Otsuka, 1994).
However, in a country where infrastructure is underdeveloped, it is difficult for poor
farmers in rural area to have access to chemical fertilizer due to its high transaction costs.
Moreover, unlike lowland rice farming, which is most sustainable, upland farming requires
maintenance of soil fertility by applying organic fertilizer in addition to chemical fertilizer.
Hence, many farmers in the highlands of Kenya apply organic fertilizer which is made from
feces collected from stall-fed cows as depicted in Figure 2.1. Farmers grow feed grass such as
Napier grass, which repels pests, and feed it to improved cows in the stalls. Then, farmers
collect the cows’ feces and create manure from it. Many of them plant a hybrid maize variety
and apply both manure and chemical fertilizer on the plot. Moreover, they often intercrop
maize with legumes that fix nitrogen from the atmosphere, which improves soil fertility. It is
important to emphasize that this system combines technological advantages from two
agricultural revolutions, one that occurred in England in the 18th century and another that was
achieved in Asia in the 20th century. We hypothesize that the emerging farming system has
potential to boost maize productivity significantly in Sub-Saharan Africa.

Figure 2.1 Organic green revolution in East Africa
Source: Revised figure 4. From Otsuka and Yamano, 2005

52

2.3 Descriptive Analysis
2.3.1 Data
In order to analyze the determinants of adoption of the new maize farming system and
its impact on maize and entire crop yields, including yield of leguminous crops, and milk
production, data are taken from a survey called RePEAT. This data set was jointly collected
by the National Graduate Institute for Policy Studies (GRIPS), the World Agroforestry
Center, and Tegemeo Institute of Agricultural Policy and Development in Kenya. The
RePEAT survey is originally based on a survey conducted by the Smallholder Diary Project
(SDP) that collected data from more than 3,300 households randomly selected from
communities in the Central, Rift Valley, Nyanza, and Western, and Eastern provinces in
Kenya by the International Livestock Research Institute, Nairobi. In 2004, the RePEAT
survey randomly selected 99 sub-locations, which is the smallest administrative unit, and up
to 10 households from each of the selected sub-locations, which results in a sample of 899
households. The second round of the RePEAT survey was conducted in 2012, which revisited
751 households that were interviewed in 2004. Thus, the attrition rate is 16.5%. Attrition
weights are estimated and used to control for potential attrition bias in all regressions.
Because our focus is maize farmers, we limit our sample to farmers who grow maize on at
least 20% of their farm land in our analysis. After this eligibility rule is applied, our final
panel sample is composed of 622 households in 96 sub-locations in 2004 and 2012.
The RePEAT survey includes detailed household information on agricultural
activities, land use, demographics, education, assets, nonfarm income, agricultural
expenditure, and consumption. In the survey area, farmers are endowed with land parcels
each of which is typically subdivided into multiple plots to grow multiple crops. Each parcel
has unique ID that is traceable overtime, which makes it possible to compare the maize

53

production on the same parcel across two crop seasons within the same year or over the two
survey years. However, plot numbers are given in each cropping season in a year and thus
these are not traceable even across crop seasons in a given year. In our sample, 622
households had 958 parcels in 2004 and had 880 parcels in 2012. There are main and short
cropping seasons in the survey area. The agricultural production data were collected for all
crops in all plots for both the main and the short seasons. In 2004, 991 plots were grown with
maize in the main season and 561 were grown with maize in the short seasons. In 2012, the
corresponding figures are 877 plots and 479 plots, respectively. This gives us 1,552 maize
plot level sample observations in 958 parcels in 2004 and 1,356 maize plot observations in
880 parcels in 2012. To address extreme values or outliers, we drop the outcome variables if
their values are more than the 99th percentile of each variable.
Table 2.1 Sample household and sub-location characteristics
2004
Mean

S.D.

Testing
difference
in meansa

2012
S.D.

Mean

Household characteristics
Number of households

622

622

Female-headed households (%)

22%

(41)

29%

(46)

***

Head completed primary education (%)

35%

(48)

41%

(49)

**

Age of the head (years)

55.89

(13.9)

61.01

(14.2)

***

Value of productive asset (KSh)

49,394

(184,421)

35,050

(155,685)

Value of asset (KSh)

80,829

(201,970)

65,933

(169,348)

Household size

6.6

(2.9)

7.1

(3.2)

***

Household members between 15 & 64

3.6

(2.0)

4.4

(2.4)

***

Number of dependents

2.8

(1.9)

2.6

(1.7)

**

(1.8)

**

(0.7)

***

(1,616)

***

(37)

***

Owned land size (ha)
1.7
(2.4)
1.5
Owned land size per household
0.6
(0.9)
0.4
members between 15 & 64 (ha)
Sub-locations characteristics
Number of sub-locations
96
Sub-location population density
744
(1,123)
1,101
(persons/km2)
Time to the nearest big town (min by
98
(48)
79
car)
*** and ** indicate significance at 1 and 5%, respectively.
a
Significance testing of the difference in means between 2004and 2012.

54

96

Table 2.1 shows socioeconomic characteristics of the sample households and sublocations. According to this table, the proportion of female-headed households has increased
from 22% to 29% and that of household’s head who completed primary education raised
from 35% to 41% from 2004 to 2012. The typical household head has become older by 5
years. The average values of productive assets and total non-land assets have decreased by
about 14,000 Kenyan Shieling (KSh) and 15,000 KSh respectively from 2004 to 2012.11 The
average household size and the number of household working age (15-64 years) members
increased by 0.5 and 0.8 respectively, and the number of dependents has decreased by 0.2
over time. The size of owned land was small already in 2004, i.e., 1.7 hectares, indicating that
the population pressure was severe in Kenya. Owned land size has shrunk to 1.5 hectares
over the eight-year period, which clearly leads to a decrease in the land-labor ratio over time.
The sub-location population density (persons per square kilometer (km2)) has increased from
744 to 1,101 over time. Due to the fact that the land-labor ratio has reduced and the
population pressure has risen over time, it is clear that it is necessary to increase agricultural
productivity to avoid food insecurity in rural Kenya. Transportation infrastructure has
improved over time in Kenya as evidenced by the shortened time distance from the center of
sub-location to the nearest big town by a motor vehicle by 9 minutes, which indicates that
accessibility to agricultural inputs and output markets and information could have improved
over time.

2.3.2 Maize and Milk Production in Kenya
Table 2.2 provides production data in Kenya based on maize plot data per cropping
season in 2004 and 2012. The size of a maize plot has shrunk over time, which is consistent
11

Throughout this chapter, all prices are converted to the real price setting 2009 as a base year. The consumer
price index for 2004 is 66.03 and that for 2012 is 103.53.

55

with the declining trend in the owned land size. The adoption rate of hybrid maize, however,
has increased from 49% to 72%, expenditures for chemical inputs other than chemical
fertilizer, which include herbicides, pesticides, and fungicides, have risen from 88 KSh per
hectare to 176 KSh per hectare from 2004 and 2012, and expenditure of hired labor also has
increased from 2,941 KSh per hectare to 3,973 KSh per hectare. Though the ratio of
intercropping with legumes has declined by 6%, the quantity of intercropped legume seeds
has been raised by 5 kg per hectare over time. Both the adoption rate of manure and the
quantity of manure applied per hectare have risen significantly over time. In contrast, the
adoption rate of chemical fertilizer and its applied quantity, which is converted into the total
weight (in kg per hectare) of primary nutrients in terms of nitrogen (N), phosphorus (P2 O5 ),
and potassium (K 2 O5 ) contained in fertilizers (hereafter, NPK), have stagnated from 2004 to
2012. The maize yield has increased by about 40% and value of all crop production including
maize and all other intercropped crops of a maize plot has increased by 21%. Similarly,
sample households experienced a growth in their net crop income, defined as total value of
all crop production minus all paid costs associated with crop production including costs of
chemical and organic fertilizer, other chemical inputs, seed, and hired labor, by 21% over
time.
Table 2.3 shows the amount of fertilizer application and land productivity by the type
of maize seeds on a maize plot per cropping season. The adoption of hybrid maize seeds is
associated with a higher yield and value of harvest than that of local seeds by about 63% and
68%, respectively. Consistently, the proportion of plots with chemical fertilizer application is
higher for hybrid seeds than for local seeds by 37%, and the quantity of chemical fertilizer
applied per hectare is also greater for the hybrid seed plots than for the local seed by 31 kg
per hectare. In contrast to chemical fertilizer use, the proportion of manure used is same for

56

local seed plots and hybrid plots. However, when we look at the quantity of manure applied
per hectare, it is greater for hybrid seeds than for local seeds. This indicates that rural farmers
in Kenya know the importance of applying both chemical and manure to realize the yield
potential of the hybrid seeds.
Overall, it is indicated that maize farmers in the highlands of Kenya spontaneously
began exerting efforts to intensify land use under the increasing population pressure on the
limited land resources.
Table 2.2 Crop production of the maize plots per cropping season
2004
Mean

2012
S.D.

Mean

S.D.

Testing
difference
in meansa

1,552
1,356
Number of plots
0.38
(0.42)
0.34
(0.31)
***
Maize plot size (ha)
49%
(50)
72%
(45)
***
Hybrid maize seeds (%)
78%
(42)
72%
(45)
***
Intercrop with legumes (%)
39%
(49)
48%
(50)
***
Manure applied (%)
70%
(46)
71%
(45)
Chemical fertilizer applied (%)
20
(25)
25
(25)
***
Intercropped legumes seeds (kg/ha)
970
(2,554)
1385
(2,729)
***
Quantity of manure (kg/ha)
b
46
(62)
44
(50)
Quantity of chemical fertilizer (kg/ha)
88
(376)
176
(506)
***
Cost of other chemical inputs (KSh/ha) c
2,941
(5,625)
3,973
(5,684)
***
Cost of hired labor (KSh/ha)
1,363
(1,452)
1,909
(1,446)
***
Quantity of maize yield (kg/ha)
41,733
(43,285)
50,701
(43,652)
***
Value of crop production (KSh/ha)
d
32,101
(39,441)
38,918
(39,589)
***
Net crop income (KSh/ha)
*** and * indicate significance at 1 and 10%, respectively.
a
Significance testing of the difference in means between 2004and 2012.
b
Quantity of chemical fertilizer is measured in NPK equivalent.
c
This includes herbicides, pesticides, fungicides, and other chemical input.
d
Net crop income is defined as the value of all crop production minus all paid costs associated with crop
production.

57

Table 2.3 Means of yield and fertilizer application by seed type in the maize plots per
cropping season in 2012
Type of maize seeds
Local seeds
381
1,315
34,151

Hybrid seeds
975
2,143
57,215

All
1,356
1,909
50,701

Testing
difference
a
in means

Number of maize parcels
Maize yield (kg/ha)
Value of crop production (KSh/ha)
Manure
Manure applied (%)
48%
48%
48%
Quantity Applied (kg/ha)
1,070
1,509
1,385
Chemical fertilizer
Chemical fertilizer applied (%)
45%
82%
71%
Quantity Applied (kg/ha)
22
53
44
*** and * indicate significance at 1 and 10%, respectively.
a
Significance testing of the difference in means between local seeds and hybrid seeds.

***
***

***
***
***

It is a mistake to examine only maize fields if we are interested in the impacts of new
maize-based farming system because keeping improved dairy cows is an integral part of this
farming system. Table 2.4 displays the slight decline in the number of improved cows and the
total number of cows from 2004 to 2012 in the RePEAT data, though these changes are not
statistically significant. However, the quantity of milk produced per cow by local, improved,
and both local and improved cows all increased over time. Milk production per improved
dairy cow is about four times greater than that of a local cow, which demonstrates the much
higher productivity of improved cows over local cows. The use of improved dairy cows is
reminiscent of the White Revolution realized in India a few decades ago (Kajisa and
Palanichamy, 2013).

58

Table 2.4 Milk production per household in a year
2004
Mean

2012
S.D.

Mean

662

Number of households

S.D.
662

Number of local cows

1.3

(4.8)

1.3

(4.5)

Number of improved cows

1.9

(2.9)

1.8

(2.5)

Number of total cows

3.2

(5.2)

3.1

(4.8)

HH with improved cows (%)
Quantity of milk produced per cow for HH
owning only local cows (liter/cow)
Quantity of milk produced per cow for HH
owning only improved cows (liter/cow)
Quantity of milk produced per cow for HH
owning local & improved cows (liter/cow)
Quantity of milk produced per cow for all HH
(liter/cow)

0.57

(0.5)

0.56

(0.5)

154

(222)

182

(211)

695

(619)

841

(665)

336

(307)

396

(296)

511

(570)

624

(627)

29,268

(35,912)

27,683

(35,729)

Value of milk produced (KSh/cow)

Testing
difference
in meansa

***

***

20,922
(29,498)
22,127
(30,916)
Net milk income (KSh/cow)
*** indicates significance at 1%.
a
Significance testing of the difference in means between 2004and 2012.
b
Net milk income is defines as the value of milk produced minus all the paid costs associated with milk
production.
b

2.4 The Agriculture Intensification Index
It is difficult to measure the overall effect of the farming system, which consists of
multiple changes in input uses and production practices, by simply looking at individual
elements of the new farming system separately because their effects on agriculture production
could be interactive. In fact, many changes are expected to be complementary. In such a case,
if we analyze the impacts of each change on outcome variables by estimating the production
function by using each input and technology separately as an explanatory variable, we could
miss the interacting effects of multiple changes. Although it is theoretically possible to
specify the general form of production function, such as translog, it is empirically difficult to
estimate such a function due to the limited degree of freedom and high correlation among
various elements of the new farming system.12 Therefore, it will be useful to construct a

12

Table A2.1 shows both household and plot level matrices of the pairwise correlation coefficients of input uses
that consist of the new maize farming system. All the inputs are positively correlated and the correlation

59

single index that represents the degree of adoption of the new maize farming system. This
single index should incorporate the important multiple indicators from each dimension of
agriculture intensification in the system.
This study uses principal component analysis (PCA) to construct an index of
agricultural intensification. PCA is a variable reduction procedure which decomposes
variations in the variables included in the analysis into components (Darnell, 1994). A
component is a linear combination of weighted explanatory variables, in such a way that the
component accounts for a maximal amount of variance in the explanatory variables
(Cavatassi, Davis, and Lipper, 2004). Since the first component captures the greatest
proportion of total variation, it will be used as an agricultural intensification index in our
analysis. The component is constructed based on the factor scores which are used as weights
for each explanatory variable to calculate an index which represents the degree of agricultural
intensification.
For this study, we generate two agricultural intensification indices, one at the
household level and the other at the plot level. The household level agricultural
intensification index is computed by the following formula (Filmer and Prichett, 1998):
𝐻𝐼𝑖𝑡 = ∑𝑁
𝑖=1 𝐹𝑘 [

(𝑥𝑖𝑘 −𝑋𝑘 )
𝑆𝑘

],

(1)

where 𝐻𝐼𝑖 is the household level agricultural intensification index for household 𝑖 which
follows a normal distribution with a mean of zero, 𝐹𝑘 is the factor score for the variables 𝑘 in
the PCA model, 𝑥𝑖𝑘 is the variable 𝑘 of household 𝑖, and 𝑋𝑘 and 𝑆𝑘 are the mean and standard
deviation of the variable 𝑘. The PCA model includes a dummy variable for hybrid maize seed
adoption, quantity of intercropped legume seeds with maize, quantity of manure per hectare,
quantity of chemical fertilizer converted in NPK per hectare, and number of improved cows
coefficients are mostly statistically significant.

60

per hectare, as these input variables represent household level agricultural intensification of
the new maize farming system.
Table 2.5 Factor loading for maize production intensification index in maize plots
Pooled years
Household level
Individual elements
Hybrid maize seeds (=1)
0.46
Quantity of intercropped legume seed (kg/ha)
0.09
0.41
Quantity of manure (kg/ha)
Quantity of chemical fertilizer (kg/ha)a
0.59
Number of improved cows (numbers/ha)
0.51
KMO
0.60
0.31
Proportion variation explained
Mean of agriculture intensification index generated from
0.00
pooled data
1.24
SD of agriculture intensification index
Plot level
Individual elements
Hybrid maize seeds (=1)
0.56
Quantity of intercropped legume seed (kg/ha)
0.43
0.34
Quantity of manure (kg/ha)
Quantity of chemical fertilizer (kg/ha) a
0.62
KMO
0.56
Proportion variation explained
0.35
Mean of agriculture intensification index generated from
0.00
pooled data
1.19
SD of agriculture intensification index
a
Quantity of chemical fertilizer is measured in NPK equivalence.

2004

Factor loadings
0.48
0.03
0.38
0.60
0.51
0.59
0.32
-0.126
1.32
Factor loadings
0.56
0.38
0.27
0.69
0.49
0.34

2012

0.41
0.10
0.45
0.59
0.52
0.57
0.29
0.124
1.14

0.57
0.45
0.36
0.59
0.57
0.36

-0.181

0.204

1.22

1.12

Similarly, the plot level agricultural intensification index is constructed as follows:
𝑃𝐼𝑖𝑝𝑠 = ∑𝑁
𝑖𝑝𝑠=1 𝐺𝑙 [

(𝑧𝑖𝑝𝑠𝑙 −𝑍𝑙 )
𝑇𝑙

],

(2)

where 𝑃𝐼𝑖𝑝 is the plot level agricultural intensification index of household 𝑖 on maize plot p in
the cropping season s, 𝐺𝑙 is the factor score for the variables 𝑙 in this model, 𝑧𝑖𝑝𝑠𝑙 is the
variable 𝑙 of household 𝑖 on maize plot p in cropping season s, and 𝑍𝑙 and 𝑇𝑙 are the mean and
standard deviation of the variable 𝑙. This PCA model includes the same variables as in the
household level intensification index with exception of the number of improved cows per
61

hectare, since though the number of improved cows per hectare is one of the key variables of
the new maize farming system, this variable is only observable in the household level data.
As both 𝐻𝐼𝑖 and 𝑃𝐼𝑖𝑝𝑠 becomes greater, farming is supposed to be more intensified. Since the
data used for the analysis consist of two rounds of panel data, it is necessary to create indices
which can be compared over time. Therefore, the pooled data from two rounds of panel data
are used to estimate both intensification indices.
Table 2.5 shows the factor loadings of the individual elements accounting for both
household and plot level agricultural intensification indices. The principal components
explain 31% of the variance in the 5 variables for the household level model and 35% of that
in the 4 variables for the plot level model. Factor loading, which provides direction and
weight for each variable, shows that the quantity of chemical fertilizer applied and number of
improved cows account for a large part of the agricultural intensification in the household
level model and hybrid seed adoption and the quantity of chemical fertilizer applied
contribute greatly to the agricultural intensification in the plot level model. The KaiserMeyer-Olkin (KMO) measure of sampling adequacy takes a value between 0 and 1, and
higher KMO values indicate that the correlation between pairs of the explanatory variables
could be explained by the other explanatory variable (Kaiser, 1974). The KMO of the
household level index is 0.60 and that of the plot level index is 0.56, and it is usually
considered that PCA is acceptable if a value of KMO is more than 0.5. The factor loadings
for both indices obtained from the pooled samples of the 2004 and 2012 surveys display
similar patterns, which indicates that it is acceptable to use indices created from pooled data.
The result shows that agricultural intensification indices have increased from -0.126 to 0.124
from 2004 to 2012 at the household level and from -0.181 to 0.204 at the plot level,
indicating that agricultural intensification has advanced even in the short period of 8 years.

62

Table 2.6 Mean of crop production by quartile of the agriculture intensification index in
maize plots in 2012
Quartile of agriculture intensification index
1st
2nd
3rd
4th
11%
85%
95%
96%
Hybrid maize seeds (%)
50%
66%
79%
91%
Intercrop with legumes (%)
39%
44%
46%
60%
Manure applied (%)
32%
69%
87%
96%
Chemical fertilizer applied (%)
11
17
26
45
Intercropped legumes seeds (kg/ha)
528
762
1042
3134
Quantity of manure (kg/ha)
a
9
23
49
94
Quantity of chemical fertilizer (kg/ha)
b
54
118
189
334
Cost of other chemical inputs (KSh/ha)
2,083
3,458
4,709
5,213
Cost of hired labor (KSh/ha)
1,247
1,664
2,064
2,606
Quantity of maize yield (kg/ha)
27,503
40,384
52,122
79,475
Value of crop production (KSh/ha)
23,901
32,076
38,142
58,648
Net crop income (KSh/ha) c
0.32
0.38
0.37
0.28
Maize plot size (ha)
a
Quantity of chemical fertilizer is measured in NPK equivalence.
b
This includes herbicides, pesticides, fungicides, and other chemical inputs.
c
Net crop income is defined as crop production minus all paid costs associated with crop production.

Table 2.6 provides evidence that the agricultural intensification index captures the
degree of intensification of each input quite well by looking at crop production on maize
plots per cropping season by the quartile of the plot level index in 2012. As shown in the
table, there are upward trends in all individual input uses, as the quartile of the agricultural
index goes up. Consistently, outcome variables such as maize yields, value of production
from all crops, and net crop income increase as the degree of agricultural intensification
deepens. These findings indicate that the farmers’ effort of agricultural intensification is
likely to pay off in rural Kenya. Furthermore, it is interesting to note that households that
belong to the greatest quartile of the index have the smallest operated maize plot size, which
is consistent with the negative correlation between farm size and agricultural intensification
widely observed in Sub-Saharan Africa in recent years (Larson et al., 2014).

63

2.5 Estimation Strategy
2.5.1 Determinants of the New Maize Farming System Adoption
Following the literature on agricultural intensification, this study focuses on
population pressure as the driving force that accelerates agricultural intensification. Boserup
(1965) argues that a rise in population density will change the relative prices of land and
labor, which increases the demand for new inputs such as fertilizer, irrigation water,
improved seeds, and herbicide in order to intensify land use. This leads to an increase in input
use per unit of area, which is regarded as agricultural intensification. In this way, population
pressure accelerates intensive use of labor and other non-land inputs, which facilitates a shift
of farming system from extensive, such as slash and burn farming, to intensive, such as
sedentary multi-cropping farming with higher agricultural productivity (Otsuka and Place,
2001). Similarly, Hayami and Ruttan (1985) argue that changes in relative input scarcities
would bring about changes in farmers’ behaviors and institutions to adapt to new conditions,
which is called the “induced innovation hypothesis.” In their hypothesis, it is hypothesized,
as in the Boserupian view, that population pressure decreases a wage rate relative to a land
price, which increases the demand for labor and non-land input use, thereby enhancing land
productivity. Empirical evidence shows that population pressure is associated with smaller
land size and higher agricultural intensification (Josephson, Ricker-Gillbert, and Florax,
2014; Muyanga and Jayne, 2014; Ricker-Gillbert, Jumbe, and Chamberlin, 2014). Following
the existing literature, this study employs the community level population density and the
ratio of a household’s own land to family labor as proxies for population pressure on land in
order to explore its impact on agriculture intensification.
To assess the effects of the population pressure and other variables to explain
agricultural intensification, we consider estimation of the following reduced form equation

64

based on seasonal maize plot level data:
𝐼𝑖𝑝ℎ𝑣𝑑𝑠𝑡 = 𝛼0 + 𝛼1 𝑃𝑜𝑝𝑣𝑑𝑡 + 𝛼2 𝐿𝐿ℎ𝑣𝑑𝑡 + 𝛼3 𝑃𝐿𝑖𝑝ℎ𝑣𝑑𝑠𝑡 + 𝛼4 𝑃𝑟𝑣𝑑𝑡 + 𝛼5 𝑋ℎ𝑣𝑑𝑡 +
𝛼6 𝐷𝑖𝑠𝑡𝑣𝑑𝑡 + 𝛼7 𝐷𝑖𝑣𝑑 +𝛼8 𝐷𝑡 + 𝛼9 𝐷𝑖𝑣𝑑 ∗ 𝐷𝑡 + 𝑆𝑆𝑠 + 𝛽𝑝ℎ𝑣𝑑 + 𝜀𝑖𝑝ℎ𝑣𝑑𝑠𝑡 , (3)
where 𝐼𝑖𝑝ℎ𝑣𝑑𝑠𝑡 is the agricultural intensification index or one of the four agriculture input or
practice variables of interest, i.e., the amount of manure applied per hectare, amount of
chemical fertilizer converted into the NPK applied per hectare, adoption of hybrid maize
seed, and amount of intercropping legume seed planted. All variables pertain to maize plot i
in parcel p of household h in sub-location v in division d in cropping season s in year t.13
𝑃𝑜𝑝𝑣𝑑𝑡 is a sub-location level population density (persons per km2). 𝐿𝐿ℎ𝑣𝑑𝑡 is a ratio of
household’s own land size to a number of working age (15-64) household members.
𝑃𝐿𝑖𝑝ℎ𝑣𝑑𝑠𝑡 is plot land size. 𝑃𝑟𝑣𝑑𝑡 is a vector of sub-location level output and input prices
including a maize price, a diammonium phosphate (DAP) price, which is the price of most
popular chemical fertilizer in the survey area, an average hybrid maize seed price, and a wage
rate of hired labor in agriculture. 𝑋ℎ𝑣𝑑𝑡 is a vector of household control variables including a
number of working age (15-64) household members, a dummy variable for female head,
household head’s age, a dummy variable for head with primary education, value of livestock,
and a soil carbon content of the main maize plot which represents soil fertility of household’s
farm land. Some soil samples were lost or spoiled in the laboratory and thus a dummy
variable for no soil information is created and included in the regressors in order to avoid the
loss of the observations without soil sample information. 𝐷𝑖𝑠𝑡𝑣𝑑𝑡 is a travel time from the
center of sub-location to the nearest big town by a motor vehicle. 𝐷𝑖𝑣𝑑 and 𝐷𝑡 are division
and time dummies. Division and time interaction terms are also added to control for the
impact of time specific localized shocks that could affect both agricultural intensification and
13

Both division and sub-location are types of administrative regions in Kenya. There are 43 divisions, which
divided into 96 sub-locations in our sample data.

65

population pressure. 𝑆𝑆𝑠 is a short season dummy. 𝛽𝑝ℎ𝑣𝑑 is a household-parcel fixed effect
that intends to capture time-invariant parcel characteristics such as soil type and land quality
and time-invariant household level factors such as farmer management ability, household risk
preferences, and unmeasured household wealth, which could be correlated with population
density and the land-labor ratio and input use simultaneously. The existence of 𝛽𝑝ℎ𝑣𝑑 would
cause OLS estimates to be biased and inconsistent. Because of the availability of plot level
production data for the same parcels in different seasons and different years, we can purge
𝛽𝑝ℎ𝑣𝑑 by estimating equation (3) using a household-parcel level fixed effects estimation
approach. Our main interest is the estimated parameters of 𝛼1 and 𝛼2 .

2.5.2 Impact of the New Maize Farming System on Agricultural Production
To examine the impact of the new maize farming system on agricultural productivity,
the impact of each individual element of the new farming system is estimated separately.
Additionally, in order to measure the impact of the entire farming system, the effect of the
agricultural intensification index is also estimated. The following model is used to examine
individual and overall effects:
𝑄𝑖𝑝ℎ𝑣𝑑𝑠𝑡 = 𝛿0 + 𝛿1 𝐼𝑖𝑝ℎ𝑣𝑑𝑠𝑡 + 𝛿2 𝑃𝐿𝑖𝑝ℎ𝑣𝑑𝑠𝑡 + 𝛿3 𝑋ℎ𝑣𝑑𝑡 +𝛿4 𝐷𝑖𝑠𝑡𝑣𝑑𝑡 + 𝛿5 𝐷𝑖𝑣𝑑 +𝛿6 𝐷𝑡 +
𝛿7 𝐷𝑖𝑣𝑑 ∗ 𝐷𝑡 + 𝑆𝑆𝑠 + 𝜃𝑝ℎ𝑣𝑑 + 𝜇𝑝ℎ𝑣𝑑𝑡 + 𝜀𝑖𝑝ℎ𝑣𝑑𝑠𝑡 ,

(4)

where 𝑄𝑖𝑝ℎ𝑣𝑑𝑠𝑡 is one of the three output variables of interest, which are the physical maize
yield per hectare, value of all crop production, and net crop income which is defined as value
of all crop production minus all paid costs. As in the determinants of intensification
regression models, the existence of the time-invariant unobservable factor (𝜃𝑝ℎ𝑣𝑑 ) would
cause the OLS estimates to be biased and inconsistent. To deal with this, we first estimate
equation (4) using household-parcel fixed effects model approach. However, even after the
66

time-invariant household-parcel characteristics are controlled for, there are still concerns that
the time-variant household and parcel level factors, 𝜇𝑝ℎ𝑣𝑑𝑡 could affect both intensification
and agricultural outputs simultaneously. To deal with this problem, we take advantage of a
subsample of parcels for which the production data is available for at least one plot from both
seasons or more than one plot in any one cropping season in a given year. Such subsamples
of parcels allow us to estimate 𝛿1 from within household-parcel-year variation.
Outputs from a new maize farming system accrue not only from crop production but
also from milk production. Therefore, the following models are also employed in order to
capture the effect of the maize-based farming system on total value of crop and milk
production and net income from the crop and milk production:
𝑌ℎ𝑣𝑑𝑡 = 𝜋0 + 𝜋1 𝐻𝐼ℎ𝑣𝑑𝑡 + 𝜋2 𝐿ℎ𝑣𝑑𝑡 + 𝜋3 𝑋ℎ𝑣𝑑𝑡 +𝜋4 𝐷𝑖𝑠𝑡𝑣𝑑𝑡 + 𝜋5 𝐷𝑖𝑣𝑑 +𝜋6 𝐷𝑡
+𝜋6 𝐷𝑖𝑣𝑑 ∗ 𝐷𝑡 + 𝜌ℎ𝑣𝑑 + 𝜀ℎ𝑣𝑑𝑡 ,

(5)

where 𝑌𝑙𝑘𝑗𝑡 is alternately the value of crop and milk production per hectare or net income per
hectare defined as value of crop and milk production minus all paid costs associated with
crop and milk production. 𝐻𝐼ℎ𝑣𝑑𝑡 is the household level intensification index. 𝐿ℎ𝑣𝑑𝑡 is
household’s land endowment. As indicated above, the unobservable fixed effects, 𝜌ℎ𝑣𝑑 ,
would result in inconsistent estimates. Hence, the household fixed effects model is estimated
for equations (5).
Even though the intensification appears to increase land productivity and profitability,
it is not clear whether agricultural intensification also contribute to overall household income.
Though intensification increases crop income, household income could reduce in total if
intensification requires large amount of family labor and a household reduces labor allocation
to non-farm activity. Therefore, we also conduct household fixed effects estimation to
examine the effect of maize farming intensification on household net non-farm income and

67

net total income, which is a sum of net crop income, net livestock income, wage income, net
income from self-owned business and transfer income such as remittances and pensions, by
using same specification of equation (5). The dependent variables are household net non-farm
per capita and net total income per capita.

2.6 Estimation Results
2.6.1 Determinants of Adoption of the New Maize Farming System
Table 2.7 shows the estimation results of the new maize-based farming system
adoption model. In columns (1) to (5), the specifications explaining quantity of manure per
hectare, quantity of NPK equivalent chemical fertilizer use per hectare, adoption of hybrid
maize seed dummy, quantity of intercropped legume seeds planted per hectare, and the
agriculture intensification index on a maize plot per cropping season are estimated by the
household-parcel level fixed effects model.
The econometric results confirm that population pressure is indeed the driving force
for the emergence of the new farming system. For example, sub-location level population
density has a positive and significant impact on hybrid seeds adoption and the agriculture
intensification index. Additionally, the land-labor ratio has a negative and significant effect
on chemical fertilizer use and the agriculture intensification index. These estimation results
support our hypothesis that population pressure encourages input use intensification. It is
observed that plot size has consistent negative and significant impacts on all technology
adoption except hybrid maize seed, which also indicates that a constraint on farm size
facilitates intensification.

68

Table 2.7 Estimation results of the determinants of input intensification per cropping season
(parcel fixed effects model, plot level data)a

Explanatory variables
Log of sub-location
population density (ppl/km2)
Log of owned land size per
working adult (ha)
Log of cultivated plot size
(ha)
Log of maize price (KSh/kg)
Log of DAP price (KSh/kg)
Log of hybrid maize seed
price (KSh/kg)
Log of farm wage rate
(KSh/day)
Log of HH size
Female-headed (=1)
Head's age
Head completed primary
education (=1)

Organic
fertilizer
(t/ha)

Chemical
fertilizer
(10kg/ha)b

Hybrid
maize seeds
(=1)

Intercropping
legume seeds
(kg/ha)

Intensification
index

(1)

(2)

(3)

(4)

(5)

0.470

0.340

0.152*

5.227

0.328*

(0.722)

(0.907)

(0.0782)

(4.364)

(0.194)

0.0688

-0.370**

-0.00952

-1.056

-0.0681*

(0.118)

(0.177)

(0.0167)

(0.973)

(0.0387)

-0.544***

-0.985***

0.0172

-4.513***

-0.231***

(0.104)

(0.198)

(0.0159)

(0.923)

(0.0425)

0.205

0.0141

0.0421*

-0.397

0.0570

(0.209)

(0.290)

(0.0221)

(1.491)

(0.0605)

1.087*

-2.450**

-0.0203

1.971

-0.150

(0.604)

(1.032)

(0.104)

(5.492)

(0.232)

0.0197

0.550

-0.0834

-1.316

-0.0460

(0.466)

(0.940)

(0.103)

(4.556)

(0.213)

-0.0932

-1.785

-0.0497

2.083

-0.201

(0.466)

(1.193)

(0.0853)

(5.167)

(0.216)

0.277

0.610*

0.0391

-0.144

0.137

(0.269)

(0.328)

(0.0426)

(2.025)

(0.0847)

-0.333

0.0621

0.0318

1.893

0.0349

(0.250)

(0.532)

(0.0525)

(2.340)

(0.113)

-0.00688

0.0139

-0.00182

0.135

0.00229

(0.00895)

(0.0226)

(0.00166)

(0.0827)

(0.00447)

-0.104

0.679

-0.00915

-1.420

0.0275

(0.237)

(0.550)

(0.0388)

(2.309)

(0.101)

0.138**

0.0964

0.0105

-0.147

0.0391

(0.0688)

(0.124)

(0.0106)

(0.642)

(0.0252)

-0.145

1.090

0.0119

1.401

0.104

(0.491)

(0.842)

(0.0738)

(4.539)

(0.176)

-1.846

-4.778*

-0.0979

-2.723

-0.711

(1.812)

(2.505)

(0.278)

(18.52)

(0.640)

0.147

30.34**

0.487

-19.76

1.011

(8.946)

(12.02)

(1.344)

(85.02)

(3.022)

Observations

2,879

2,884

2,908

2,883

2,831

R-squared

0.068

0.164

0.189

0.106

0.155

Log of value of productive
assets (KSh)
Log of carbon in the soil
Log of time to big town (min
by car)
Constant

Number of parcels
1,118
1,119
1,122
1,120
1,113
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2012 and divisions, and year 2012, short season, and no carbon information
dummies are included in all regressions.
b
Quantity of chemical fertilizer is measured in NPK equivalence.

69

As expected, the chemical fertilizer use is negatively and statistically significantly
affected by DAP price (price of the most popular chemical fertilizer in the survey area). It
also appears that the chemical fertilizer and organic fertilizer substitute each other, as
supported by the fact that the coefficient of DAP price in the model of quantity of manure is
positive and significant at 10% level. While the hybrid seed price has the expected negative
sign in the coefficient of adoption of hybrid seed, it is statistically insignificant, suggesting
seed price is not a substantial factor affecting farmers’ decision on whether to adopt hybrid
seed varieties or not. On the other hand, farmers’ adoption decisions are positively and
significantly influenced by maize price, which is not surprising if farmers are profit
maximizers and hybrid seed varieties have yield advantages over the conventional varieties.

2.6.2 Impact of the New Maize Farming System on Agricultural Production
Table 2.8 shows the impact of individual input use and intercropping on land
productivity alternatively measured by maize yield per hectare, value of all crop production
per hectare, and net crop income per hectare on a maize plot per cropping season. The
equation (4) is estimated in two ways. We first present the household-parcel fixed effects
results for each of three measures of land productivity (columns 1, 3, and 5) and then the
household-parcel-year fixed effects results (columns 2, 4, and 6).
The household-parcel fixed effects model shows that adoption of hybrid maize is
found to contribute to 12% and 13% increases in maize yield and value of all crop
production. Additionally, the household-parcel-year fixed effects estimation indicates that the
adoption of hybrid maize would increase net crop income by 16% respectively. Quantity of
intercropped legume seeds is shown to have an almost zero impact on maize yield. This is
unsurprising because by intercropping maize with legume, the area of maize planted is

70

smaller in an intercropped field than a pure-stand field. Thus, it is not surprising that
intercropping with legume has no effect on maize yield. However, farmers could obtain
compensation from revenue from legume harvest in addition to revenue from maize and thus
total crop revenue from an intercropped field could be more than a pure-stand field. In fact,
an increase in intercropped legume seeds by 10 kg raises value of crop production by from 3
to 4% and net crop income by about 4%. Hence, although intercropping with legumes on a
maize plot does not increase maize yield, farmers can obtain higher revenue and income from
the intercropped production of legumes. In addition, as legumes enhance soil nutrients by
fixing nitrogen from the atmosphere, intercropping with legumes could contribute to a gain in
total crop revenue in the longer run. Both household-parcel fixed effects and householdparcel-year fixed effects estimations show that the additional application of manure by one
ton per hectare is expected to increase maize yield, value of all crop production, and net
income from all crops by about 2-3%, 3%, and 3%, respectively. Similarly, additional
application of chemical fertilizer by 10 kg per hectare is expected to increase maize yield and
value of all crop production by about 2-3% and 2%, respectively. However this positive
impact disappears in net crop income, implying that chemical fertilizer application increases
yields but not net income dues to its high costs. There are consistent negative effects of farm
size on all outcome variables, which demonstrates the inverse relationship between farm size
and agricultural productivity.14

14

In order to check if there are interacted effects of adoption of hybrid maize seeds and other input use, Table
A2.2 shows the estimation results of the effects of input intensification including interaction terms between
adoption of hybrid maize seeds with intercropping with legume, manure application, and chemical fertilizer
application. Though individual effects of each input still remains mostly positive and significant, most effects of
the interaction terms are insignificant. This seems to contradict with the descriptive finding in which many
farmers use hybrid seeds and fertilizers at the same time and they achieve higher yields than those who don’t.
One possible explanation is that there are mainly two types of farmers, who use inputs all together and achieve
high yields and who don’t. Thus, interaction effects might be difficult to observe in the interaction terms.

71

Table 2.8 Estimation results of the effects of input intensification on crop production per
cropping season (parcel and parcel-year fixed effects models, plot level data)a
Log of maize yield
(kg/ha)

Log of value of crop
production (KSh/ha)

Explanatory variables
Hybrid maize seeds (=1)

(1)

Parcel
-year
(2)

0.124**

0.0792

(0.0526)

(0.0646)

(0.0582)

(0.0848)

(0.0672)

(0.0924)

Intercropping legume seeds (kg/ha)

0.000314

-0.00100

0.0039***

0.00290**

0.0041***

0.00429***

Type of fixed effects model

Parcel

(3)

Parcel
-year
(4)

0.125**

0.0806

Log of net crop income
(KSh/ha)c

Parcel

(5)

Parcel
-year
(6)

0.0835

0.156*

Parcel

(0.0009)

(0.00114)

(0.00100)

(0.00135)

(0.00112)

(0.00149)

Manure (t/ha)

0.0275***

0.0176*

0.0321***

0.0313***

0.0324***

0.0194

(0.00843)

(0.00949)

(0.00903)

(0.0116)

(0.0107)

(0.0120)

Chemical fertilizer b (10kg/ha)

0.0290***

0.0180***

0.0215***

0.0103

0.00533

-0.00974

(0.00522)

(0.00631)

(0.00614)

(0.00915)

(0.00633)

(0.00868)

-0.457***

-0.530***

-0.387***

-0.450***

-0.333***

-0.435***

(0.0406)

(0.0447)

(0.0470)

(0.0555)

(0.0434)

(0.0646)

Log of cultivated plot size (ha)
Log of household size

0.128

0.116

0.0784

(0.0964)

(0.0904)

(0.0822)

Female-headed (=1)

-0.0960

-0.0858

-0.0510

(0.114)

(0.110)

(0.103)

Age of head

0.00154

0.00130

-0.000914

(0.00416)

(0.00407)

(0.00366)

0.0907

0.0197

0.188*

Head completed primary education
(=1)
Log of value of productive assets
(KSh)
Log of carbon
Log of time to big town (min by car)
Constant

(0.101)

(0.0873)

(0.103)

0.00802

-0.0387

-0.00229

(0.0247)

(0.0250)

(0.0269)

0.0499

-0.0789

0.250

(0.172)

(0.153)

(0.199)

-0.692

-0.691

-0.256

(0.496)

(0.467)

(0.549)

8.721***

6.266***

12.89***

9.680***

10.18***

9.405***

(2.232)

(0.0777)

(2.129)

(0.0973)

(2.506)

(0.113)

Observations

2,810

2,810

2,810

2,810

2,809

2,809

R-squared

0.732

0.737

0.522

0.532

0.810

0.782

1,110
1,803
1,113
1,805
1,113
1,805
Number of fixed-effects
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Short season dummy is included in all regression. Interaction terms between year 2012 and divisions, and year 2012, and
no carbon information dummies are included in regression of (2), (4) and (6).
b
Quantity of chemical fertilizer is measured in NPK equivalence.
c
Net crop income is defined as crop production minus all paid costs associated with crop production.

72

Table 2.9 Estimation results of the effects of the intensification index on crop production per
cropping season (parcel and parcel-year fixed effects models, plot level data)a
Log of maize yield
(kg/ha)

Explanatory variables

(1)

Parcel
-year
(2)

Intensification index

0.155***

Type of fixed effects model

Log of cultivated plot size (ha)
Log of household size
Female-headed (=1)
Age of head
Head completed primary education
(=1)
Log of value of productive assets
(KSh)
Log of carbon
Log of time to big town (min by car)
Constant

Log of value of crop
production (KSh/ha)

(3)

Parcel
-year
(4)

0.0817***

0.185***

(0.0214)

(0.0261)

-0.463***
(0.0399)
0.137

Log of net crop income
(KSh/ha)b

(5)

Parcel
-year
(6)

0.126***

0.129***

0.0864**

(0.0248)

(0.0360)

(0.0263)

(0.0366)

-0.534***

-0.397***

-0.457***

-0.342***

-0.430***

(0.0441)

(0.0459)

(0.0546)

(0.0426)

(0.0638)

-

0.115

0.0748

(0.0960)

(0.0900)

(0.0826)

Parcel

Parcel

Parcel

-0.103

-0.0892

-0.0521

(0.113)

(0.109)

(0.102)

0.00138

0.00148

-0.000736

(0.00411)

(0.00405)

(0.00370)

0.102

0.0202

0.180*

(0.102)

(0.0868)

(0.102)

0.0104

-0.0384

-0.00188

(0.0245)

(0.0249)

(0.0269)

0.0574

-0.0771

0.245

(0.171)

(0.153)

(0.200)

-0.747

-0.704

-0.219

(0.504)

(0.470)

(0.543)

9.192***

6.386***

13.23***

9.864***

10.20***

9.580***

(2.268)

(0.0605)

(2.138)

(0.0755)

(2.476)

(0.0897)

Observations

2,810

2,810

2,810

2,810

2,809

2,809

R-squared

0.730

0.736

0.521

0.530

0.810

0.780

Number of fixed effects

1,110
1,803
1,113
1,805
1,113
1,805
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Short season dummy is included in all regression. Interaction terms between year 2012 and divisions, and year 2012, and
no carbon information dummies are included in regression of (2), (4) and (6).
b
Quantity of chemical fertilizer is measured in NPK equivalence.
c
Net crop income is defined as crop production minus all paid costs associated with crop production.

It may not be possible to capture the whole impact of the new maize farming system
only by estimating the impact of an individual practice on agriculture production. In order to
examine the effect of the entire new maize farming system, we re-estimated equation (4) by
replacing all the individual intensification practices by the single agricultural intensification

73

index on the right hand side of the equation. The estimation results using both the householdparcel fixed effect and household-parcel-year fixed effect panel estimation methods are
reported in Table 2.9. Estimation results are generally consistent for both fixed effects models
but estimated coefficients in parcel-year fixed effect models are smaller than those of parcel
fixed effect models, suggesting possible positive bias in parcel fixed effect models which fail
to control time variant household and parcel level unobservables. The results show significant
and positive effects of the agricultural intensification index on all outcome variables for both
models. An increase in the intensification index by one standard deviation would raise maize
yield per hectare by 18% and 10%, value of all crop production per hectare by 22% and 15%,
and net crop income per hectare by 15% and 10% in the parcel fixed effects model and in the
parcel-year fixed effect model respectively. Consistent with the results in Table 2.8, we
observe the negative impacts of the farm size on outcome variables, which confirms the
inverse farm size-productivity relationship.
Since the new maize farming system aims to increase output not only from crop
production but also from milk production, Table 2.10 illustrates the impacts of agricultural
intensification on the total value of crop and milk production per hectare (column 1) and the
sum of crop and milk net income per hectare (column 2) estimated from household level
panel data. Consistent with the findings in Table 2.9, the effects of agriculture intensification
are positive and significant on both outcome variables. A rise in the intensification index by
one standard deviation would increase the value of crop and milk production per hectare by
36% and net crop income per hectare by 34%. The estimation results also indicate that the
household head’s age is negatively related to both crop and milk production and net income,
indicating that a household with a younger head tends to have higher agricultural
productivity.

74

Table 2.10 Estimation results of the effects of the intensification index on agriculture
production per year (household fixed effects model, household level data)a
Log of value of crop &
milk production (KSh/ha)

Log of net crop & milk
income (KSh/ha)b

(1)

(2)

0.293***
(0.0302)
-0.0354
(0.0446)
0.0274
(0.0791)
-0.0284
(0.0986)
-0.00477*
(0.00282)
-0.00583
(0.0701)
0.00929
(0.0218)
-0.104
(0.158)
-0.337
(0.451)
12.65***
(2.038)
1,195
0.389
619

0.277***
(0.0382)
-0.0146
(0.0655)
-0.0430
(0.105)
-0.161
(0.107)
-0.00897***
(0.00344)
-0.0262
(0.0919)
-0.0125
(0.0289)
-0.189
(0.225)
-0.541
(0.552)
13.89***
(2.447)
1,195
0.524
619

Explanatory variables
Intensification index
Log of owned land size (ha)
Log of household size
Female-headed (=1)
Head's age
Head completed primary education (=1)
Log of value of productive assets (KSh)
Log of carbon
Log of time to big town (min by car)
Constant

Observations
R-squared
Number of households
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2012 and divisions, and year 2012, short season, and no carbon
information dummies are included in all regressions.
b
Net crop and milk income is defined as the value of crop and milk production minus all paid costs
associated with crop and production.

75

Table 2.11 Estimation results of the effects of the intensification index on non-farm and total
household income per year (household fixed effects model, household level data)a
Log of net non-farm income
per capita (KSh)

Log of net total income per
capita (KSh)b

(1)

(2)

0.0787
(0.0820)
0.231*
(0.118)
-0.295
(0.199)
-0.496*
(0.266)
-0.0128
(0.00873)
-0.251
(0.223)
0.0931*
(0.0538)
-0.181
(0.402)
1.003
(1.132)
-0.964
(0.656)
5.095
(5.120)
1,192

0.168***
(0.0386)
0.172***
(0.0447)
-0.545***
(0.0889)
-0.305***
(0.117)
-0.00703*
(0.00374)
-0.117
(0.0921)
0.0631**
(0.0250)
-0.176
(0.218)
-0.490
(0.640)
-0.730**
(0.312)
13.25***
(2.886)
1,192

Explanatory variables
Intensification index
Log of owned land size (ha)
Log of household size
Female-headed (=1)
Head's age
Head completed primary education (=1)
Log of value of productive assets (KSh)
Log of carbon
Log of time to big town (min by car)
Constant

Observations
R-squared
Number of households
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2012 and divisions, and year 2012, short season, and no carbon
information dummies are included in all regressions.
b
Net total income is computed as the sum of net crop income, net livestock income, wage income, net
income from self-owned business and transfer income.

Furthermore, Table 2.11 shows the estimation results of the effect of maize farming
intensification on the net non-farm income per capita (column 1) and the net total household
income per capita (column 2) based on household level panel data. There is a potential
concern that the positive effect of increase in agricultural intensification on crop income
could be offset by a reduction of non-farm income if intensification requires large amount of
76

family labor and a household reduces its labor allocation to non-farm activities. The positive
and insignificant coefficient on intensification index in the off-farm equation allays such
concern (column 1). Finally, intensification has significant and positive effect on total income
as supported by the positive and statistically significant coefficient of intensification index on
the total income equation (column 2). In terms of the magnitude of effect, one standard
deviation increase in the intensification index causes net total income to increase by 21%.
The results on other variables are also mostly consistent with expectation. For example, land
access is significantly and positively related to both non-farm income and total income.
Female-headed households are worse off than male-headed households, as they earn 50% and
31% less non-farm income and total income than male-headed households holding other
factors constant. The value of productive assets is positively associated with non-farm
income and total income.

2.6.3 Impact of the New Maize Farming System on Profit
One limitation of the data is that it did not collect information on family labor use on
crop production and thus we could not estimate an impact of the new farming system on
economic profits for which family labor cost is deducted. This may cause bias in the
estimated effect of the input intensification on outcome variables.15 If intensification requires
households to use more family labor which is not captured in data, an impact of
intensification on agricultural production or income would be overestimated. Though the
survey team did not collect family labor use data from all plots, they collected it from the
largest pure stand maize plot in the main cropping season. If a household does not have a

15

The concern is mainly related to the quantification of the net effects on agricultural productivity and
agricultural income, the results on off-farm income and total income are not affected as the labor use is
internalized in the measurement of off-farm income and total income.

77

pure-stand maize plot, family labor use information on the largest intercropped maize plot in
main season was collected. Based on this additional information in the dataset, we could
compute a profit, which is defined as value of crop production minus all the costs associated
with production including family labor on the largest maize plot. This enables us to check if
the impact of agricultural intensification differs between net crop income and profit at least
for the largest plot.
Table 2.12 Crop production of the largest pure-stand maize plot or the largest intercropped
maize plot in the main cropping season
2004
Mean

2012
S.D.

Mean

S.D.

Testing
difference
in meansa

426
426
Number of plots
0.48
(0.45)
0.41
(0.34)
***
Maize plot size (ha)
52%
(50)
76%
(43)
***
Hybrid maize seeds (%)
44%
(50)
60%
(49)
***
Manure applied (%)
71%
(46)
75%
(43)
Chemical fertilizer applied (%)
942
(2,567)
1525
(2,464)
***
Quantity of manure (kg/ha)
b
53
(66)
43
(42)
***
Quantity of chemical fertilizer (kg/ha)
891
(763)
686
(708)
***
Family labor (hours/ha)
237
(398)
84
(207)
***
Hired labor (hours/ha)
1,180
(876)
778
(736)
***
Total labor (hours/ha)
c
106
(411)
184
(407)
***
Cost of other chemical inputs (KSh/ha)
4,088
(6,429)
4,859
(6,029)
*
Cost of hired labor (KSh/ha)
1,661
(1,330)
2,071
(1,404)
***
Quantity of maize yield (kg/ha)
47,541 (40,774) 58,546 (44,362)
***
Value of crop production (KSh/ha)
Net crop income from all crops
36,920 (39,759) 45,246 (39,623)
***
(KSh/ha)d
34,225 (38,841) 42,992 (39,542)
***
Crop profit from all crops (KSh/ha)e
*** and * indicate significance at 1 and 10%, respectively.
a
Significance testing of the difference between columns (b) and (c)
b
Quantity of chemical fertilizer is measured in NPK equivalent.
c
This includes herbicides, pesticides, fungicides, and other chemical input.
d
Net crop income is defined as the value of all crop production minus all paid costs associated with
crop production.
e
Net crop income is defined as the value of all crop production minus all costs associated with crop
production including family labor costs.

78

Table 2.13 Estimation results of the effects of the intensification index on crop production in
the main cropping season (parcel fixed effects models, the largest maize plot level data)a

Explanatory variables
Intensification index
Log of cultivated plot size (ha)
Log of household size
Female-headed (=1)
Age of head
Squared age of head
Head completed primary education (=1)
Log of value of productive assets (KSh)
Log of carbon
Log of time to big town (min by car)
Constant

Log of net crop
income (KSh/ha)b

Log of crop profit
(KSh/ha)c

(1)

(2)

0.0789*
(0.0468)
-0.441***
(0.0922)
0.170
(0.107)
-0.0492
(0.158)
-0.00187
(0.0256)
-0.0000145
(0.000216)
0.223
(0.155)
-0.0224
(0.0386)
0.0583
(0.193)
-0.415
(0.446)
-0.0109
(0.240)
0.0328
(0.124)
11.50***

0.101*
(0.0539)
-0.433***
(0.0988)
0.107
(0.115)
-0.0726
(0.176)
-0.00613
(0.0275)
0.0000156
(0.000230)
0.347*
(0.179)
-0.00587
(0.0389)
0.00378
(0.209)
-0.461
(0.501)
-0.142
(0.259)
0.0379
(0.141)
11.75***

Observations
R-squared
Number of fixed effects
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Year 2012 and no carbon information dummies are included in all regressions.
b
Net crop income is defines as the value of crop production minus all paid costs associated with crop
production.
e
Crop profit is defined as the value of all crop production minus all costs associated with crop
production including family labor costs.

Table 2.12 displays crop production data on the largest pure-stand maize plot or the
largest intercropped maize plots in the main cropping season. The means of most inputs and
outputs have the same trend as in Table 2.2; crop yield and revenues increased with the level
of intensification over time. Additionally, crop profit has increased by 26% over time. In
79

contrast to other input use, family labor use, hired labor use, and total labor use have
decreased significantly over time. Though hired labor use measured in working hours has
decreased, costs of hired labor have increased, which indicates that the wage rate of hired
labor in agricultural sector has risen over time. This means that intensification occurs in a
rather unexpected manner. It could be that the intensification system intensifies the use of
capital inputs to save the high cost of labor through input substitution.
Table 2.13 compares the effect of the new farming system on net crop income, which
is defined as value of crop production minus all paid costs, and crop profits, which is defined
as value of crop production minus all costs including family labor cost, using the same
subsample of plots. The results indicate that one standard deviation increase in the
intensification index would raise net crop income by about 9% and crop profit by 12%,
suggesting that the potential biases of the estimated effects of the intensification system based
on the large sample without accounting for family labor are likely to be small.

2.7 Conclusion and Policy Implication
As population pressure grows rapidly in Kenya, rural farmers have started to intensify
a farming system by adopting new inputs and production practices, including adoption of
high-yielding maize varieties, application of manure produced by improved dairy cows, and
intercropping especially of maize with legumes that could fix nitrogen. Though the
phenomenon of the new farming system has started to be paid attention among researchers,
the empirical research that assesses the driving forces and impacts of this system is limited.
Hence, this study aims to quantify the determinants of the new maize farming system and its
impact on agricultural productivity. To gauge the impact of the new farming system, this
study examines the impacts of individual inputs as well as the impact of the new maize

80

farming system by using an agricultural intensification index constructed by PCA.
Our estimation results show that an increase in sub-location level population density
raise the rate of hybrid maize seed adoption and the extent of agricultural intensification,
meanwhile a decrease in the land-labor ratio increases chemical fertilizer application and the
degree of agricultural intensification. These findings indicate that population pressure
accelerates farming intensification, consistent with the Boserupian and induced innovation
hypotheses. Furthermore, it is found that the adoption of hybrid maize seed, intercropping
legumes with maize, manure application, and chemical fertilizer application have positive
and significant impacts on land productivity. These impacts are confirmed and reinforced by
the consistent and statistically significantly positive impacts of the agriculture intensification
index not only on land productivity in terms of value of production and net income per
hectare but also on the household total income per capita.
Therefore, we conclude that the new farming system has already improved the
productivity of small-scale farmers in the highlands of Kenya. Therefore, effort for exploring
the “optimum” farming system in agricultural research center is encouraged. It can be
expected that much more significant increase in the productivity of farming could be
achieved if appropriate research is carried out and appropriate technical support and
extension services regarding this new maize farming system are provided for small-scale
maize farmers in Kenya.

81

APPENDIX

82

Table A2.1 Pairwise correlation coefficients matrix of input use
(1)

(2)

Household level
Hybrid maize seeds (=1)
1
(1)
Quantity of intercropped legume
0.0044
1
(2)
seed (kg/ha)
0.0899***
0.0598**
(3)
Quantity of manure (kg/ha)
Quantity of chemical fertilizer
0.2571***
0.0041
(4)
(kg/ha)a
Number of improved cows
0.1115***
0.0286
(5)
(numbers/ha)
Plot level
Hybrid maize seeds (=1)
1
(1)
Quantity of intercropped legume
0.0716***
1
(2)
seed (kg/ha)
0.0739*** 0.0826***
(3)
Quantity of manure (kg/ha)
Quantity of chemical fertilizer
0.2695*** 0.1653***
(4)
(kg/ha)a
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Quantity of chemical fertilizer is measured in NPK equivalence.

83

(3)

(4)

(5)

1
0.1428***

1

0.163***

0.2439***

1
0.0828***

1

1

Table A2.2 Estimation results of the effects of input intensification on crop production per
cropping season (parcel and parcel-year fixed effects models, plot level data)a
Log of maize yield (kg/ha)
Type of fixed effect model
Explanatory variables
Hybrid maize seeds (=1)
Intercropping legume seeds
(kg/ha)
Manure (t/ha)
Chemical fertilizer (10kg/ha)b
Hybrid seed * intercropping
legume seeds (kg/ha)
Hybrid seed * manure (t/ha)
Hybrid seed * chemical fertilizer
(10kg/ha)
Log of cultivated plot size (ha)
Log of household size
Female-headed (=1)
Age of head
Head completed primary
education (=1)
Log of value of productive
assets (KSh)
Log of carbon
Log of time to big town (min by
car)
Constant

Parcel
(1)

0.234***
(0.0703)
0.00153
(0.00139)
0.0314**
(0.0136)
0.0407***
(0.00853)
-0.0159*
(0.00896)
-0.00581
(0.0171)
-0.00203
(0.00167)
-0.454***
(0.0407)
0.128
(0.0963)
-0.0931
(0.114)
0.00192
(0.00416)
0.0887
(0.102)
0.00876
(0.0247)
0.0490
(0.173)
-0.696
(0.498)
8.776***
(2.243)
2,810
0.733
1,110

Parcel
-year
(2)

0.157*
(0.0860)
-0.000797
(0.00171)
0.0313*
(0.0169)
0.0279***
(0.00936)
-0.0129
(0.0101)
-0.0212
(0.0199)
-0.000432
(0.00212)
-0.526***
(0.0447)

6.227***
(0.0831)
2,810
0.738
1,803

Log of value of crop
production (KSh/ha)
Parcel
(3)

0.167**
(0.0770)
0.00403**
(0.00168)
0.0288*
(0.0158)
0.0298***
(0.00888)
-0.0114
(0.0100)
0.00474
(0.0186)
-0.000261
(0.00186)
-0.386***
(0.0471)
0.115
(0.0901)
-0.0813
(0.110)
0.00154
(0.00407)
0.0174
(0.0873)
-0.0386
(0.0250)
-0.0789
(0.153)
-0.677
(0.469)
12.80***
(2.135)
2,810
0.523
1,113

Parcel
-year
(4)

0.0943
(0.106)
0.00179
(0.00225)
0.0400*
(0.0207)
0.0178
(0.0135)
-0.0106
(0.0143)
-0.0134
(0.0248)
0.00192
(0.00252)
-0.449***
(0.0558)

9.674***
(0.106)
2,810
0.532
1,805

Log of crop net income
(KSh/ha)c
Parcel
(5)

0.0751
(0.0902)
0.00393**
(0.00187)
0.0210
(0.0180)
0.00890
(0.0108)
-0.00492
(0.0119)
0.0163
(0.0214)
0.000365
(0.00229)
-0.334***
(0.0436)
0.0777
(0.0820)
-0.0472
(0.103)
-0.000806
(0.00368)
0.187*
(0.103)
-0.00272
(0.0268)
0.252
(0.198)
-0.230
(0.548)
10.06***
(2.496)
2,809
0.811
1,113

Parcel
-year
(6)

0.0939
(0.116)
0.00240
(0.00235)
0.0167
(0.0210)
-0.00841
(0.0137)
-0.00273
(0.0150)
0.00497
(0.0250)
0.00331
(0.00291)
-0.437***
(0.0653)

9.436***
(0.119)
2,809
0.783
1,805

Observations
R-squared
Number of fixed effects
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Short season dummy is included in all regression. Interaction terms between year 2012 and divisions, and year 2012, and
no carbon information dummies are included in regression of (2), (4) and (6).
b
Quantity of chemical fertilizer is measured in NPK equivalence.
c
Net crop income is defined as crop production minus all paid costs associated with crop production.

84

REFERENCES

85

REFERENCES

Barrett, C. B., Carter, M. R., & Timmer, C. P. (2010). A century-long perspective on agricultural
development. American Journal of Agricultural Economics, 92(2), 447–468.
Boserup, E. (1965). The conditions of agricultural growth. The Economics of agrarian change
under population pressure. London: Allen & Unwin.
Cavatassi, R., Davis, B., & Lipper, L. (2004). Estimating poverty over time and space:
construction of a time-variant poverty index for Costa Rica (ESA Working Paper 04-24).
Roma, Italy: FAO.
Darnell, A. C. (1994). A dictionary of econometrics. Brookfield, VT: Edward Elgar:
David, C. C., & Otsuka, K. (1994). Modern rice technology and income distribution in Asia.
Boulder, CO: Lynne Rienner Publishers.
Evenson, R. E., & Gollin, D. (2003). Crop variety improvement and its effect on productivity:
The impact of international agricultural research. Wallingford, Oxford: CAB International.
Filmer, D., & Pritchett, L. H. (2001). Estimating wealth effects without expenditure Data - Or
tears: An application to educational enrollments in states of India. Demography, 38(1),
115–132.
Gregory, D. I., & Bumb, B. (2006). Factors affecting supply of fertilizer in Sub-Saharan Africa
(Agriculture and Rural Development Discussion Paper 24). Washington, D.C.: World Bank.
Hayami, Y., & Godo, Y. (2005). Development economics: From the poverty to the wealth of
nations. Oxford: Oxford University Press.
Hayami, Y., & Ruttan, V. W. (1985). Agricultural development: An international perspective,
1880-1960. Baltimore, MD: Johns Hopkins University Press.
Jayne, T. S., Govereh, J., Wanzala, M., & Demeke, M. (2003). Fertilizer market development: a
comparative analysis of Ethiopia, Kenya, and Zambia. Food Policy, 28(4), 293–316.
Josephson, A. L., Ricker-Gilbert, J., & Florax, R. J. G. M. (2014). How does population density
influence agricultural intensification and productivity? Evidence from Ethiopia. Food
Policy., 48, 142-152
Kaiser, H. F. (1974). An index of factorial simplicity. Psychometrika, 39(1), 31–36.
86

Khandker, S. R., Bakht, Z., & Koolwal, G. B. (2009). The poverty impact of rural roads:
evidence from Bangladesh. Economic Development and Cultural Change, 57(4), 685–722.
Kikuchi, M., & Hayami, Y. (1978). Agricultural growth against a land resource constraint: a
comparative history of Japan, Taiwan, Korea, and the Philippines. The Journal of
Economic History, 38(04), 839–864.
Kydd, J., Dorward, A., Morrison, J., & Cadisch, G. (2004). Agricultural development and propoor economic growth in sub-Saharan Africa: potential and policy. Oxford Development
Studies, 32(1), 37–57.
Larson, D. F., Otsuka, K., Matsumoto, T., & Kilic, T. (2014). Should African rural development
strategies depend on smallholder farms? An exploration of the inverse-productivity
hypothesis. Agricultural Economics, 45(3), 355–367.
Muyanga, M., & Jayne, T. S. (2014). Effects of rising rural population density on smallholder
agriculture in Kenya. Food Policy, 48, 98-113
Otsuka, K., & Place, F. (2001). Land tenure and natural resource management: A comparative
study of agrarian communities in Asia and Africa. Baltimore, ML: IFPRI books.
Otsuka, K., & Yamano, T. (2005). The possibility of a green revolution in sub-Saharan Africa:
Evidence from Kenya. Journal of Agricultural and Development Economics, 2(1), 8–19.
Otsuka, K., Estudillo, J. P., & Sawada, Y. (2008). Rural poverty and income dynamics in Asia
and Africa. London: Routledge.
Reardon, T., Barrett, C., Kelly, V., & Savadogo, K. (1999). Policy reforms and sustainable
agricultural intensification in Africa. Development Policy Review, 17(4), 375–395.
Ricker-Gilbert, J., Jumbe, C., & Chamberlin, J. (2014). How does population density influence
agricultural intensification and productivity? Evidence from Malawi. Food Policy, 48, 114128
Timmer, C. P. (1969). The turnip, the new husbandry, and the English agricultural revolution.
The Quarterly Journal of Economics, 375–395.

87

CHAPTER 3: THE EFFECT OF LAND DISTRIBUTION ON INCOME AND POVERTY
REDUCTION: EVIDENCE FROM KENYA

3.1 Introduction
Land is a primal productive asset for rural farmers in Sub-Saharan Africa. Therefore,
access to land has non-negligible impact on agricultural production and income for rural
households (Gunning et al., 2000; Nargis and Hossain, 2006). However, past studies have shown
that land distribution in Sub-Saharan Africa is highly skewed and land is highly concentrated
(Griffin et al., 2002; Jayne et al., 2003; Barret et al., 2005; Holden et al., 2009). This is a
worrisome situation for rural households in Sub-Saharan Africa because unequal land
distribution is often found to be detrimental to income growth (Deininger and Squire, 1998;
Vollrath, 2007). Since most of households in rural areas in Sub-Saharan Africa depends on
agriculture for their livelihoods (Quan, 2000), land distribution and land access could have farreaching implications on rural household’s welfare in Sub-Saharan Africa.
There are several potential channels through which land inequality may affect
household’s income and poverty. First, land inequality can affect rural household’s income and
poverty through its possible effect on credit. For instance, the landless and land constrained
households face challenges to access credit to invest in productive and human capital due to lack
of valuable collateral (Stiglitz, 1969; Loury, 1981; Galor and Zeira, 1993; Binswanger et al.,
1995; Benabou, 1996; Aghion and Bolton, 1997; Aghion et al., 1999). Second, land inequality
creates unequal factor ratios across households that would persist in the presence of market
imperfections (Feder, 1985; Bardhan and Udry, 1999). The persistent imbalance in the factor
ratios is likely to be translated into persistent low productivity and inefficiency. Third, land

88

distribution could affect income and poverty through its effect on provision of public goods and
services, as well as social and political cohesion within a community. For example, a constraint
on wealth could limit households’ ability to express their concerns in politics (Bourguignon and
Verdier, 2000). High inequality could hinder local communities to provide growth enhancing
public goods and services that benefit all residents (Cardenas, 2003). Additionally, high unequal
wealth distribution could induce public ‘bads’ such as social unrest and strife between inhabitants
in a community which could undermine economic growth (Conning and Robinson, 2007).
Fourth, the inverse relationship between farm size and agricultural productivity suggests that
skewed land distribution among farmers is associated with lower land productivity in a region
(Schultz, 1964; Bardhan, 1973; Feder, 1985; Hayami and Otsuka, 1993; Binswanger et al., 1995;
Vollrath, 2007; Hazell, 2011). In this regard, policy makes are increasingly aware of the
important implication of relationship between asset distribution and economic growth (Bardhan
et al., 2000; World Bank, 2005).
Though quite a few studies have explored the relationship between asset distribution and
agricultural productivity, income growth, or poverty, most of them are based on country level
data. While informative, cross-country studies are likely to suffer from a number of data or
methodological concerns. One concern about cross-country studies is related to data
comparability. The way in which data are aggregated to the national level and the definitions of
certain variables could be very different across countries. Additionally, findings based on crosscountry analysis have limited policy use for policies targeted at the community and/or at the
household level. From a policy perspective, it is important to understand how asset inequality
affects households from different economic or social backgrounds. Cross-country studies are

89

unable to provide an answer to this type of question.
Therefore, our study aims to contribute to the debate on the relationship between land
inequality and income and poverty based on micro level panel household survey data.
Specifically, we will assess the impact of land distribution on agricultural productivity, income
and poverty of rural households in Kenya based on household level panel data from 5 rounds of
repeated household survey from 1997 to 2010. The long panel data allow us to estimate a panel
regression to control for time-invariant unobservable that could be associated with community
level land inequality and household economic outcomes. The heterogeneous effects of land
inequality for different groups of households are also explored. Finally, village level panel data
are used to identify the determinants of village level land distribution.
The estimation results reveal several interesting findings. First, population density is
negatively related to land distribution in a village. As village population density increases by 100
persons per square kilometer (km2), the Gini coefficient of own land size would reduce by 0.004
- 0.005. Second, agricultural land inequality has consistent and negative effects on agricultural
land productivity, as an increase in the Gini coefficient of own agricultural land by 0.1 would
reduce the value of crop production per acre and net crop income per acre by 4.3% and 5.1%.
Third, unequal distribution of agricultural land has negative effects on household per capita
income and household poverty status. An increase in the Gini coefficient by 0.1 would decrease
the net crop income per capita, net livestock income per capita, net non-farm income per capita,
net total income per capita, and the probability of being out of poverty 4.1%, 6.6%, 5.0%, 3.8%
and 2.8% respectively. Last but not least, the negative income and poverty effects of land
inequality are much more significant and larger in magnitude for poor and land-constrained

90

households. High level education, however, does not seem to mitigate the negative effects of
land inequality.
The rest of this essay consists of 5 sections. Section 3.2 describes how unequal land
distribution affects economic growth and poverty and land distribution in Kenya. Section 3.3
explains the estimation strategy to pursue the purpose of this paper. Section 3.4 describes data
which is used for this study and shows sample characteristics. Section 3.5 presents the estimation
result. Finally, Section 3.6 summarizes the major findings and discusses the policy implications.

3.2 Background
3.2.1 Why Does Land Inequality Affect Economic Growth and Poverty?
Land inequality could affect income growth in a community through multiple channels
including credit access, efficiency loss due to imbalance in factor ratios across households,
political and social cohesion, and loss of productivity because the existence of inverse sizeproductivity relationship.
First, unequal land distribution could affect income growth through its effect on credit. If
households wish to raise their productivity but their human and physical capital is constrained,
they need financing for investment. However, if information asymmetry between lenders and
borrowers prevails, credit rationing will arise (Stiglitz and Weiss, 1981). In this environment,
borrowers could access credit if they use their own assets such as lands as collateral. Hence, it
would be difficult for land-constrained or landless households to access credit to invest in human
and physical capital if credit market failure prevails. Additionally, limitation of access to credit
not only influences accumulation of productive capital by limiting investment, but also inhibits

91

formulation of the spill-over effects of growth enhancing technologies (Galor and Zeira, 1993).
Therefore, higher land inequality could lead to lower aggregate income growth in a region. If
credit markets function perfectly, the relationship between land inequality and income growth
breaks down, as the distribution of resources no longer determines household income growth.
The credit market imperfection in the developing world has long been documented (Binswanger
et al., 1995; Benabou, 1996; Aghion and Bolton, 1997; Aghion et al., 1999).
Second, inequality in land distribution hinders aggregate productivity through the
inefficiency effect of imbalance in factor ratios across households within a community. If all
markets function perfectly and agricultural production function technology exhibits constant
return to scale, the ratio of land and other production factors among all households in a region
would be equalized by market equilibration and thus the initial endowment of land would not
matter for production efficiency (Feder, 1985; Bardhan and Udry, 1999). Even though there is no
land market, the efficient outcome could be achieved if other factor markets function perfectly.
Therefore, if land markets or other factor markets function well, the initial distribution of land
would not matter for the economic growth. However, past studies point out the fact that
imperfection of rural factor markets prevails in developing areas (Binswanger and Rosenzweig,
1986; De Janvry, Fafchamps, and Sadoulet, 1991). Hence, imbalance in the factor ratios inherited
from the skewed land distribution persists, leading to a long-term inefficiency and low
productivity.
Third, land inequality is hypothesized to have negative effect on community’s political
and social cohesion and public good provisions. Limitation of wealth could make it difficult for
individuals to lobby and to express their concerns in politics (Bourguignon and Verdier, 2000;

92

Deininger et al., 2009). Additionally, it may be easier for more equal society to reach a consensus
of provision of public goods and services that could benefits all residents than more unequal
society (Durlauf, 1996; Cardenas, 2003; Benjamine et al., 2011). Unequal asset distribution not
only could hinder from provision of growth enhancing public goods, but also could generate
public ‘bads’ such as social strife and unrest which could pose huge social costs and undermine
economic growth (Conning and Robinson, 2007, Deininger et al., 2009). In fact, unequal land
distribution since post-colonial time became one of the main causes of violence which was
triggered by the president election in 2007 in Kenya (Kimenyi and Ndung’u, 2005).
Fourth, the inverse relationship between farm size and productivity is another possible
reason why land inequality lowers aggregate productivity. The hypothesis of inverse relationship
between land size and land productivity simply explains that small farmers are more productive
than large farms because of monitoring costs of hired labor (Schultz, 1964; Hayami and Otsuka,
1993; Binswanger et al., 1995) or for some other reasons (Benjamin, 1995). While large farms
usually require hired labor for agricultural production and landholders need to pay costs for
monitoring, small farms, which could be cultivated only by family labor, generally do not require
monitoring costs. Under this setting, land productivity of small farms would be higher than that
of large farms (Bardhan, 1973; Feder, 1985; Vollrath, 2007). Highly unequal land distribution
could mean that few landholders have a large share of land and many others have small pieces of
land. If small landholders could not access more land up to the amount at which they maximize
land productivity given their endowment, and large landholders are not productive as much as
small landholders due to the high monitoring costs, then aggregate agriculture productivity could
be enhanced simply by redistributing some land from large farmers to small landholders and the

93

landless.
The limited number of previous studies on land inequality and economic growth,
productivity and poverty are mostly based on aggregate data. For example, Deininger and Squire
(1998) showed the negative relationship between initial land distribution and economic growth
and Vollrath (2007) confirmed the negative effects of land distribution on agricultural
productivity, based on county level data. Additionally, Ravallion and Datt (2002) found that the
high initial proportion of landless households in a state inhibits economic growth of non-farm
sector and poverty reduction based on state level panel data in India. We are not aware of any
study that explores the relationship between land inequality and household’s income and poverty
status using household panel data. We aim to fill in this knowledge gap here.

3.2.2 Land Distribution in Kenya
Before colonization, land was allocated to households in a community and managed
collectively by a community in Kenya. Though there were no official land certificates, the user
rights were recognized and protected by a community (Muyanga, 2013). However, during the
colonial time, the Crown Land legislation was enacted, which established “scheduled” land for
white settlers and “reserves” for natives (Okoth-Ogendo, 1991; Syagga, 2010). The scheduled
areas were fertile land in highland but the reserves were generally marginal and unfertile land.
The colonial government took 2.8 million hectare of fertile land, which was about a half of high
agriculturally potential land, from communities and allocated to the white settlers (OkothOgendo, 1991; Mamdani, 1996; Kimenyi and Ndung’u, 2005, Kanyinga, 2009). As a result, the
reserves became congested as population grew, which led to overuse and overgrazing of land

94

(Kanyinga, 2009; Muyanga, 2013).
Few years before independence, the British government and the World Bank launched the
Million-Acre Settlement Scheme, which aimed at transferring white settler’s farms to Kenyan.
The land-transferring program was based on market-based “willing-seller/willing buyer” system.
This resulted in the fact that most of land was purchased by the wealthy ethnic majority group,
such as Kikuyu (Leo, 1989). Additionally, distribution of state owned land was used as a political
tool to favor the patrons of rulers by providing public land to the economic and political elites
(Muyanga, 2013). This resulted in skewed agricultural land distribution in Kenya, where a small
number of wealthy elites had large underutilized farmland, while the large number of famers
experienced reduction of per capita farm sizes. This fostered long-lasting animosities among the
ethnic groups, which became one of the causes of social unrest that happened at the time of the
presidential election in 2007 (Kimenyi and Ndung’u, 2005).
Currently the primary mode of land acquisition for smallholders is via inheritance and
land market transactions in Kenya. The Kenyan government supports market transactions of land
by the National Land Policy (NLP), which is the framework for land access and administration.
NLP provides and ensures land rights, makes regulation of the markets, and invests in
infrastructure (Republic of Kenya, 2009).

3.3 Estimation Strategy
Before quantifying the effects of land distribution on productivity, income and poverty,
we first explore the determinants of land distribution. Bardhan et al. (2014) showed how
population growth can affect land inequality. Population growth in a community could be

95

explained in two ways, by exceeding the birth rate over the death rate, and by exceeding the
number of immigrants over out-migrants. When a household divides land for inheritance to more
than one child, a household would have smaller land than before. Land inequality tends to fall if
a household with large landholding divides their land at a faster rate than a household with small
landholding and vice versa. Immigrants usually access land by land sale or land rental markets. If
a large piece of land is divided and sold to new residents, the land inequality within a community
could fall. However, if a large number of poor landless immigrants settled in a community, land
inequality of that community could rise. Bardhan et al. (2014) empirically showed that land
reform reduced land inequality by household divisions and market transactions, but the effect
was offset by an increase in the number of landless immigrants in a state in India. This study
employs the population density to measure the effects of population pressure to land on land
distribution.
The model to identify determinants of land distribution based on village level data can be
expressed as following:
𝐺𝑖𝑛𝑖𝑣𝑡 = 𝛼0 + 𝛼1 𝑃𝑜𝑝𝑣𝑡 +𝛼2 𝑍𝑣𝑡 + 𝛼3 𝐷𝑡 + 𝛽𝑣 + 𝜀𝑣𝑡

(1)

where 𝐺𝑖𝑛𝑖𝑣𝑡 is the Gini coefficient of owned agricultural land in village v in year t measuring
the degree of land inequality in a community, 𝑃𝑜𝑝𝑣𝑡 is population density (persons per km2), 𝑍𝑣𝑡
is a vector of village characteristics including the average size of land owned by an average
household, average household size, average distance from a residence to the tarmac road,
average distance from a residence to the extension service in a village, the proportion of
households participating in the rental market, the proportion of households receiving credit, and
the amount of total annual rainfall, and 𝐷𝑡 is a vector of year dummies that capture time trends.

96

The existence of 𝛽𝑣 , village level time-invariant unobserved factors that are correlated with Ginivt
and Popvt (or Zvt) causes OLS estimator of equation (1) biased and inconsistent. To eliminate 𝛽𝑣
and obtain a consistent estimator, we estimate the first differenced version of equation (1) as
∆𝐺𝑖𝑛𝑖𝑣𝑡 = 𝛼1 Δ𝑃𝑜𝑝𝑣𝑡 + 𝛼2 ∆𝑍𝑣𝑡 + 𝛼3 ∆𝐷𝑡 + Δ𝜀𝑣𝑡

(2)

The coefficient of interest is 𝛼1 in this model.
To estimate the effects of land inequality on agricultural production, income, and poverty,
we specify the reduced form equation as
𝑦𝑖𝑣𝑧𝑡 = 𝛾0 + 𝛾1 𝐺𝑖𝑛𝑖𝑣𝑧𝑡 + 𝛾2 𝑋𝑖𝑣𝑧𝑡 + 𝛾3 𝐴𝐸𝑍𝑧𝑡 + 𝛾4 𝐷𝑡 + 𝛾5 𝐴𝐸𝑍𝑧𝑡 ∗ 𝐷𝑡 + 𝛿𝑖𝑣𝑧 + 𝜖𝑖𝑣𝑧𝑡

(3)

where 𝑦𝑖𝑣𝑧𝑡 is alternately the value of total crop production per acre or per capita, net crop
income which is the difference between the value of crop production and input costs per acre or
per capita, net livestock income which is the difference between the value of livestock
production and input costs per capita, net non-farm income which is net revenue from non-farm
sectors per capita, net total income which is a sum of net crop, livestock and non-farm income
per capita, or poverty status (= 1 if income per capita is above the national poverty line, = 0
otherwise). The subscripts, i, v, z, t stand for household, village, agro-ecological zone and year.
𝑋𝑖𝑣𝑧𝑡 is a vector of household characteristics including endowment of agricultural land, a dummy
for female-headed household, a dummy for head with primary education, a dummy for
households receiving credit, head’s age, the value of total asset, distance from homestead to the
tarmac road, and distance from homestead to extension service. 𝐴𝐸𝑍𝑧𝑡 is a vector of agroecological dummy variables, which is expected to capture common shocks at the level of agroecological zone. Interaction terms between agro-ecological zone dummies and year dummies are
also included to control for the time specific localized shocks at each agro-ecological zone. 𝛿𝑖𝑣𝑧

97

includes village time-invariant unobserved factors (e.g., unmeasured weather, technology,
productivity, land quality and so on) and household level factors (e.g., household head’ innate
ability, risk preference and so on). Estimation of (3) by OLS generates biased and inconsistent
estimates because 𝛿𝑖𝑣𝑧 could affect 𝑦𝑖𝑣𝑧𝑡 and 𝐺𝑖𝑛𝑖𝑣𝑧𝑡 simultaneously. Hence, we take the first
difference of the equation (3) to eliminate 𝛿𝑖𝑣𝑧 as:
Δ𝑦𝑖𝑣𝑧𝑡 = 𝛾1 Δ𝐺𝑖𝑛𝑖𝑣𝑧𝑡 + 𝛾2 Δ𝑋𝑖𝑣𝑧𝑡 + 𝛾3 Δ𝐴𝐸𝑍𝑧𝑡 + 𝛾4 Δ𝐷𝑡 + 𝛾5 ΔAEZ𝑧𝑡 ∗ 𝐷𝑡 + Δ𝜀𝑖𝑣𝑧𝑡 .

(4)

The coefficient of interest in this estimation is 𝛾1 .
We also explore the heterogeneous effects of land inequality on income and poverty
across households of different socio-economic backgrounds. First, we added to equation (4) an
interaction term between the Gini coefficient and a dummy for households with value of total
assets in the lowest quartile in a village. Doing so allows us to assess whether land inequality
affects poor and rich households differently. Second, we added to equation (4) an interaction
terms between the Gini coefficient and a dummy for households whose agricultural land
ownership is in the lowest quartile in a village to assess whether the effects of land inequality are
different across households with different land endowments. Finally, to assess whether land
inequality affects households with different levels of education, we add to equation (4) an
interaction term between the Gini coefficient and a dummy for heads who have completed
secondary education.

3.4 Data and Sample Characteristics
Our study relies on the panel household data from a repeated nationwide household
survey in rural Kenya conducted by Tegemeo Institute of Egerton University in Kenya. The

98

survey started in 1997 and the sampling frame was designed in consultant with the Kenya
National Bureau of Statistics. In the sampling frame, 24 administrative districts were chosen to
represent all major agro-ecological zones in Kenya. Then, proportionally to population in agroecological zones, one or more administrative divisions were picked from a selected district. After
that, villages and households were randomly drawn from a selected division. Eventually, the
survey interviewed 1540 households from 111 villages in 39 divisions in 24 districts and in 8
agro-ecological zones in 1997. The subsequent surveys were conducted in 2000, 2004, 2007, and
2010. The attrition rate between two consecutive surveys is about 5%. A test of the existence of
systematic attrition in the data was conducted and the F-ratio test indicates no evidence of
systematic attrition.
This household survey includes information on household characteristics, assets,
geographical information, agriculture production and different income sources as well as
community characteristics. The relatively long panel household data allow us to identify the
impacts of land inequality on agricultural productivity, income, and poverty by the panel
estimation methods. The village level Gini coefficient of agricultural land is used as a measure of
land inequality in a community. To ensure the Gini coefficient is a reliable measure of land
inequality, we excluded all the villages with sample size less than 10 households in our analysis.
To deal with outlier issues, we further drop the observation with the value of production and all
income in the top 99% of the sample.
The data used to estimate the determinants of land Gini is from multiple sources. While
the Gini coefficient and many of the right hand side variables in equation (2) are calculated from
the above-mentioned household panel data, we only rely on other data sources to calculate one of

99

the key right hand side variables – population density of each village. Specifically, the population
density is based on the Global Rural Urban Mapping Project dataset (GRUMP), and the global
land cover dataset (GlobCover2009) assembled by Chamberlin (2012). The village level
population density is computed as the ratio of total population by total arable land area in a
village.
Table 3.1 Household characteristics
1997

2010

2010
Western
High
Traditional
Potential
& Western
Maize
Highlands
Zone
Zone
25%
25%

National

National

Eastern &
Western
Lowlands
Zone

Female-headed HH (%)

14%

27%

34%

Head completed primary education (%)

54%

60%

56%

58%

60%

65%

Age of the head (years)

50.3

60.4

60.3

59.1

60.0

63.0

Household size

6.6

5.5

6.2

5.7

6.0

3.8

Owned farm size (acres)

3.2

2.9

2.5

2.3

4.8

1.8

Value of total asset (KSh)

238,030

312,636

242,368

142,216

432,179

434,680

Rented-in lands (%)

18%

17%

12%

22%

24%

10%

Received credit (%)

40%

54%

50%

55%

43%

86%

Distance to tarmac road (km)

8.2

7.1

8.6

6.5

6.4

5.0

Distance to extension service (km)

5.4

5.4

6.6

4.7

6.1

3.7

Value of crop production (KSh/acre)

45,171

40,869

25,302

38,999

29,242

87,707

Net crop income (KSh/acre)a

38,164

37,309

23,780

35,142

25,261

81,528

20,924

21,403

12,800

18,416

18,581

44,406

30,941

14,432

5,512

9,700

23,549

21,341

8,546

18,525

19,970

11,077

18,260

24,338

59,632

57,636

41,002

40,655

62,317

98,324

Net crop income (KSh/capita)

a

Net livestock income (KSh/capita)

b

Net non-farm income (KSh/capita)
Net total income (KSh/capita)d

c

Headcount poverty ratio (%)
44%
38%
50%
49%
37%
a
Net crop income is defined as the value of crop production minus input costs.
b
Net livestock income is defined as the value of livestock production minus costs.
c
Net non-farm income is defined as the revenue from non-farm sector minus costs.
d
Net total income is summation of net crop income, net livestock income, and net non-farm income.

100

Central
Highland
Zone
23%

7%

3.4.1 Household Characteristics
Table 3.1 describes characteristics of households from 1997 to 2010. Following Jin and
Jayne (2013), we divide the total sample into four main zones based on the agro-ecological
conditions and agricultural productivity potential. They are respectively Eastern and Western
Lowlands, Western Traditional and Western Highlands, High Potential Maize, and Central
Highlands. The agricultural productivity potential is relatively low in lowlands and high in
highlands.
There are notable intertemporal and interzonal differences in household characteristics in
many aspects. The average proportion of female-headed households increased from 14% to 27%
from 1997 to 2010. This varies from 23% in Central Highland Zone to 34% Eastern and Western
Lowlands Zone in 2010. The completion rate of primary education of a household head increased
from 54% to 60% over time ranging from 56% in Eastern and Western Lowlands Zone to 65% in
Central Highland Zone. The average household head aged by about 10 years, which is not
surprising as the panel data tracks same households for 13 years. The mean sizes of household
and own farmland shrank by 1.1 and 0.3 acres respectively over time. While Central Highland
Zone has the smallest mean household size (3.8) and own farmland (1.8 acres), Eastern and
Western Lowlands Zone has the largest household size (6.2) and High Potential Maize Zone has
the biggest own farm (4.8 acres). The average total value of assets has increased by 31% over
time. The value of total assets in Central Highland Zone (434,680 Kenyan Shieling (KSh)16) is
about three times as large as that in Western Traditional and Western Highlands (142,216 KSh).
The change in proportion of households who rented-in land over time is negligible. The land

16

Throughout this chapter, all prices are converted to the real price setting 2010 as a base year. The consumer price
index for 1997, 2000, 2004 and 2007 are 28.62, 29.77, 36.78, 47.9, and 64.86 respectively.

101

rental is most active in High potential Maize Zone and least active in Central Highland Zone.
The access to credit has improved over time as the proportion of households who received credit
increased by 14% from 1997 to 2010. The credit access is most active in Central Highland Zone
(86%) and least in High potential Maize Zone (43%). The transportation infrastructure seems
improved in Kenya, as the average distance from homestead to the tarmac road shortened by 1.1
km overtime. However, the average distance from home to the extension service stays same over
time. While Central Highland Zone has best infrastructure, as the distances to the tarmac road
and to extension service are the shortest, Eastern and Western Lowlands has poorest
infrastructure as these distances are farthest.
This study uses the value of all crop production per acre and net crop income per acre,
which is the difference between the value of crop production and input costs, to measure
agricultural land productivity. Though the average value of crop production per acre has declined
by 4,300 KSh over time, the net crop income per acre stayed almost the same from 1997 to 2010.
There are huge disparities of the value of crop production per acre and net crop income per acre
across zones in 2010, as the value of crop production and net crop income in Central Highland
Zone are more than 3 times higher than those in Eastern and Western Lowlands Zone.
Household net total income is the summation of net crop income, net livestock income,
and net non-farm income. Each net income is computed as gross income minus costs. The
change in the average net crop income per capita is negligible over time. The net livestock
income per capita significantly dropped from 1997 to 2010, which is probably due to an outbreak
of Rift Valley fever, a viral disease communicable to animals such as cows, sheep, and goats, in
2006 and 2007. It is highest in High Potential Maize Zone and lowest in Eastern and Western

102

Lowlands Zone. The net non-farm income per capita has increased more than doubly over time.
The net non-farm income per capita of households in Central Highland Zone is more than double
in Western Traditional and Western Highlands. The net total income per capita has declined
slightly over time. There are also great disparities of net total income per capita among zones.
The net total income per capita of households in Central Highland Zone is two times as great as
that in Eastern and Western Lowlands Zone. The headcount poverty ratio reduced by 6% over
time. The poverty ratio is significantly lower for households in Central Highland Zone with an
average ratio at 7%.
Based on the household level descriptive statistics, it is evidenced that there are huge
gaps in economic conditions across zones in Kenya. Though Central Highland Zone has the
smallest household size and own farm size, the agricultural intensification seems to be most
advanced, as the agricultural land productivity is much higher than other areas. Additionally,
Central Highland Zone has most active non-farm sector economy. As a result, Central Highland
Zone has the highest income level and lowest poverty ratio in Kenya.

3.4.2 Population Density and Land Distribution in Kenya
Table 3.2 describes village level population density and agricultural land distribution over
time and across zones. The average village level population density in Kenya grew from 619
people per km2 to 893 people per km2 from 1997 to 2010. On the other hand, the average village
level Gini coefficient of own agricultural land dropped by 0.11 (from 0.47 to 0.36), suggested
that agricultural land became more equally distributed over time. The more equalized land
distribution over time is also evidenced in the Lorentz curves in figure 3.1. Interestingly, Central

103

Highland Zone which has the highest village level population density (5 times greater than that
in High Potential Maize Zone) also has the lowest village level Gini coefficient.
Table 3.2 Village characteristics
1997

2010

National

Population density (persons/km2)

619

893

Gini coefficient of own farm size

0.47

0.36

0.36

2010
Western
High
Traditional
Potential
& Western
Maize
Highlands
Zone
Zone
749
368
0.33

Central
Highland
Zone
1,912

0.44

0.31

0

.2

.4

.6

.8

1

National

Eastern
&
Western
Lowlands
Zone
925

0

.2

.4
.6
Cumulative share of population
1997
equality

.8

1

2010

Figure 3.1 Lorenz curve of national own agricultural landholding distribution over time (19972010)

104

10.6
10.55
10.5
10.45
10.4
10.35

linear fit
.2

.4
.6
Gini coefficint of owned farm land

95% CI
.8

Figure 3.2 Correlation between log of income per capita and the Gini coefficient of own farm
land
Note: The fitted line is regression line of log of household net total income per capita (KSh) on
the village level Gini coefficient of owned farm land based on pooled household data. The
regression coefficient and the standard error are -.107 and 0.13. The gray area is 95% confidence
interval.
Figure 3.2 shows the negative correlation between the household net total income and the
village level Gini coefficient of own farm. In the next section, this relationship is further assessed
by panel regression analysis that controls for time, regional, and household characteristics.

105

Table 3.3 Determinants of the Gini coefficient of owned farm lands (village first-differencing
and fixed effects model, village level data)
Gini coefficient
FD
FE
Explanatory variables
(1)
(2)
2
Population density (100 persons/km )
-0.00511**
-0.00401*
(0.00212)
(0.00202)
Log of average household's owned land size (acre)
-0.0271
-0.0390
(0.0282)
(0.0323)
Log of average household size
0.0346
-0.00422
(0.0394)
(0.0457)
% of households who rented-in lands
0.0508
0.0332
(0.0367)
(0.0392)
% of households who received credit
-0.0115
-0.0290
(0.0234)
(0.0297)
Log of average distance to tarmac road (km)
-0.0153
-0.00145
(0.0137)
(0.0167)
Log of average distance to extension advice (km)
0.00345
-0.00278
(0.00930)
(0.0103)
Log of quantity of annual rainfall (mm)
-0.00357
0.0342
(0.0268)
(0.0313)
year 2000 dummy
0.00444
(0.00443)
year 2004 dummy
-0.0562***
-0.0551***
(0.0151)
(0.0143)
year 2007 dummy
-0.0493***
-0.106***
(0.0107)
(0.0161)
year 2010 dummy
-0.00295
-0.104***
(0.0137)
(0.0185)
Constant
0.00569*
0.238
(0.00323)
(0.315)
Observations
355
459
Number of village
100
R-squared
0.103
0.370
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.

106

Table 3.4 Effect of the village Gini coefficient of owned farm land on agricultural land
productivity (household first differenced model, household level data)a
Log of value of crop
Log of net crop
production (KSh/acre)
income (KSh/acre)b
VARIABLES
(1)
(2)
Gini coefficient of owned farmland
-0.425**
-0.514***
(0.166)
(0.181)
Log of owned farm size
-0.0245***
-0.0223***
(0.00676)
(0.00747)
Log of household size
0.0212
0.00980
(0.0360)
(0.0402)
Head's age
-0.000530
-0.00140
(0.00237)
(0.00256)
Female-headed (=1)
-0.00442
-0.0810
(0.0642)
(0.0689)
Head completed primary education (=1)
0.000531
-0.0265
(0.0416)
(0.0464)
Received credit (=1)
0.0121
-0.0185
(0.0263)
(0.0279)
Log of value of total assets (KSh)
0.0277*
0.0365**
(0.0146)
(0.0153)
Log of average distance to tarmac road (km)
-0.000505
0.00779
(0.0184)
(0.0204)
Log of average distance to extension service
-0.0134
-0.0129
(km)
(0.0114)
(0.0128)
Log of rainfall in the main cropping season
0.0541
0.0699*
(mm)
(0.0377)
(0.0381)
Constant
0.565***
0.814***
(0.151)
(0.181)
Observations
4,505
4,506
R-squared
0.207
0.602
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2012 and agro-ecological zone, agro-ecological zone and year
dummies are included in all regressions.
b
Net crop income is defined as the value of crop production minus input costs.

107

3.5 Estimation Results
3.5.1 Determinants of Land Distribution
The regression results for the determinants of village land distribution are reported in
Table 3.3. For robustness check, both first differenced (column 1) and fixed effect (column 2)
estimation results are reported. Population density has significant and negative impact on
inequality of household’s own agricultural land within a village. An increase in population
density of 100 persons per km2 would lead to a reduction in the Gini coefficient by 0.005 in the
first differenced model and by 0.004 in the fixed effect model. This could indicate that
population pressure to land accelerates fragmentation of large land and distributed land to
smallholders. Both the first differenced model and the fixed effect model show a declining trend
of the Gini coefficient after 2000, which is consistent with the evidence in the descriptive
analysis.

3.5.2 Effects of Land Distribution on Agricultural Land Productivity
Table 3.4 presents the results on the effects of land inequality on land productivity which
are measured by log of the value of all crop production per acre (column 1) or by log of net crop
income per acre (column 2). As expected, land inequality has significant and negative effects on
agricultural productivity. An increase in the Gini coefficient by 0.1 would lead to a reduction of
the value of crop production and net crop income by 4.3% and by 5.2% respectively. The results
also show statistically significant inverse relationship between land size and productivity, a
phenomenon that has long been observed and studied (Schultz, 1964; Bardhan, 1973; Feder,
1985; Hayami and Otsuka, 1993; Binswanger et al., 1995; Vollrath, 2007; Hazell, 2011). An

108

increase in own farm size by 1 acre would lead to a decline in the value of crop production per
acre by 2.5% and net crop income per acre by 2.2%. The presence of inverse farm-size and
productivity relationship implies that redistributing land from large farmers to land constrained
smallholders in the same community would be productivity enhancing. The positive and
significant coefficient on the value of total assets is also consistent with expectation. In the
presence of credit and other market imperfection, wealthy households have better ability to
afford more and better inputs and to own and use agricultural machinery and bullocks. Majority
of the other variables also have expected sign though mostly insignificant. For example, rainfall
has positive effect on productivity (though only significant in the case of the net crop income),
but female headship has negative effect on crop productivity.

3.5.3 Effects of Land Distribution on Income and Poverty
Table 3.5 exhibits the estimation results of the effects of land distribution on household
income and poverty. Net household income is classified into 4 categories: net crop income
(column 1), net livestock income (column 2), net non-farm income (column 3), and net total
income which is the summation of net crop income, net livestock income, and net non-farm
income (column 4). The poverty status is measured by a dummy variable (“non-poor”), which
equals one if the net income per capita is above the national poverty line and zero otherwise
(column 5).

109

Table 3.5 Effect of village Gini coefficient of owned farm land on income and poverty
(household first differenced model, household level data)a

b

Log of net
livestock
income
(KSh/capita)

Log of net
non-farm
income
(KSh/capita)

c

d

(1)
-0.410**
(0.195)

(2)
-0.656**
(0.307)

(3)
-0.500*
(0.269)

(4)
-0.383**
(0.171)

0.0542***
(0.00775)
-0.842***
(0.0445)
0.00470
(0.00290)
-0.0682
(0.0771)
0.0351
(0.0498)
0.0326
(0.0296)

-0.00379
(0.0119)
-0.620***
(0.0606)
0.00213
(0.00397)
0.0733
(0.112)
0.0545
(0.0747)
0.0383
(0.0441)

0.0151
(0.0103)
-0.439***
(0.0547)
-0.00240
(0.00376)
-0.328***
(0.0975)
0.256***
(0.0637)
0.0558
(0.0380)

0.0322***
(0.00652)
-0.746***
(0.0396)
0.00309
(0.00230)
-0.109
(0.0668)
0.162***
(0.0403)
0.0581**
(0.0254)

0.144***
(0.0172)
0.00386
(0.0213)

0.300***
(0.0264)
0.0198
(0.0314)

0.0978***
(0.0207)
-0.0403
(0.0284)

0.153***
(0.0162)
0.00314
(0.0193)

(5)
-0.276***
(0.0943)
0.0135**
*
(0.00373)
-0.248***
(0.0194)
0.000357
(0.00128)
-0.145***
(0.0356)
0.0522**
(0.0229)
0.0301**
(0.0139)
0.0573**
*
(0.00742)
-0.00349
(0.00993)

-0.00400
(0.0139)
0.0322
(0.0395)
0.707***
(0.174)
4,508
0.637

0.0237
(0.0196)
0.180***
(0.0629)
-1.097***
(0.207)
4,516
0.847

-0.00620
(0.0174)
-0.0779
(0.0528)
0.282
(0.186)
4,528
0.798

0.000150
(0.0111)
-0.0118
(0.0337)
-0.0467
(0.126)
4,523
0.453

0.00954
(0.00628)
0.0311
(0.0191)
0.142*
(0.0767)
4,589
0.098

Log of net
crop income
(KSh/capita)
VARIABLES
Gini coefficient of owned farm land

Log of owned farm size
Log of household size
Head's age
Female-headed (=1)
Head completed primary education (=1)
Received credit (=1)

Log of value of total assets (KSh)
Log of average distance to tarmac road (km)
Log of average distance from extension service
(km)
Log of rainfall in the main cropping season (mm)
Constant

Log of net
total income
(KSh/capita)

Non-poor
dummyf

e

Observations
R-squared
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2012 and agro-ecological zone, agro-ecological zone and year dummies are included in all
regressions.
b
Net crop income is defined as the value of crop production minus input costs.
c
Net livestock income is defined as the value of livestock production minus costs.
d
Net non-farm income is defined as the revenue from non-farm sector minus costs.
e
Net total income is summation of net crop income, net livestock income, and net non-farm income.
f
Non-poor dummy is equal to one if income per capita is above national poverty line.

110

Like in the case of productivity, the coefficient on land inequality is consistently negative
and statistically significant for all models. An increase in the Gini coefficient by 0.1 would
reduce the net crop income per capita, net livestock income per capita, net non-farm income per
capita and net total income per capita by 4.1%, 6.6%, 5.0%, and 3.8%, respectively. A same
change in the Gini coefficient would also decrease the chance of an average household to fall
into poverty by 2.8%. Additionally, access to own agricultural land has a positive and significant
impact on the net crop income, net total income and poverty status. An increase in own farm size
of 1 acre raises the net crop income by 5.4%, net total income by 3.2%, and the chance of getting
out of poverty by 1.4%.
The effects of other household’ characteristics on income and poverty have mostly
expected signs. For example, a female-headed household would earn lower non-farm income (by
33%) and is more likely to fall into poverty (15%) than a male-headed household holding other
factors constant. Head’s education has significant and positive effects on non-farm income per
capita, total income per capita and poverty status, and positive but insignificant effects on crop
income and livestock income. Specifically, a household with its head who completed primary
school education or higher would earn 26% higher non-farm income and 16% higher total
income than other households, and reduce the probability of the household to fall below poverty
line by 5%. Another highly consistent result is the positive and significant coefficients on the
value of total assets in all models. Doubling the value of total assets would increase the net crop
income, net livestock income, net non-farm income, and net total income by 14%, 30%, 10%,
and 15 % respectively, and would reduce the likelihood of a household to fall into poverty by
6%.

111

Table 3.6 Heterogeneous effect of village Gini coefficient of owned farm land on income and poverty (household first differenced
model, household level data)a

Explanatory variables
For those whose value of total assets are lowest quartile in a village
Gini coefficient of owned farm land (α)
Gini coefficient * lowest 25% group of value of assets in village (β)
Log of value of total assets (KSh)
Test α+β=0 (p-value)
For those whose owned farm sizes are lowest quartile in a village
Gini coefficient of owned farm land (α)
Gini coefficient * lowest 25% group of owed farms in village (β)
Log of owned farm land (acre)
Test α+β=0 (p-value)
For those whose head completed secondary education
Gini coefficient of owned farm land (α)
Gini coefficient * head completed secondary education dummy (β)
Head completed secondary education dummy

Log of net crop
income
(KSh/acre)b

Log of net
livestock
income
(KSh/capita)c

Log of net nonfarm income
(KSh/capita)d

Log of net total
income
(KSh/capita)e

Non-poor
dummyf

-0.536***
(0.181)
0.0808
(0.0901)
0.0444**
(0.0174)
0.0201

-0.539*
(0.309)
-0.414***
(0.132)
0.260***
(0.0285)
0.0026

-0.443
(0.272)
-0.197*
(0.115)
0.0786***
(0.0232)
0.0228

-0.326*
(0.172)
-0.202**
(0.0804)
0.133***
(0.0183)
0.0039

-0.251***
(0.0944)
-0.0914**
(0.0434)
0.0485***
(0.00814)
0.0007

-0.548***
(0.180)
0.369***
(0.0977)
-0.00187
(0.00953)
0.3859

-0.629**
(0.307)
-0.299*
(0.158)
-0.0202
(0.0148)
0.0058

-0.492*
(0.269)
-0.0863
(0.134)
0.0104
(0.0127)
0.0508

-0.338**
(0.171)
-0.497***
(0.0917)
0.00497
(0.00803)
0

-0.260***
(0.0944)
-0.183***
(0.0526)
0.00343
(0.00468)
0

-0.455**
(0.195)
-0.142
(0.306)
0.0644
(0.141)
0.0434

-0.463
(0.329)
-1.080**
(0.480)
0.373*
(0.216)
0.0012

-0.563*
(0.295)
0.169
(0.403)
0.123
(0.191)
0.3155

-0.311*
(0.187)
-0.273
(0.253)
0.154
(0.115)
0.0183

-0.252**
(0.101)
-0.0645
(0.156)
0.0249
(0.0722)
0.0386

Test α+β=0 (p-value)
The numbers in parentheses are robust standard errors.
***, **, and * indicate significance at 1, 5, and 10%, respectively.
a
Interaction terms between year 2012 and agro-ecological zone, agro-ecological zone and year dummies are included in all regressions.
b
Net crop income is defined as the value of crop production minus input costs.
c
Net livestock income is defined as the value of livestock production minus costs.
d
Net non-farm income is defined as the revenue from non-farm sector minus costs.
e
Net total income is summation of net crop income, net livestock income, and net non-farm income.
f
Non-poor dummy is equal to one if income per capita is above national poverty line.

112

3.5.4 Heterogeneous Effects of Land Distribution on Income and Poverty
The estimation results so far show negative effects of unequal land distribution on land
productivity, household income and poverty. It is also important to understand how these effects
vary across households with different socio-economic backgrounds. Table 3.6 displays the results
of models to explore the heterogeneous effects of land inequality across households along three
dimensions: non-land asset ownership (top panel), land endowment (middle panel), and head’s
education (bottom panel).
The results show considerable heterogeneous effects of land inequality across households
in different groups. The heterogeneous effects are especially strong between households with
different levels of asset ownership (non-land asset or land alike). For an average household in the
lowest asset quartile, an increase in the Gini coefficient of 0.1 would reduce the net livestock
income per capita, net non-farm income per capita, net total income per capita, and probability of
being a non-poor household by an additional 4.1%, 2.0%, 2.0% and 1% respectively (top panel).
This indicates that relatively poor households in a village are more likely to be negatively
affected by unequal land distribution than others. Similarly, a constraint on land access would
exacerbate the negative effects of land inequality; for households in the lowest land quartile, an
increase in the Gini coefficient would decrease the net livestock income per capita, net total
income per capita, and probability of being a non-poor household by an additional 3.0%, 5.0%,
and 1.8% (middle panel). On the other hand, the coefficient of the interaction term between the
Gini coefficient and a lowest quartile farmland dummy in the model of net crop income per acre
is positive and significant, which could be due to the inverse land size and productivity
relationship. The heterogeneous effects along the dimension of head’s education is more mixed

113

(bottom panel). While land inequality has significant and negative effect on non-farm income for
the households whose heads did not complete secondary education, the effect is not significant
any more for those with heads having completed secondary education or higher. But on the other
hand, land inequality has even greater negative effect on household’s livestock income for a
household with a more educated head. Overall, the effects of land inequality on household total
income and poverty status are not significantly different between households with heads of
different levels of education.

3.6 Conclusion and Policy Recommendation
Though the relationship between unequal asset distribution and economic growth has
long been discussed among researchers, most of the empirical studies on this topic were based on
country level data and the empirical studies based on micro-level data are surprisingly few.
Therefore, this study attempts to contribute to this important topic by assessing the effects of
land distribution on agricultural productivity, income and poverty using a long household panel
data in Kenya.
Our findings are generally consistent with the large theoretical literature on the
relationship between asset distribution and economic growth. First, we find that the village level
land inequality has significant and negative effects on land productivity. Increasing village level
Gini coefficient of agricultural land by 0.1 would cause a reduction of the value of crop
production per acre and net crop income per acre by 4.3% and 5.1% each. Second, unequal
agricultural land distribution is negatively related to household income and poverty status. If the
village level Gini coefficient of own farmland increases by 0.1, the net crop income per capita,

114

net livestock income per capita, net non-farm income per capita, net total income per capita, and
chance of being out of poverty would decrease by 4.1%, 6.6%, 5.0%, 3.8% and 2.8%
respectively. Third, we find that the effects of land inequality vary with households of different
socio-economic backgrounds. In particular, the negative effects of land inequality are
significantly larger for households that are poorer and endowed with smaller amount of
agricultural land. Improving land distribution would lead to disproportionate benefits for the
households in the bottom of the poverty bracket.
The implication of our findings is clear. A more equalized land distribution is productivity
and welfare enhancing for rural farmers in Kenya in particular and in Sub-Saharan Africa
countries in general. The question that remains unanswered is how to reduce land inequality.
There are several ways to promote equal land distribution including, land reform, promotion of
land sales and rental markets, land taxes, and so on. Empirical studies find that effects of these
policies are mixing. For example, Yamano et al. (2009) argue that both land sales and land rental
markets improve equity and efficiency among rural farmers in Kenya. Additionally, Jin and
Jayne (2013) find that land rental market transfers land from large landholders to small ones.
However, others argue that the land reform based on market-based “willing-seller/willing buyer”
system implemented in South Africa has resulted in large-scale commercial farming by the black
elite at the expense of the poor smallholders (Lahiff and Cousins, 2005). Bardhan et al. (2014)
reveals that land distribution program in West Bengal lowered land inequality but this effect was
canceled out by by the rising inequality caused by inequality rising effects of population growth.
Assuncao (2008) find that the land reform implemented in Brazil during 1992 and 2003 raised
land inequality among households. Therefore, more research is needed in order to find effective

115

policies to lower land inequality.
Additionally, in order to identify the causes for land distribution, we need to have data
from a large number of communities with more complete and accurate information on historical
as well as current social, cultural, economic, political and institutional conditions. Based on the
limited information, we find population pressure is helping equalize land distribution but the
magnitude of effect is extremely small, which implies that land inequality is likely to be related
to many other factors for which we do not have information. Therefore, further studies about the
determinants of land distribution based on more detailed community level data are required in
order to formulate the effective policies to improve land distribution.

116

REFERENCES

117

REFERENCES

Aghion, P., & Bolton, P. (1997). A theory of trickle-down growth and development. The Review
of Economic Studies, 64(2), 151–172.
Aghion, P., Caroli, E., & Garcia-Penalosa, C. (1999). Inequality and economic growth: the
perspective of the new growth theories. Journal of Economic Literature, 1615–1660.
Bardhan, P., Bowles, S., & Gintis, H. (2000). Wealth inequality, wealth constraints and economic
performance. Handbook of Income Distribution, 1(10), 541–603.
Bardhan, P. K. (1973). Size, productivity, and returns to scale: An analysis of farm-level data in
Indian agriculture. The Journal of Political Economy, 1370–1386.
Bardhan, P., Luca, M., Mookherjee, D., & Pino, F. (2014). Evolution of land distribution in West
Bengal 1967–2004: Role of land reform and demographic changes. Journal of Development
Economics, 110, 171–190.
Bardhan, P., & Udry, C. (1999). Development microeconomics. New York: Oxford University
Press
Barret, C. B., Bezuneh, M., Clay, D. C., & Reardon, T. (2005). Heterogeneous Constraints,
Incentives and Income Diversification Strategies in Rural Africa. Quarterly Journal of
International Agriculture, 44(1), 37-60.
Benabou, R. (1996). Inequality and growth. In B. Bernanke & J. Rotemberg, (Eds.), NBER
Macroeconomics Annual 1996, Volume 11 (pp. 11–92). Cambridge, MA: MIT Press.
Benjamin, D., Brandt, L., & Giles, J. (2011). Did Higher Inequality Impede Growth in Rural
China?. The Economic Journal, 121(557), 1281–1309.
Binswanger, H. P., Deininger, K., & Feder, G. (1995). Power, distortions, revolt and reform in
agricultural land relations. Handbook of Development Economics, 3, 2659–2772.
Binswanger, H. P., & Rosenzweig, M. R. (1986). Behavioral and material determinants of
production relations in agriculture. The Journal of Development Studies, 22(3), 503–539.
Bourguignon, F., & Verdier, T. (2000). Oligarchy, democracy, inequality and growth. Journal of
Development Economics, 62(2), 285–313.
Cardenas, J.-C. (2003). Real wealth and experimental cooperation: experiments in the field lab.
Journal of Development Economics, 70(2), 263–289.
118

Chamberlin, J. (2013) Data Extracted from the GIS for Household Survey Points. East Lansing
MI: Michigan State University, mimeo.
Conning, J. H., & Robinson, J. A. (2007). Property rights and the political organization of
agriculture. Journal of Development Economics, 82(2), 416–447.
Deininger, K., Jin, S., & Nagarajan, H. K. (2009). Land Reforms, Poverty Reduction, and
Economic Growth: Evidence from India. Journal of Development Studies, 45(4), 496–521.
Deininger, K., & Squire, L. (1998). New ways of looking at old issues: inequality and growth.
Journal of Development Economics, 57(2), 259–287
De Janvry, A., Fafchamps, M., & Sadoulet, E. (1991). Peasant household behaviour with missing
markets: some paradoxes explained. The Economic Journal, 1400–1417.
Durlauf, S. N. (1996). A theory of persistent income inequality. Journal of Economic Growth,
1(1), 75–93.
Feder, G. (1985). The relation between farm size and farm productivity: The role of family labor,
supervision and credit constraints. Journal of Development Economics, 18(2), 297–313.
Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of agricultural innovations in
developing countries: A survey. Economic Development and Cultural Change, 255–298.
Galor, O., & Zeira, J. (1993). Income distribution and macroeconomics. The Review of Economic
Studies, 60(1), 35–52.
Griffin, K., Khan, A. R., & Ickowitz, A. (2002). Poverty and the Distribution of Land. Journal of
Agrarian Change, 2(3), 279–330.
Gunning, J. W., Hoddinott, J., Kinsey, B., & Owens, T. (2000). Revisiting forever gained:
Income dynamics in the resettlement areas of Zimbabwe, 1983–96. Journal of Development
Studies, 36(6), 131–154.
Hayami, Y., & Otsuka, K. (1993). The Economics of Contract Choice: An Agrarian
Perspective.Oxford: Oxford University Press
Hazell, P. (2011, January). Five Big Questions about Five Hundred Million Small Farms. Paper
presented at the IFAD Conference on New Directions for Smallholder Agriculture, Rome,
Italy. Retrieved from http://www.ifad.org/events/agriculture/doc/papers/hazell.pdf

119

Holden, S., Otsuka, K., Place, F., (2009). Land markets and poverty in perspective. In S. T.
Holden, K. Otsuka & F. M. Place (Eds.), The Emergence of Land Markets in Africa (pp. 273296). Washington, DC; RFF Press
Jayne, T. S., Yamano, T., Weber, M. T., Tschirley, D., Benfica, R., Chapoto, A., & Zulu, B.
(2003). Smallholder income and land distribution in Africa: implications for poverty
reduction strategies. Food Policy, 28(3), 253–275.
Jin, S., & Jayne, T. S. (2013). Land rental markets in Kenya: implications for efficiency, equity,
household income, and poverty. Land Economics, 89(2), 246–271.
Kanyinga, K.(2009). Land redistribution in Kenya.In H.P. Binswanger-Mkhize, C. Bourguignon,
& R. van den Brink (Eds.) Agricultural Land Redistribution: Toward Greater Consensus
(pp.87-118) Washington D.C.: World Bank.
Kimenyi, M. S., and Ndung’u, N. S. (2005) Sporadic Ethnic Violence: Why Has Kenya Not
Experienced a Full-blown Civil War? In P. Collier & N. Sambanis (Eds), Understanding
Civil War: Volume 1: Africa (pp.123-153) Washington D.C: World Bank.
Lahiff, E., & Cousins, B. (2005). Smallholder Agriculture and Land Reform in South
Africa. IDS Bulletin, 36(2), 127-131
Leo, C. (1989). Land and Class in Kenya, Harare: Nehanda Publishers.
Loury, G. C. (1981). Intergenerational transfers and the distribution of earnings. Econometrica,
843–867.
Mamdani, M. (1996). Citizen and subject: Contemporary Africa and the legacy of late
colonialism. Princeton: Princeton University Press.
Muyanga, M. C (2013). Smallholder Agriculture in the Context of Increasing Population
Densities in Rural Kenya (Doctoral dissertation). Retrieved from
http://etd.lib.msu.edu/islandora/object/etd%3A904/datastream/OBJ/view
Nargis, N., & Hossain, M. (2006). Income dynamics and pathways out of rural poverty in
Bangladesh, 1988–2004. Agricultural Economics, 35, 425–435.
Okoth-Ogendo, W.H.O. (1991). Tenants of the Crown: Evolution of Agrarian Law and
Institutions in Kenya. Nairobi: ACTS Press.
Quan, J. (2000). Land tenure, economic growth and poverty in sub-Saharan Africa. In C. Toulmin
& J. Quan (Eds.), Evolving land rights, policy and tenure in Africa (pp. 31-74) London:
DFID/IIED/NRI.

120

Ravallion, M., & Datt, G. (2002). Why has economic growth been more pro-poor in some states
of India than others? Journal of Development Economics, 68(2), 381–400.
Republic of Kenya. (2009). Sessional Paper No. 3 of 2009 on National Land Policy. Nairobi:
Ministry of Lands.
Schultz, T. W., & others. (1964). Transforming Traditional Agriculture. New Haven, CT: Yale
University Press.
Stiglitz, J. E. (1969). Distribution of Income and Wealth Among Individuals. Econometrica,
37(3), 382–397.
Stiglitz, J. E., & Weiss, A. (1981). Credit rationing in markets with imperfect information. The
American Economic Review, 393–410.
Syagga, P. (2010). Public land, historical land injustices and the new Constitution. (Constitution
Working Paper 9). Nairobi: Society for International Development (SID)
Vollrath, D. (2007). Land Distribution and International Agricultural Productivity. American
Journal of Agricultural Economics, 89(1), 202–216.
World Bank. (2005). World Development Report 2006: Equity and Development. Retrieved from
http://www-wds.worldbank.org/external/default/WDSContentServer/IW3P/IB/2005/09/20/00
0112742_20050920110826/Rendered/PDF/322040World0Development0Report02006.pdf
Yamano, T., Place, F. M., Nyangena, W., Wanjiku, J. & Otsuka, K. (2009). Efficiency and
Equity Impacts of Land Markets in Kenya. In . S. T. Holden, K. Otsuka & F. M. Place (Eds.),
The Emergence of Land Markets in Africa (pp.93-111). Washington, DC; RFF Press

121