a , .
. .va!
{ohm/fly»?!

a
“‘5‘“;

J.
.1

73’, P
22

V.

.

‘ o

,‘ .
w.

‘N.
‘l

n 3, a .
. . . u. 5
. . . "J§.'f~u.vr..vi.
. f .1 134...“)...
H aglvv; . . :. .
r! 1
a v Tau-1.31“: . V A

5. Irrr . v‘... .r.
5.1%
:33

.D. v

r $4....I‘I1IIQII’

9 I. "Z! .l.
"EV'v

to!"

5

I... 1;..rlpfl.v-
.1110}?! qt! 1

. . . ¢ Ky:

.1..d x...)

H
mic"

s
N

I O ‘ ‘-
l.!.)v| I)

s
L
.

\

\b (\Ju‘v.1‘¢.-‘v.rl‘ I. .I
{3.5.4} .I ’I..d
rl It’ll Iliu-

; 7!..rl\.".

 

 

Litter... 071.1raauezi. {3333-53. 1.... : "\W
‘ ‘ 7 . { .

Ill'iiilii'l'li'illiliiiliiiflllfilfliIiiil’ilfliilfil

3 1293 00910 9939

This is to certify that the
dissertation entitled
Analysis of Hood-Energy Production and Consumption
Strategies Among Small-Scale Farlers in Central Kenya

presented by

Albert Makanga Hwangi

has been accepted towards fulfillment
of the requirements for

Doctoral degree in Forestry

 

 

a jor professor I

Date 1 2/2 3/?2

 

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

 

_4‘—__ __—-_—-_—t_—

 

hh‘

; LIBRARY I
(Michigan State
University J

i
I
P

 

PLACE IN RETURN BOX to remove this checkout from your record. '
TO AVOID FINES return on or before date due.

DATE DUE DATE DUE DATE DUE

J

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU In An Affirmative Action/Equal Opportunity Institution ‘
cmma

ANALYSIS OF WOOD-ENERGY
PRODUCTION AND CONSUMPTION STRATEGIES

AMONG SMALL-SCALE FARMERS IN CENTRAL KENYA

BY

Albert Makanga Mwangi

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Forestry

1992

ABSTRACT

ANALYSIS OF WOOD-ENERGY PRODUCTION AND CONSUMPTION
STRATEGIES AMONG SMALL-SCALE FARMERS IN CENTRAL KENYA

BY
Albert Makanga Mwangi

This study focuses on wood-energy production and
consumption strategies among small-scale farm households in
central Kenya. The specific objectives were: (1) to
determine how households had responded to specific wood-
energy policies, and (2) to identify factors that were
associated with household adoption or non-adoption of the
strategies. Different programs aimed at addressing wood-
energy shortages in Kenya were initiated or strengthened
during the 19803. They included fuelwood or multipurpose
tree planting, development and dissemination of improved
stoves and fireplaces and promotion of increased
accessibility to wood-energy substitutes. Household
adoption levels for policy-supported-strategies have
remained low despite promotion efforts.

Cross-sectional survey data from two villages in Nyeri
district were collected and used to determine the factors
associated with adoption of the Kenya Ceramic Jiko, the
"Kuni Mbili" stove/fireplace, kerosene stoves, electric
cookers, and fuelwood or multipurpose tree planting.
Adoption rates varied from as low as 1 percent for
electricity to 43 percent for the Kenya Ceramic Jiko.

‘Different parametric and non-parametric tests were used to

evaluate differences between the two villages. Probit
analysis was used to examine the factors associated with
household adoption or non-adoption. Important policy
variables included extension visits per year, income levels,
years of formal education received by head of household,
access to different fuels, area of farm-land owned,
household size, and locational characteristics of the
villages.

Policy recommendations included: use of research
results to direct policy: improvement of information flows
between policy makers, extension agents, and technology-
users: increased support of agroforestry: and better program
coordination. Recommendations for further research
included: examining more areas where efficiency gains in
energy production and consumption can be made, extending the
study to cover the drier parts central Kenya, and conducting
regular case studies in order to better understand the

adoption process over time.

ACKNOWLEDGEMENTS

I would like to express my sincere thanks to my
dissertation director Dr. Larry Leefers, and also to my
other committee members Drs. Dan Chappelle, Karen Potter-
Witter, and Michael Gold. Their support, encouragement and
constructive criticism enabled me to complete this work.
Dr. Leefers took time from his busy schedule to visit me in
Kenya where I was collecting data. He also went through
many drafts of this dissertation.

Many thanks also go to my nine research assistants
(Ndagi, Mugo, Wambui, Wanjiku, Julia, Kimani, Kiongo,
Githua, and Ndirangu) who worked hard in the data collection
process. Others who were invaluable in this process
included Lawrence Wang’ondu, James Wahome, and local
administration officials in Kihugiru and Island Farms sub-
locations of Nyeri District. Officials from the Ministries
of Environment and Natural Resources, Energy and Regional
Development, and Economic Development and Planning were also
very helpful. This study could not have been carried out
without the cooperation of many farmers in the two study
areas and I am very grateful to them.

I am very grateful to all the organizations and

institutions that supported my studies and research. The

 

International Development and Research Center (IDRC)
provided me with a research grant that made this project
possible. The Fulbright Program supported my studies at
Michigan State University. Moi University granted me study
leave for the duration of my studies and also provided
transport during the data collection phase. This assistance
is gratefully acknowledged.

The support of my wife Margaret and our children Danson
and Joy was vital to the completion of my program. Margaret
consistently encouraged me to keep moving and also helped me
a lot with the typing and editing. Thank you very much for
your patience and understanding and may God richly bless

you.

TABLE OF CONTENTS

L I ST 0F TABLE S O O O O O O O O O O O O O O 0
LIST OF FIGURES . . . . . . . . . . . . . . .

I. INTRODUCTION . . . . . . . . . . . . .
Background and Problem Setting . . . .
Resource and Population Concerns
Wood-energy Programs

Specific Policies . .

Fuelwood Use . . . .

Research Objectives . . .
Scope and Limitations . .
Dissertation Organization

II. LITERATURE REVIEW . . . . . . . . . .
Research and Development Work by Various
Implications of Wood-Energy Shortages .
The Search for Solutions . . . . . . . .
Summary . . . . . . . . . . . . . . .-.

III. RESEARCH METHODS . . . .
Conceptual Framework . . .
Model Specification . . .
Variable Selection . . . .
Sampling Design and Prose dur

Sampling Frame . . .
Survey Type . . . . .
Survey Pretesting . .
Sample Size . . . . .
Administering the Survey .
Data Entry and Initial Analyse
Summary of Survey Efforts . .
Parametric and Non-parametric Tests
Econometric Models . . . . . . . . .
Summary . . . . . . . . . . . . . .

IV. RESULTS AND DISCUSSION . . . . . . . . .
Tests for Mean Differences . . . . . .
Discussion of Mean Difference Results .

Village Comparisons . . . . . . . .
Adopter/Non-adopter Comparisons . .

iii

Agencies

Duration of Use . . . . . . . . . . . . . . . . . 89
Attributes of Strategies . . . . . . . . . . . . . 89
Probit Results . . . . . . . . . . . . . . . . . 93
Assessing Estimation Results . . . . . . .‘. 93

Adoption of Kenya Ceramic Jiko in Village 1
(Model KCJl) . . . . . . . . . . . . . 98

Adoption of Kenya Ceramic Jiko in Village 2
(Model KCJZ) . . . . . . . . . . . . . . 101

Adoption of Kuni Mbili Stove/Fireplace in

Village 1 (Model KUMl) . . . . . . . . 105
Adoption of "Kuni Mbili" Stove/Fireplace in

Village 2 (Model KUM2) . . . . . . . . 107
Adoption of Kerosene Stove in Village 1

(Model KESl . . . . . . . . . . . 109
Adoption of Kerosene Stove in Village 2

(Model KESZ) . . . . . . . . . . . . 112

 

 

 

Adoption of Electricity in Village 2 . . . 114
Adoption of Tree Planting in the Preceding 10
Years in Village 1 (Model TPH1) . . . 116
Adoption of Tree Planting in the Preceding 10
Years in Village 2 (Model TPH2) . . . . 119
Combined Data Models . . . . . . . . . . . . 121
Discussion of Decision Variables . . . . . 123
Farm Area . . . . . . . . . . . . . . . 124
Financial Information . . . . . . . . 125
Household Characteristics . . . . . . . 126
Exposure to Information . . . . . . . . 126
Quantities of Fuel Used per Month . . . 127
Choice of Cooking and Heating
Technologies . . . . . . . . . . . 129
Household Fuel Choice . . . . . . . . . 132
Land Tenure Structures . . . . . . . . . 134
Predictive Ability of the Models . . . . . . 135
Overall Model Performance . . . . . . . . . . 139
Summary . . . . . . . . . . . . . . . . . . . . . 139

V. SUMMARY AND CONCLUSIONS . . . . . . . . . . . . . . 144
Policy Implications . . . . . . . . . . . . . . . 147
Recommendations for Future Research . . . . . . . 151

APPENDICES . . . . . . . . . . . . . . . . . . . . . . 156
1. Variable Description . . . . . . . . . . . . . 156
2. UCRHS Approval Letter . . . . . . . . . . . . 160
3. Domestic Energy Production and Consumption

Strategies Survey . . . . 161
4. Variables with High Correlation Coefficients

in Village 1. . . . . . . . . 177
5. Variables with High Correlation Coefficients

in Village 2 . . . . . . . . . . . . . . 178

6. Coefficients for Model KCJC . . . . . . . . . 179
7. Coefficients for Model KUMC . . . . . . . . . 180

iv

 

8. Coefficients for Model KESC . . . . . . . . . 181
9. Coefficients for Model TPHC . . . . . . . . . 182

REFMCES I O O O O O O O O O O O O O O O O O O O O O 1 8 3

 

 

LIST OF TABLES

Fuelwood and charcoal use by different sectors .

Research variables and their expected
relationship to adoption . . . . . . . . . . .

Comparison of mean values using the parametric

t-test . . .

Comparison of mean values using the U-test . . .

Adoption rates

(percentages) for policy

supported strategies in the two villages . . .
Number of households using different

combinations of strategies . . . . . . . . . .
Comparison of adopters and non-adopters,

Independent

t-teSt o e o e o o o e o o o o e 0

Comparison of adopters and non-adopters, Mann-
Whitney U-test . . . . . . . . . . . . . . . .
Average duration of technology use in years . . .

Attributes for
Probit results
Jiko (KCJl)
Probit results
Jiko (KCJZ)
Probit results

stove/fireplace

Probit results

stove/fireplace

Probit results

Stove (KESl) in

Probit results

different strategies . . . . . . .
for the adoption of Kenya Ceramic
in village 1. . . . . . . . . .
for the adoption of Kenya Ceramic
in Village 2. . . . . . .
for the adoption of "Kuni Mbili"

(KUMl) in Village 1 . . . .
the adoption of "Kuni Mbili"

(KUM2) in Village 2 . . . . .
the adaption of Kerosene
Village 1 . . . . . . .
the adoption of Kerosene

for

for

for

Stoves (KEsz) in Village 2 . . . . . . . . . .
Comparison of electricity user means against
Village 2 Means . . . . . .

Probit results

for the adoption of tree planting

(TPHl) in the preceding 10 years in Village 1

Probit results

for the adoption of tree planting

(TPH2) in the preceding 10 years in Village 2

Average prices

for different fuels and

strategies for Village 1 and Village 2 . . . .
Pearson's correlation coefficients for important

old and new

technologies used for cooking and

heating in Village 2 . . . . . . . . .
Pearson Correlation Coefficients for Quantities
of Fuels Used per Month in Village 1 . . . . .

vi

14
46

77
81

83
85
87
as
90
91
99

102

106

108

111

113

115

118

120

128

131

133

 

 

 

4.22 Pearson's correlation coefficients for

quantities of fuels used per month in
Village 2 . . . . .

O O O O O O O O O O O O O 133
4.23 Predictive capability of the models . . . . 136
4.24 Percentages correctly predicted for all models . 138
4.25 Summary of coefficient signs . . . . . . . . . . 140

vii

 

H F‘H
u 53H

UQU
UMP

LIST OF FIGURES

 

Agro-climatic zone map of Kenya . . . . . . . . 2
Projected 1989 per capita land availability for

Kenya . . . . . . . . . . . . . . . . . . . . . 4
Map of central Kenya showing the location of the

two villages . . . . . . . . . . . . . . . . . 18
Flowchart of the adoption process . . . . . . . 39

Flowchart of steps in the analysis process . . . . 43
Comparison of ordinary least squares estimate
and "typical" S-shaped relationship with a
dichotomous dependent variable . . . . . . . . 69

viii

 

I. INTRODUCTION

Was

This study addresses wood—energy related policies in
central Kenya. In this context, wood energy refers
collectively to fuelwood, charcoal, sawdust or any other
form in which wood is used for domestic cooking and heating
in the rural areas of Kenya. Fuelwood and charcoal are the
principal forms of energy used by most rural Kenyans. Here,
the wood-energy shortage has reached a crisis state. In
facing this crisis, energy choices must be made. Wood-
energy production and consumption choices include various
strategies that rural households take (1) to increase the
supply of wood energy and suitable substitutes and (2) to
ensure that available supply is efficiently used.
Individuals, the government and development agencies are

addressing the problem with various measures.

W

The wood-energy problem is severe in both high and low
potential agricultural regions of Kenya. Figure 1.1 shows
the agro-climatic zones of Kenya. The zones are based on a

Food and Agriculture Organization (FAO) study and are widely

 

 

2

Figure 1.1 Agro-climatic zone map of Kenya

Ecological potential 0! Kenya

    

E the" ”tom-a: am; (Ii (I)

g m “I (N)

unto-em ems (V)
D Ind you IV!)

Source: Juma and Namuye, 1987.

3
used as a system of land classification. The high potential
agricultural areas examined in this study are mainly in the
region including agro-climatic zones I, II, and III which
have an average annual rainfall of 800-1700 mm (Teel, 1984).
The typical natural vegetation in zone I is moist forest
with average annual rainfall of 1100-1700mm. Zone II has
average annual rainfall of 1000-1600mm: the typical natural
vegetation is moist to dry forest. Dry forest and moist
woodland are the typical natural vegetation in zone III
where the average annual rainfall is 800-1400 mm. The low
potential areas include marginal (semi-humid to semi-arid,
zone IV), semi-arid (zone V), arid (zone VI), and very arid
(zone VII) agro-climatic zones which are characterized by
low rainfall and high evapo-transpiration rates.

Low agricultural potential areas, which are more
ecologically delicate and susceptible to degradation, have
lower population densities than high potential areas. Most
of Kenya's rural population, however, live in the high
potential arable lands where there are many competing land
uses. These are the areas with critical fuelwood needs
(Bradley et_§1‘, 1985). Figure 1.2 shows the 1989 projected
land availability per capita for different districts in
Kenya (Odingo, 1988). Total district area and projected
1989 population were used to compute the values shown. Most
high potential areas had a per capita land availability of

less than ten hectares (1 hectare = 2.47 acres). Nyeri

 

 

4

Figure 1.2 Projected 1989 per capita land availability for
Kenya

ETHIOPI

I'IVWOS

l
l
I
I
I
l
I
l
: :1
I |"'|
I U
l C
| W
l r—
I -
I O
l
l
l
1

:0-9 locum Menu:
are one:

< I0 Huh Donny
Setumt n:-

 

Source: Odingo, 1988.

5
district had 4.5 hectares per capita. The low levels of
technological inputs such as fertilizers make maintaining
long-term land productivity difficult for these areas.

The Central Province of Kenya is representative of high
potential areas. It has 3.1 million of the country's total
1989 population of about 23 million (G.K., 1991). The two
study sites are located in Nyeri district which has a
population of 613,000 (ibid.). The other four districts in
the province (Murang'a, Nyandarua, Kirinyaga, and Kiambu)
hold the rest of the population. The arable areas of these
districts are relatively densely populated. Land holdings
are generally small and intensively used. Salem and Van Nao
(1981) noted that as population increases, agriculture
becomes more intensive and the amount of land left under
fallow at any one time is reduced considerably. This has
been the case in much of central Kenya. The population has
grown rapidly in the last generation or so, and land has

become increasingly scarce.

Heed:enersx.£resrsms

To address wood-energy needs, many programs were
initiated or strengthened in the country during the 1980s.
Funding for programs has involved investments by individual
farmers, the government, private volunteer organizations and
international development agencies.

Scherr (1989) noted the activities of some government

6
agencies involved in environmental conservation and energy
programs. In 1971, the government established the Rural
Afforestation and Extension Services (R.A.E.S.) program
under the Ministry of Environment and Natural Resources.
Its mandate was to promote tree planting outside the
national forest reserves by farmers, organizations, groups,
and institutions. The RAES program has evolved into the
Forest Extension Services Division (FESD) established in
1989 and given a broad mandate. The Division encourages the
different groups mentioned above to grow trees for firewood,
poles, fodder, fruit, shelterbelts, and aesthetics among
other reasons. The Rural Tree Development Support Project
(RTDSP) under the FESD supports the development of woodfuel
production, inter-agency cooperation, and the dissemination
of wood energy development and use packages that can help
farmers become self-sufficient in wood energy (FESD
Newsletter, 1990). Other programs are also included.

The Ministries of Agriculture and Livestock have been
promoting tree planting for soil conservation and fodder
production. From its inception in 1980, the Ministry of
Energy had a specific mandate to promote tree planting for
fuelwood and to encourage the efficient utilization of all
forms of energy. The Ministry has used an "agroforestry
centers" model to establish a network of centers which
promote tree planting and efficient energy use throughout

the country. The six major centers include Wambugu Farmers

 

7
Training Center in the central highlands (the region where
this study was conducted), Mtwapa Center in the coastal
region, Bukura Center in the western highlands, Ngong Center
near Nairobi, Kitui Center for the arid/semi-arid regions,
and the Kisii Center for the Lake Victoria basin (MOERD,
n.d.).

S 'E' E J' .
Several policy alternatives for dealing with the wood-
energy problem in the rural areas of Kenya have been
proposed or implemented. They include increasing wood-
energy supplies, promoting alternatives to wood energy, and
encouraging efficient energy use. Another option is
switching to faster cooking foods. This option has,
however, not been given much emphasis by the agencies
reviewed in this study. In many cases, energy alternatives
are promoted by the government. In others, farmers have
recognized their needs and embarked on their own programs.
Specific policies that have been proposed or
implemented include the following:
(1) Planting more trees on family farms for fuelwood and
charcoal (G.K., 1989),
(2) Expanding the land area reserved for government
afforestation programs (G.K., 1989),
(3) Providing a wood-energy conservation scheme based on

reducing the effective price of kerosene and

 

 

 

8

encouraging its use as an energy alternative (G.K.,

1989),

(4) Developing and promoting the use of more efficient
stoves especially the Kenya Ceramic Jiko, the "Kuni
Mbili" stove/fireplace, and sawdust stoves (Joseph and
Kinyanjui, 1982: Jones, 1988), and

(5) Accelerating rural electrification so that more
families can use electricity for cooking, heating and
lighting (G.K., 1986).

The first strategy, tree planting, has received the
greatest policy and resource allocation emphasis so far.
This alternative assumes that owners of private land can
increase the amount of biomass available for energy purposes
through adopting various types of agroforestry practices or
by establishing farm woodlots. Critical issues in promoting
tree planting include the following: the specific wood
products desired, where trees should be planted, the species
that should be planted, the source(s) of seedlings, the
target group(s) for planting programs, tree tending and
harvesting, and finally, the distribution of benefits from
trees among members of the same household who may have
different interests in the trees. A good understanding of
these issues is essential to the effective promotion of tree
planting by small-scale farmers.

Tree growing is one possible use of land. Benefits

from tree growing include cash, labor savings, environmental

9
protection, increased availability of consumable wood
products, or even increased leisure time by cutting down on
the amount of time needed to collect fuelwood, for example.
For some farmers, particularly the poor ones, growing trees

that can be sold as fuelwood or other products in times of

 

emergency cash needs, such as school fees for the children
or medical expenses, can be an important consideration.

Due to the existing population pressures in the high
potential agricultural areas, it is unlikely that any
private land will ever be turned over to the government for
tree planting projects as proposed in the second strategy.
This option would probably work only in areas that have low
population or low arable potential. In central Kenya, such
land is limited. Private land owners can, however, increase
the amount of land under tree management through
agroforestry or planting woodlots. Also, the existing
public forest lands can be managed to provide more of the
fuelwood needs of rural communities. In fact, some villages
are becoming more dependent on public forests. In the
Island Farms sublocation (hereafter referred to as Village
1), farmers initially settled on fairly well-wooded land
parcels under the Million Acre resettlement scheme after
Kenya got independence from Britain in 1963. Landless
Kenyans were given low-interest, long-term loans to acquire
land that was formerly held by British settlers. Over the

years, farmers have cut down most of the original trees on

 

10
their land and now rely heavily on national forest lands for
fuelwood and charcoal.

The third strategy, use of kerosene as a wood-energy
substitute, was proposed in the 1989-93 National Development
Plan. Generally, only a small proportion of rural
households use or are likely to use kerosene for their daily
cooking and heating. In Kenya, profit margins from the
retail sale of kerosene have historically been relatively
small. In 1983 for example, the ratio of retail to border
price was 1.07 (Kosmo, 1989). This meant that kerosene at
retail outlets was only 7 percent more expensive than the
price at the port of entry. Retailers were therefore making
a very small profit. Large subsidies may be needed to
ensure a shift from the consumption of fuelwood to kerosene,
and this may lead to high foreign exchange costs since Kenya
is an oil-importing country. Results for similar policies
have not been good in countries like Zambia and Indonesia
where they have been tried (Foley, 1986: Mercer and Soussan,
1992).

An early effort to improve efficiency and convenience
in cooking and heating was the introduction of the
traditional metal stove in the early 1900's: it has been
adopted widely (Juma and Namuye, 1987). This is partially
due to the simplicity of the technology. It can be made by
local blacksmiths using scrap sheet metal and very few

tools. Unfortunately, this charcoal-burning stove loses a

 

 

11
lot of heat and wastes a lot of fuel.

Efficiency improvements in wood and charcoal use are
important policy considerations. Macklin (1984) noted that
the complete dissemination of improved stoves could save up
to 50 per cent of the firewood and charcoal being used at
the time of his study. There would also be additional
savings in terms of the time used in fuelwood collection.
These estimates are based on the unrealistic assumption of
complete adoption. The data from the study areas indicate
that adoption levels for many of the technologies are still
low. This emphasizes the importance of an adoption study.

Traditional pit kilns are used in the production of
most of the charcoal in Kenya. These kilns are inefficient
and have conversion efficiencies of 9 to 14 percent (ibid.).
More efficient charcoal kilns have conversion efficiencies
of up to 30 percent. Shah et_alL (1992) found that
commercial partial-combustion kilns had carbonization
efficiencies of between 26 and 38 percent on a dry wood
basis. Such commercial kilns, however, require higher
capital outlay and large quantities of output to be viable
investments. When conversion losses are added to losses due
to inefficient stoves, it becomes apparent that total heat
loss can be enormous. More research is needed to establish
how better charcoal-making technology can be widely adopted.

More efficient stoves such as the Kenya Ceramic Jiko could

12
make an even greater impact if appropriate charcoal-making
technology were also adopted.

The Kenya Ceramic Jiko (KCJ) is an improved charcoal
stove that reduces energy losses by using a ceramic lining
for insulation. Hyman (1986) reported that households could
save 25 percent of their charcoal expenditures by switching
to the improved stove. This stove is a recent technology in
Kenya: it was developed in the early 1980s. Juma and Namuye
(1987) noted that even though the adoption of the stove
appears to be increasing, there is no reliable information
on its production and use. In addition, they also noted
that the profile of the early adopters of the stove was
unclear, and that the impact of such technologies on forest
and tree resource depletion was also not well understood.

The "Kuni Mbili" stove or fireplace is a newer
technology designed to replace the three-stone fireplace.
The traditional three-stone fire place consists of three
stones placed near each other in a triangular design so that
a cooking pot can be placed on the stones. The fuelwood is
set on fire directly under the cooking pot. The "Kuni
Mbili" stove/fireplace prevents heat loss from burning
fuelwood by using a ceramic lining similar to the one used
in the KCJ. The fireplace is designed and fitted in a fixed
position in the kitchen while the stove is a portable unit.
This is one of the important strategies considered in this

study.

 

 

13

The final strategy, rapid rural electrification, is an
important component of Kenya's overall energy program and
has been funded through government investments and subsidies
to the Kenya Power and Lighting Company (Anonymous, 1989).
Nearly all electricity used in Kenya is produced by
hydroelectric power plants. Between 1979 and 1989, the
output of hydroelectric power in Kenya increased by 88 per
cent (World Resources Institute, 1992). One major
geothermal power station (Olkaria in the Rift Valley
Province) is also in use. The government projects
electricity demand to grow by 6.2 percent per year in the
1990s (G.K., 1989). The government expects that "wider
distribution of electricity, coupled with rising incomes,
will reduce the demands for kerosene for lighting and for
fuelwood" (G.K., 1986). One facet of this study examines
whether, and if so, why, some households have shifted away
from using wood energy to using electricity for cooking and
heating. Electricity was available only in Kihugiru sub-
location (hereafter referred to as Village 2), which became
a beneficiary of the rural electrification program in the

mid-19703.

W
Fuelwood and charcoal are the main sources of domestic
energy for cooking and heating for most Kenyan households.

Table 1.01 shows 1983 fuelwood and charcoal use levels by

 

 

l4
sectors (MOERD, 1986). At that time, between 75-80 percent
of the national energy needs were met by fuelwood and
charcoal (ibid.). Rural households were using 74 percent of
the fuelwood and 36 percent of the charcoal. Urban
households, on the other hand, used only 1 percent of the
fuelwood and 45 percent of the charcoal. The trends are
unlikely to have changed dramatically during the last eight

years. Industries and institutions used much lower

Table 1.01 Fuelwood and charcoal use by different sectors

 

Percent of National Consumption

 

 

Sector Fuelwood Charcoal
Rural households 74 36
Urban household 1 45
Informal industry 15 17
Large industry 8

Institutions

 

Source: MOERD, 1986.

quantities of fuelwood than rural households.

In urban areas where 19 per cent of the Kenyan
population lives (G.K., 1991), household energy-use
strategies are quite different from those in rural areas.
Kimuyu (1990) evaluated the effect of urbanization and
changes in the structure of the Kenyan economy between 1963-
85 and reported that there was a significant effect on the

demand for many fuels. He concluded that "spatial

 

15
population distribution and chosen development strategies
create energy requirements of specific magnitudes and
structure." An earlier study by the Government of Kenya
(G.K.) also showed that urban and rural households have
different energy-use patterns (G.K., 1980). In general,
urban households are more dependent on commercial fuels such
as kerosene, gas, and electricity whereas rural households
are more dependent on fuelwood and charcoal.

It is estimated that 46 percent of the fuelwood used in
Kenyan households comes from family farms, 28 percent from
national forests, and 25 percent from range lands (G.K.,
1989). These proportions may vary considerably at regional
and sub-regional levels. Baker (1986) noted that national
forest reserves covered 3.5 percent of the national
territory in 1970 and had declined to 2.5 percent in 1980.
Due to the limited area of national forest lands, fuelwood
will increasingly have to be produced on family farms.

Households appear to have paid little attention to some
of the government supported programs stated above. Although
one objective of the rural electrification program is to
supply energy for domestic use for example, it is does not
appear to have substituted for fuelwood in any significant
amount in Village 2. In fact, only 1 percent of the

households surveyed in Village 2 reported that they used

 

16
electricity for cooking and heating. Factors associated
with the adoption of various energy strategies need to be

critically examined.

E l :1. !'
In Kenya, the wood-energy production and consumption

strategies arising from different government policies

include planting more trees on family farms, using more
kerosene as a fuelwood substitute, using more efficient
fuelwood and charcoal stoves (the KCJ, and the "Kuni Mbili"
stove/fireplace), and rural electrification. The specific
objectives of this study are:

(1) to determine how study households have responded to
specific wood-energy policies which were put in place
or strengthened during the 1980s, and

(2) to identify factors that are associated with household
adoption or non-adoption of the strategies.

Results of this research will provide a basis for

recommendations on suitable implementation approaches for

wood-energy policies.

Current literature on fuelwood supply and demand
situations in Kenya and other developing countries supports
the need for adoption studies such as the current one (e.g.
Bradley et_a11. 1985: FAO, 1989: FAO, 1990). Observed
wood energy needs of study area households also appear to be

relatively high. This research has a potential contribution

 

17
in answering questions related to better provision of energy
to rural households. In addition, wood energy shortages may
have adverse nutrition impacts on the household members
especially in situations where certain foods may be avoided

due to high energy requirements for their cooking.

5 i I' '! !'

This research involves a case study of two villages in
the highlands of central Kenya. The general location of the
villages is shown in Figure 1.3. The rationale for
selecting these two villages is explained in Chapter 3.
Small-scale agriculture, which incorporates food crops, cash
crops and livestock, is the main activity in both villages.
Village 1 relies more on horticultural and dairy products as
sources of income while Village 2 relies on tea and coffee.

The area covered by this study represents a small
proportion of the otherwise large region of central Kenya.
The ethnic composition of the two villages in the region is
fairly homogeneous. There are, however, local geographic
and economic variations. Caution must therefore be
exercised in extending case study results to other regions.
The current study is based on a survey of 306 farmers. A
comprehensive study of the whole region would require a much
larger sample and more resources.

As a one-time cross-sectional study, the results

generally reflect the situation in the two villages at the

 

18

Figure 1.3 Map of central Kenya showing the location of the

two villages

1 \
'2 m
I”
m 3 ‘

. “In
'30 ’n

u 5 ‘_ Y

’ It}, 1n _' .
-— . m m

1 n »
no --
‘_ m

f
m 1:

Source 3

3" US

0
”S

l «a r pmcm

t3...
Nai I'uhi

Survey of Kenya, n.d.

db

5 CENTRAL

7.1111 KHHA
£1» ,
/\/,C§1f;_,V1llage 1

“Village 2

  
   
    
 

PROV! NFF.

Q20

EASTERN
PROVINCE

19
time of the study (July-September 1991) . Thus, it is
'difficult to explain some village differences which may be
due to historical reasons. In the mid-1970s, for example,
world coffee prices were very good and many farmers in
Village 2 were able to build good permanent homes and to
support the development of the rural infrastructure. At the
time of the survey, however, coffee prices were very low,
and this tended to give the impression that reported incomes
were not very significant in explaining residents' higher
standard of living relative to Village 1. To the extent
possible, relevant anecdotal historical information is
included in the discussion of the situations in these
villages.

Probit analysis was used to examine the factors
associated with adoption or non-adoption of different wood-
energy production and consumption strategies. For each of
the specified wood-energy production and consumption
strategies, a binary probit model was developed. Since the
two villages were statistically different for many
independent variables, probit models were generated
independently. Data for the two villages were also combined
and used to generate probit models for the enlarged sample.
The assumption here was that the two villages reflected two
strata from different socio-economic and bio-physical
conditions within central Kenya. Dummy variables were used

to test for village effect in the combined models.

20
E' ! !' ; . !'0

This dissertation is organized into five chapters.
Chapter 1 is the introduction. Chapter 2 reviews different
wood energy production and consumption policies and relevant
studies that have been carried out in Kenya and elsewhere.
The importance of wood energy to rural households and the
constraints associated with obtaining adequate supplies is
also discussed. The roles of different government and non-
governmental agencies are examined in the context of the
adoption or non-adoption of different strategies.

Chapter 3 presents the conceptual and analytical
methods used in the study. A theoretical justification is
presented for selecting probit analysis over the general
linear model, the linear probability model, the logit model,
or the Tobit model. Rationales are also presented for
selecting the two villages, the sample sizes, and the survey
procedure. Parametric and non-parametric tests used for
comparative analysis are also explained.

Chapter 4 presents a discussion of results. They
include parametric and non—parametric test results,
correlation analysis, and probit models generated for the
various strategies in the two villages. A discussion is
also included of the goodness of fit and the significance of
the models in explaining factors associated with adoption or
non-adoption of wood-energy production and consumption

strategies.

21
Finally, policy implications of this study are
discussed in Chapter 5. Recommendations for future research

are also presented.

II. LITERATURE REVIEW

Literature related to wood-energy production and
consumption is discussed in this chapter along with
technologies that increase the efficiency of fuelwood and
charcoal. Possible alternatives to fuelwood and charcoal
are also discussed. Literature on the diffusion of
innovations and analytical techniques used is discussed in
Chapter 3. Relative to Chapter 1, specific aspects include
an expanded discussion of research and development work
carried out by various agencies, implications of wood-energy
shortages, different approaches to the problem, and policy

and household responses.

B l l E J ! H l l M . E .
Following the oil crisis of the early 1970s and the
Sahelian drought in the same period, many African countries
were forced to re-evaluate their energy policies as well as
the delicate nature of their environments. In Kenya, a
rather slow process of evaluating household and industrial

energy needs was initiated. There were few studies on the

22

23
wood-energy problem in the 1970s. Additional studies on the
fuelwood problem in Kenya, however, appeared throughout the
19808.

Organizations involved in planning and executing wood-
energy programs include different government ministries and
agencies and non-governmental organizations. The major
government organizations include the following: Ministry of
Energy and Regional Development (MOERD), Ministry of
Environment and Natural Resources (MENR), Ministry of
Agriculture (MOA), Ministry of Livestock Development (MOLD),
Kenya Forestry Research Institute (KEFRI), Kenya
Agricultural Research Institute (KARI), Permanent
Presidential Commission on Soil Conservation and
Afforestation (PPCSCA), and national universities. Non-
governmental organizations include the Green-Belt Movement,
CARE-Kenya, and Kenya Energy Non-Governmental Organizations
(KENGO), among others. Most of these organizations have
primarily pursued an agroforestry (production) approach to
the fuelwood problem. Agroforestry addresses the supply
side of the fuelwood shortage. The PPCSCA, KARI, KEFRI,
KENGO, and CARE-Kenya all participate in programs that
include multi-purpose tree planting. The Ministry of Energy
and KENGO have also been actively involved with demand-side
policies. Kenyatta University has a stove testing unit at

its appropriate technology center while Moi and Egerton

24
Universities are involved in agroforestry teaching and
research.

Agroforestry has emerged as a popular strategy of
dealing with wood-energy deficits among farmers. The
International Council for Research in Agroforestry (ICRAF)
based in Kenya is an international agricultural research
center. It has several field research sites in Kenya.
ICRAF (1990) quotes Paul Harrison (author of Ihe_§reening_gf
Africa) and agrees with his argument that: "Agroforestry is
not only the most promising approach to reforestation and
the supply_gf_fuelwggd, it is also, in yield-boosting forms
like windbreaks and alley cropping, the most hopeful avenue
for intensifying African agriculture over the next five to
ten years, increasing food production and reducing exposure
to drought with few or no outside or imported inputs.
Agroforestry is arguably the single most important
discipline for the future of sustainable development in
Africa." Some reports suggest that agroforestry for
fuelwood has not been very successful since private farmers
place a higher priority on higher-value products such as
building poles and pulpwood (Mercer and Soussan, 1992).
Organizations promoting agroforestry need to appreciate its
potentials and limitations.

The International Council for Research in Agroforestry
was founded in 1977 and opened its headquarters in Nairobi,

Kenya in 1978 (ICRAF, 1990). The primary objective of ICRAF

25
was to support agroforestry research and disseminate
agroforestry information among developing countries. In the
mid-19805, the objective was reviewed and more emphasis is
now given to generating appropriate agroforestry
technologies and strengthening national research capacity.
In fulfilling its objectives, ICRAF hopes to increase the
social, economic, and nutritional well-being of people in
developing countries. In their combined efforts to increase
the growing of multi-purpose tree species by farmers, all of
these agencies are performing useful roles.

In 1981, the United Nations Conference on New and
Renewable Sources of Energy was held in Nairobi, Kenya. The
consensus at the conference and in subsequent policy
discussions was that most developing countries would
continue to rely heavily on wood for their domestic energy
needs in the foreseeable future (Engelhard, 1992).

Countries such as Kenya therefore needed to come up with
appropriate institutions for formulating and implementing
appropriate policies for new and renewable sources of
energy. The new sources of energy included technologies
such as methanol (a gasoline substitute) which had not been
widely used in the past. Renewable sources of energy, on
the other hand, included energy sources whose stock could be
regenerated. Examples included biomass (fuelwood, charcoal,
crop residues etc.), biogas, wind, and solar power. The new

Ministry of Energy in Kenya was given the responsibility of

 

26
exploring the potential of all these technologies.

The pursuit of new and renewable sources of energy has
generally produced mixed results. Foley (1992) noted that
many of the new and renewable sources of energy programs
advocated and supported during the 19805 failed for various
reasons. Failed programs included solar energy, small
hydroelectric projects, biomass gasification, biogas, and
ethanol production. Failure was mainly due to technical and
maintenance problems, high costs, and unjustified pessimism
about oil prices. He suggested that there is a need for
greater rigor in economic and technical analysis. Wood-
based energy technologies are low cost compared to many of
the above technologies, and they are likely to succeed where
many other technologies have failed.

Wood-based energy technologies have been actively
supported by the Ministry of Energy and Regional
Development. The Ministry has also provided a coordinating
role for interested governmental and non-governmental
organizations. Its role and that of the Ministry of
Environment and Natural Resources were discussed in Chapter
1.

The Ministry of Energy and Regional Development came up
with an agroforestry-based institutional development
strategy in 1986 (MOERD, 1986). In 1980, it had requested
the Beijer Institute of the Royal Swedish Academy of

Sciences to perform a woodfuel cycle study to obtain some

27
empirical data on the magnitude of the woodfuel crisis and
to make suitable recommendations. The Institute recommended
increased support of agroforestry programs, development and
dissemination of efficient wood-burning and charcoal-burning
technology, and development of new and renewable forms of
energy (MOERD, 1986). Wood-energy production and
consumption programs such as the Kenya Renewable Energy
Development Project (KREDP), the Kenya Woodfuel Development
Project (KWDP), and the Special Energy Program (SEP/GTZ)
were direct results of the initial research studies and

subsequent follow up.

On a national basis, wood-based fuels represent 75 per
cent of the country’s energy resource base. Among
households, 95 per cent of the energy needs are met from
woodfuels (Macklin, 1984): fuelwood and charcoal are the
major woodfuels. Wood-energy shortages will create serious
consequences (including nutritional problems) as rural
populations grow.

Population growth rates have remained high in Kenya
although they appear to be slowly declining (G.K., 1991).
The estimated population growth rate between 1979 and 1989
was 3.34 percent. These high population growth rates may
lead to greater pressure on wood energy stocks in many of

the high potential areas.

 

28

Intensive agricultural activities and rapid population
growth have greatly increased pressures on the land.
Macklin (1984) reported that most of the high potential
areas of the country tend to be net importers of woodfuels
from surrounding areas. Many of those areas are
ecologically delicate, and the environmental consequences of
harvesting can be severe. Such consequences include
deforestation, soil degradation, erosion, pollution of
rivers, siltation of water reservoirs, and loss of economic
opportunities for many communities.

In parts of Africa, lack of fuelwood has a very direct
impact on human nutrition (Poulsen, 1978). Most staple
foods such as maize, millet, sorghum, cassava and yams
require prolonged cooking. The high costs of domestic
energy may therefore affect the health of rural populations.
In addition to the supply of more wood energy and
substitutes, more research on crops that require less

cooking time would be appropriate.

Th§_$§§19h_£QI_San§iQnS
In an overview of woodfuel supply and demand in Kenya,
Buck (1980) summarized some important socioeconomic
variables to be considered in research studies. The
variables included land tenure impacts, resettlement
programs, resistance to change, affordable energy

alternatives, and different socioeconomic characteristics of

 

 

 

29
households. These factors among others are examined in this
strategy adoption study. Empirical evidence to explain the
determinants of wood-energy consumption and production
adoption among small-scale farm households is needed.

In the early 19803 Skutsch (1983) performed an
evaluative study of village afforestation programs in
Tanzania focusing on social and economic motivations behind
successful and unsuccessful afforestation programs. Her
principal findings were that: (1) people with acute'
firewood needs were keenly interested in establishing and
maintaining woodlots, (2) people in the rural areas had
substantial knowledge about tree planting and management,
(3) extension services were not very effective in following
up the work on trees planted, (4) transportation of
seedlings to the planting sites in the planting seasons was
often inadequate, and (5) that communal forestry projects
needed to be more efficiently organized. The study did not
include a sufficient analysis of economic factors that may
motivate tree planting. Moreover, the socioeconomic
conditions on communal farms in Tanzania were significantly
different from any that one may observe in the high
potential agricultural areas of central Kenya. The study
nevertheless provided some general indicators on why people
do or do not plant trees in light of official policies that
encourage this activity.

Allen and Barnes (1985) performed a cross-national

30
analysis of deforestation for 39 countries in Africa, Asia
and Latin America for the period 1968-1978. They found that
the past rate of wood production (which included tree
cutting for fuelwood) was an important variable in
understanding the rate of deforestation. The authors
reported that the rate of deforestation in Kenya was
estimated at 2 percent per year between 1976 and 1980.
Excessive cutting of trees for fuelwood thus poses a serious
environmental problem. French (1988) had a pessimistic View
of deforestation in sub-Saharan Africa and argued that it
can only be slowed down and cannot be completely halted.

Magrath (1984) reviewed some studies on the
microeconomic aspects of agroforestry. He concluded that
microeconomic studies of agroforestry in developing
countries are very few and much more work needed to be done.
Foley (1986) agreed with this view and argued that the
economic analysis of the wood products consumer behavior is
very important if the potential impacts of proposed policies
were to be well understood. Research in this area is still
inadequate and much more remains to be done.

Many individuals and organizations look at the wood
energy problem from the physical supply side only. Bradley
(1988) cautioned that domestic energy "development programs
that stop at the mere provision of yet more trees fail to
genuinely address the issue of woodfuel energy." The

fuelwood problem is complex and requires a multifaceted and

 

31
multidisciplinary approach at both the national and local
levels. Mercer and Soussan (1992) concurred and noted that
another supply-side policy has been the manipulation of
fuelwood prices. Pricing policies seek to influence
fuelwood supply by applying royalties and stumpage fees, as
well as fuelwood taxes and subsidies. Fuelwood- and
charcoal-pricing policies do not appear to have been widely
used in Kenya. In the case of Village 1, fuelwood from
national forests appeared to be heavily subsidized compared
to fuelwood prices in transactions between farmers.

Between 1983 and 1986 the Kenya Woodfuel Development
Project carried out studies on three high agricultural
potential districts, namely, Kisii, Muranga and Kakamega
(see Figure 1.3). The studies concluded that integration of
wood production on the farm was complex and deeply rooted,
that as population density increased, the quantity of
managed biomass increased, and finally that the amount of
land devoted to tree production appeared to increase as
population density increased (Bradley, 1988). These
somewhat surprising conclusions indicate that farmers do
indeed respond to wood scarcity and higher prices associated
with it. The response mechanism, however, is not well
understood.

Fuelwood has become a market good in many of the high
agricultural potential areas of Kenya (Ngugi, 1988).

Ngugi’s study was an attempt to explain how cultural factors

32
have influenced the fuelwood shortage. The conclusion of
the study was that the shortage was due to economic and
social factors such as limited areas of farm land and
increasing population. There is, however, no statistical
analysis given to indicate the significance of different
social or economic factors. In the Baringo Fuel and Fodder
Project (BFFP) located in the semi-arid region of northern
Kenya, past project failures have been attributed to the
failure to consider local needs which are crucial in the
adoption of any new practices (Hall et_alL, 1992). In the
dissemination of technologies that address wood energy
production and consumption, local needs and preferences must
be taken into account.

Bhatia (1987) noted that reported wood energy
consumption in areas with deficits tends to conceal the
magnitude of the underlying shortages. Such shortages may
be associated with insufficient local production,
distributional constraints, low incomes, or government
policies that are in place. Data on fuelwood production and
consumption are also often poor for many developing
countries (Onyebuchi, 1989). Kenya faces many of the above
constraints and some of them are examined in this study.

In a study carried out in the Llocos Norte region of
the Philippines, selected policy options related to fuelwood
production and use were evaluated (Dixon, 1986). The study

evaluated a government-sponsored tree planting project, the

 

33
introduction of an improved wood stove, and the substitution
of a traditional wood stove with a kerosene stove. The
research was based on primary data collected through
extensive surveys. A benefit-cost analysis framework was
used to compare the policy options. The study explained
some useful policy analysis methods that can be used in
studying the fuelwood problem in Kenya.

Scherr and Muller (1989) noted that farmer surveys are
an important method of assessing the adoption of
agroforestry technologies. They also indicated that many
unanswered socioeconomic questions on these technologies
still remained. Avila (1990) concurred and also argued that
one of the goals of on—farm research should be "to assess
the impact of agroforestry technologies and to monitor and
evaluate the way farmers adopt and modify them." Schultz
(1964) argued that traditional farmers, though poor, are
efficient in the allocation of available resources. If that
argument is accepted, it can then be argued that the
adoption of wood energy production and consumption
strategies will be consistent with the resource endowment of
the farmers other things being equal. The poorest may be
expected to adopt only the most basic technology package if
any at all, while wealthier ones can be expected to consider
adoption of more advanced packages since they can bear

greater risks. The larger socio-economic and bio-physical

34
context within which the farmer operates is also important

and is discussed in Chapter 3.

SBEEQIX

Fuelwood policy analysis is crucial for Kenya and other
developing countries that are still heavily dependent on
wood energy. As already mentioned, supply-side and demand-
side policies must be comprehensively addressed. Supply-
side policies include tree planting (e.g. large-scale
plantations, social forestry, agroforestry, and natural
forest management) and fuelwood pricing. Demand-side
policies on the other hand include increasing the efficiency
of wood energy use (e.g. fuel-efficient stoves, improved
charcoal kilns) and the substitution of wood energy with
modern fuels.

Mercer and Soussan (1992) noted some of the research
gaps in the above policy issues. This study is designed to
answer some pertinent questions related to the adoption of
tree planting, fuel-efficient stoves, and wood energy
substitutes. Other research concerns they noted include
questions related to different fuels, lack of adequate data
on demand and supply elasticities, foreign exchange needs,
kerosene pricing policies, poor access or infrastructure in
rural areas, and higher costs of the more efficient

technologies.

35

This chapter has reviewed different research and
development work carried out by various agencies as well as
research that illustrates the importance of a study on the
factors associated with adoption or non-adoption of wood-
energy production and consumption strategies. Many studies
indicate that improvements in the use of wood-energy
production and consumption technologies offered to rural
households in the 19805 can still be made. Swinkels and
Scherr (1991) reviewed additional agroforestry and wood-

energy related studies.

III. RESEARCH METHODS

In this chapter, the conceptual and analytical
frameworks used in this study are discussed along with
related literature. Sampling design, sample selection,
parametric and non-parametric tests, the econometric model
selection process, and details on probit analysis are
presented. This information provides a theoretical
background against which the determinants of adoption or

non-adoption are evaluated.

W

This study is based on the diffusion of innovations and
utility maximization theories. Schultz (1964) described the
small-scale farmer as a rational economic agent who attempts
to maximize his or her utility in consumption subject to the
resource constraints that he or she faces.

In positive economic analysis, a utility function is
ordinal and can be replaced by any increasing transformation
of itself (Layard and Walters, 1978). This implies that
strategies or consumption bundles providing higher levels of
satisfaction will be preferred to those providing less.

This explanation of consumption is based on assumptions

36

 

37
that: the decision maker is rational, he/she faces various
constraints, and the individual's past experiences and
societal preferences will influence his/her individual
choice (Randall, 1987). Utility theory helps us to
systematically and conceptually understand how individuals
prefer one item over another. The utility derived from the
adoption of a particular strategy will be expressed as a
function of the characteristics of the strategy as perceived
by the individual, and the socioeconomic characteristics of
the individual.

Pharo (1982) evaluated the substitution potential of
wood energy for other fuels among households and firms in
the American economy. Firms seek to minimize costs and for
any two fuels, cost is minimized when the marginal rate of
substitution between them is equal to their price ratio.
Utility theory was an underlying assumption in his analysis.
The main objectives of the research were to: (1) evaluate
cost efficiency of woodfuels before and after the Arab oil
embargo, (2) evaluate the most efficient situations for
woodfuels, and (3) make comparisons between costs and
savings. Time series data of energy and prices over a
fourteen-year period (1967—1980) were used. Only 21 per
cent of the annual woodfuel supply potential was found to be
economical for exploitation. Woodfuels were found to be
marginally better substitutes for oil only after labor costs

were disregarded. If labor costs were included, natural gas

38
and coal appeared to be better substitutes after the oil
crisis.

Peacock (1972) examined farmer adoption of new
agricultural practices in Northeast Brazil using diffusion
of innovation theory. He examined how different
socioeconomic and personality variables affected the
adoption of new agricultural practices. In this current
study, both economic and non-economic variables are also
considered in evaluating the determinants of adoption or
non-adoption of different wood energy production and
consumption strategies. Figure 3.1 illustrates the
hypothesized household, environmental, and choice
characteristics that influence adoption or non-adoption; it
provides the conceptual model used in this study.

The attributes of the socio-economic environment
include markets, infrastructure, research, extension, and
land tenure. Household attributes include land, labor,
capital, and farmers' goals. Examples of attributes of
strategies include cost, efficiency, safety, and ease of
use. The knowledge base (experiences) for old and new
technologies is also important in deciding whether or not to
adopt. Households are assumed to be utility maximizers in
consumption and profit maximizers in production. The
observed outcome is adaption or non-adoption of specific

strategies. Variable relationships in the adoption process

 

ATTRIBUTES OF
THE SOCIO-
ECONOMIC
ENVIRONMENT

 

39

 

 

ATTRIBUTES OF
HOUSEHOLDS

 

 

 

 

 

 

 

ATTRIBUTES OF
DIFFERENT
STRATEGIES

 

 

 

 

EXPECTATIONS
FOR NEW
STRATEGIES

 

 

 

 

 

 

 

Figure 3.1 Flowchart of the adoption process

PAST EXPERIENCE
WITH NEW
STRATEGIES

 

 

 

 

 

DECISION
MAKING

 

 

 

 

 

OUTCOME
ADOPTION OR
NON-ADOPTION

 

 

 

40
are dynamic and the temporal dimension of the adoption
process is important.

Rogers (1983) described the theory on the diffusion of
innovations in a comprehensive way. Innovativeness was
described as the "degree to which an individual is rela-
tively earlier in adopting new ideas than other members of
his social system." Adoption or acceptance of various
innovations was seen as emanating from interpersonal
interactions at various social levels. The three major
challenges in classifying adopters were noted as follows:
(1) the number of adopter categories to conceptualize, (2)
the proportion of members of society to be included in each
category, and (3) the definition of the categories. Adopter
categories need to be exhaustive, mutually exclusive, and
based on one classificatory principle. Means and standard
deviations have been used in developing what are often
referred to as the ideal adopter categories (ibid.). The
five ideal categories are: innovators, early adopters,
early majority, late majority, and laggards.

The characteristics of an adopter of a particular
strategy include socioeconomic, personality and
communication behavior variables. Socioeconomic variables
include age, education level, social status, resource
ownership, and economic orientation. Personality variables
include among others attitudes towards change, attitude

towards education, fatalism and aspirations. Communication

41
behavior variables include social participation in
activities such as meetings and agricultural shows, contacts
with change agents, exposure to the mass media, knowledge of
innovations, and the degree of opinion leadership. These
variables are included in the probit model specifications
and are examined for household data in this study area.

Voh (1982) carried out a study on the adoption of
recommended farm practices in a Nigerian village using
stepwise regression analysis to calculate an adoption index.
He obtained the following parameter signs for the variables
studied: leadership role (positive), level of formal
education (positive), level of literacy (positive), number
of farm sites (negative), household size (negative), urban
contact (positive), contacts with extension agents and
various sources of advice (positive), empathy for others
(positive), and involvement in non-farm occupations
(positive). The statistically significant factors were
leadership role, level of literacy, extension contacts,
empathy, level of education, urban contact, additional
occupation, and socioeconomic status, in that order. The
model explained 32 per cent of the variability in adoption
behavior. The first four variables explained 31 per cent of
the variability. Voh concluded that it was important to

investigate what the other significant variables might be.

42

This study uses probit models to estimate adoption and
non-adoption of household energy strategies. Probit models
are based on probability units (Aldrich and Cnudde, 1975).
The model selection process is explained further in the
section on econometric models. Though model formulation
occurs late in the analysis process (Figure 3.2), it affects
the nature of questions in the survey and statistical tests.
The survey conducted was used to obtain data on strategies
and household wood-energy production and consumption
characteristics.

Data relevant to the conceptual adoption process
displayed in Figure 3.1 were collected. These data were
then used for various statistical tests (e.g., mean
differences and correlational analysis), model building and
policy analysis.

The general specification of the probit model is as
follows: I
Y, = be + b,x1 + b2x, + ....+ bnxn + e1
where:
¥,==1.if a household has adopted strategy i (i=1,2,3,4),
and 0 otherwise. i = 1 for adoption or non-adoption of
fuelwood tree planting, i = 2 for adoption or non-adoption
of the "Kuni Mbili" stove or fireplace, 1 = 3 for adoption
or non-adoption of the Kenya Ceramic Jiko (Stove), and i = 4

for adoption or non-adoption of kerosene as a fuelwood

 

SURVEY

 

43

 

 

 

 

l

t

I

 

FUEL AND
SOCIO-ECONOMIC
CHARACTERISTICS

 

 

ENERGY CONSUMPTION
AND PRODUCTION
CHARACTERISTICS

 

I
1

 

 

 

 

 

MEAN DIFFERENCES
FOR VILLAGES
1 AND 2

 

TV
4

 

 

CORRELATION
ANALYSIS FOR
VARIABLES

 

I
t

 

 

FORMULATION
OF BINARY
PROBIT MODELS

 

I
t

I

 

 

EVALUATION
OF MODELS

 

I
l

J

 

 

INTERPRETATION

 

Figure 3.2 Flowchart of steps in the analysis process

 

 

 

 

 

 

44
substitute.
bo = intercept
b1L . . . b,, = parameters
X1 ... Xn== independent variables that are associated with
the probability of adoption or non-adoption of an
individualstrategy. (Specific variables are presented in
the variable selection section).
e1 = error term

After data collection, parametric and non-parametric
tests are used to determine whether or not the two villages
are statistically different. If they are, village-specific
models for wood energy production and consumption strategies
will be developed. This can be accomplished by having
individual models (e.g., Village 1-Kenya Ceramic Jiko) or by
combining village data and using dummy variables (0, 1) to
test village effects.

After estimation, the probabilities and response
elasticities in probit models are evaluated at the mean
values of the explanatory variables. Response elasticities
represent the percentage change in the probability of
adoption of a particular strategy as a result of a one
percent change in the value of an explanatory variable
(e.g., household income, age of the head of household, and

educational level of the head of household).

 

45
H . 1] fl 1 !'

Based on studies by Voh (1982), Rogers (1983), Akinola
(1987), Dennis (1990), Polson and Spencer (1990), and Hodges
and Cubbage (1990) among others, the independent variables
and a_nriori relationships shown in Table 3.01 were
developed. Variable definitions are included as Appendix 1.

The primary attributes of an innovation thought to
influence adoption include the relative advantages of the
new innovation, compatibility with the existing social
system, how complex it is to understand and use a particular
innovation, whether an innovation can be tried on a limited
experimental basis (trialability), and whether the results
of an innovation are visible to potential adopters
(observability). Rogers (1983) confirmed that these
attributes were significant in many studies. Other studies
such as Olson (1982) and Gottschalk (1982) generated
consistent results. Olson investigated the adoption of
statewide forest planning in Michigan while Gottschalk
studied the diffusion of several educational and criminal
justice programs. Besides the five primary attributes given
above, there may be other special features that may
facilitate or hinder the adoption of innovations. This
study is primarily concerned with households as adopters of
innovations.

The operationalization of innovation adoption as a

dependent variable may take three forms (Downs and Mohr,

 

 

 

 

46

Table 3.01 Research variables and their expected
relationship to adoption

 

 

Variable‘ Expected
Relationship to
Adoption

Farm size (AREA) +
Area under cultivation (AC91) +
Net Area (NAREA) +
Land tenure structure (TNl, TN2, + or -
TN3, TNP)
Size of household (HSZ) +
Age of head of household (AGE) -
Educational level of head of +
household (EDH)
Income” (CREV, TREV, THHX) +
Attitude of head of household + or -

towards the future, i.e. economic
outlook (ECOU)

Leadership position (LEAD) +
Gender of head of household (GNDH) + or -
Contacts with extension services +
(EVPY)

Access to different sources of +
information (RHPD, NPPM, TVHPD)

Participation in agricultural shows +
and meetings (SH10, MTATT)

Attributes of the strategies (COST, + or -

CLEANLINESS, SAFETY, AVAILABILITY)
Labor availability“ (LBAV) -

 

‘See Appendix 1 for variable definitions and units of
measure.

FIncome variables were created from itemized lists of
sources of income and expenditures that were developed in
consultation with the small-scale farm households.

°Labor availability in man-days per year was computed by
listing all persons working on the farm, number of hours
spent working per day, and days per year worked.

47
1976). Binary (yes or no) answers can be used to categorize
adopters and non-adopters. Second, the duration that a
particular innovation has been used could also be used as a
dependent variable that measures adoption. Finally, the
extent of innovation adoption within a household or
organization may also be used. The three dependent
variables are conceptually different and the determining
variables may also be different. In this study, a binary
dependent variable is used. The rationale for this choice
is presented under the model selection process in the
section on econometric models.

In a recent study, Rauniyar and Goode (1992) examined
the adoption of seven modern agricultural practices by
small-scale farmers in Swaziland. The objectives of their
study were to determine if the adoption of different
practices is interrelated and to determine the nature of
that relationship if it did exist. They treated adoption as
a continuous variable. Factor analysis was then used to
develop adoption indices. However, most technology adoption
studies treat different technology choices as independent of
each other. Similarly, this study treats the adoption of
different strategies as being independent of each other.

As the research objectives indicate, the ultimate goal
is to identify the predictors of adoptive behavior with
regard to fuelwood production and consumption strategies.

Rogers (1983) indicated methods used to predict

 

48
innovativeness included multiple correlation, configura-
tional methods, and probability models. As noted
previously, this study uses a probability model in the

prediction process.

5 1. E . l E 1

Wills

For this study, the sampling frame was small-scale farm
households in central Kenya. The primary data collection
effort in the two villages was also complemented by
secondary data from several public agencies such as the
Central Bureau of Statistics (Ministry of Finance and
Planning), the Ministry of Energy, and the Ministry of
Environment and Natural Resources.

A simple random sample of households in each of the two
villages was taken. The households of interest were the
registered parcels in the study areas. The names of
landowners were obtained from land registry and verified
with cooperative society records. A simple random sample

was drawn from each village using a random number table.

W

A cross-sectional survey was used. It was selected
over trend, cohort, or panel surveys because of the time and

resources available for the field work. A trend survey

49
requires several measurements over time for the same general
population. A cohort survey on the other hand draws new
samples at different times from the same specific
population. Panel studies involve studying the same
individuals repeatedly (Finsterbusch, 1983).

Since the study was designed to investigate some
strategies which have been in place for periods of up to ten
years, an attempt to obtain some longitudinal data was made
by asking questions about strategies and attitudes in the
past. Answers to questions about the past may be less
accurate than answers to questions about the present, but in
the absence of good secondary data, this may be the only way
to get a glimpse of the past.

Face-to-face interviews were used in this study. The
low levels of literacy among the heads of households and the
limited availability of telephones greatly reduced the
potential for using telephone surveys, leave and pick-up
questionnaires, and mail surveys. Face-to-face interviews
are more expensive than; however, they generally ensure a
higher response rate and better answers where questions are
of a technical nature and may thus require further
explanation.

Dillman (1978) described telephone and mail survey
instrument design and discussed some points which are also
important in face-to-face interviews. He stressed lowering

the costs to respondents in terms of time taken, effort

 

50
required, and psychological discomforts. An effort to
maintain trust and anonymity was also required. These
factors were given due consideration and the questionnaire
was approved by the University Council for Research in Human
Subjects (UCRHS) at Michigan State University as having
taken the necessary precautions to protect the respondents

(Appendix 2).

W

The 43-question questionnaire (Appendix 3) was
pretested in both villages. Each of the enumerators
interviewed five households.1 The responses were discussed
intensively during the enumerator training sessions to
ensure that they were all clear. Some questions were
dropped while others were modified for clarity. Pretesting
helped ensure that questions provided the information
sought. For example, after finding that few households kept
good records of incomes and expenses, it was decided that
elaborate questions on monthly incomes and expenses be asked
and used to compute the annual values needed for this study.

The training process involved several day-long meetings
with all the enumerators over a period of two weeks. All

questions were discussed with them to ensure they fully

 

1Four enumerators were recruited in Village 1 and five in
Village 2. Most of them were college students. A teacher and an
agricultural extension agent were recruited as part of the Village
2 survey team.

 

51
understood the final version of the questionnaire. Several
trips to the Ministry of Energy training center at Wambugu
Farmers Training center near Nyeri town were made. They
were designed to familiarize the enumerators with
agroforestry and wood-energy technologies related to this
study. The head of household was interviewed when he or she
was present. If not available, the spouse was interviewed.
If neither was available, the person who managed the farm on

their behalf was interviewed.

W

The sample size selected was based on tests to be used,
levels of significance desired, variability in the
independent variables, funds available, and estimated
incidence of different strategies in the population.

During the questionnaire pretesting stage, enumerators
were asked to seek information on the proportion of
households using different strategies. This figure was used
to compute the minimum sample size required to meet the
objectives of the study. Sample size selection guidelines
provided by Cochran (1983) were used. The crucial steps in
selecting the sample size are:

(1) developing a statement on our expectations from the
sample
(2) finding some equation that connects the sample size

with the desired precision level

52

(3) estimating the unknown parameters

(4) deciding whether to create subgroups and then pick
subsamples among them with the sample size being the
sum of the subsamples

(5) finding a method for reconciling the differences in the
level of precision in measuring different variables

(6) evaluating the sample size to see whether sufficient
resources are available to attain it.

The above steps were followed in selecting the sample
size. A significance level of a = 0.05 was selected. This
significance level is consistent with other studies (Spector
and Mazzeo,1980; Aldrich and Cnnude, 1975). The selected
significance level allows a 1 in 20 chance of rejecting a
true null hypothesis. A very low a would imply a large
sample size. There is also an associated higher probability
of accepting a false null hypothesis. The following sample

size determination formula for a = 0.05 was used (Cochran,

1977):
n = $29
25
where:
n = sample size
P = the expected percentage of adoption (based on

pretests and secondary data when available) of a
specific strategy in a village

Q = l-P

53

ap== J 29 (standard deviation of P)
n

Based on pre-test data and discussions with enumerators, the
expected incidence of the "Kuni Mbili" stove/fireplace was
used to estimate the appropriate sample sizes. The a_prigri
expectation was that most rural households used at least the
three-stone fireplace which could be replaced with a "Kuni
Mbili" stove with a minimum investment cost. This strategy
was therefore used as a baseline against which the adoption
of new technologies was evaluated. For Village 1, P was
estimated at 10 percent and Q at 90 percent. The estimated
sample size was therefore 144. In Village 2, P was
estimated at 11 percent and Q at 89 percent. A sample size
of 157 was therefore estimated. In the actual study,
samples were slightly larger.

Finsterbusch et_alL (1983) suggested that samples of 40
to 80 could provide useful information on population
characteristics and attitudes if used cautiously. He
further noted that such relatively small samples may be
sufficient if the study is not going to play a critical role
in the decision-making process. For this study, simple
random samples of 145 households in Village 1 and 161
households in Village 2 were selected. Within each village,
all registered land parcels had equal probabilities of being
selected as part of the survey sample.

The original number of households settled in Village 1

in 1962 was about 270. At the time of the survey (July-

54
September, 1991), there were 324 registered land parcels.
This increase has been due to subdivisions of the original
parcels either for sale or for distribution to children of
the original settlers. There are also landless households
that live in the village. The dairy marketing cooperative
records listed a total of 558 members. This figure included
the following: households that had been allowed by
landowners to live on their land and derive a livelihood
either partially or completely outside their farms,
households of employees of the landowners, and households of
relatives of the landowners. The official 1979 census
figures indicated a total of 469 households and a population
density of 217 people per square kilometer (G.K., 1980).
The 1989 census results had not been published at the time
of this study.

In Village 2, land registry records indicated a total
of 565 land parcels. As in Village 1, this number only
included the registered landowners. Landless households are
also present in Village 2 as indicated by census figures.

In 1979, there were 338 households and a population density
of 368 people per square kilometer in the village (G.K.,
1980). The number of households during the 1989 national
census had not been published at the time of this study.

In designing the questionnaire, the guidelines provided
by Finsterbusch et_n1‘ (1983) and Yambo (1982) were followed

to the extent possible. The guidelines recommended clear,

55
specific, short, closed-ended, positive, objective,

comfortable and appropriate questions.

ii i . l . !l S
The 43—question questionnaire required an average of
one hour to complete. Sometimes the questioning took longer
since farmers wanted to explain some aspects of the problems
they faced in greater detail. In some cases, they would
invite the enumerators to observe the trees they had planted
as well as the cooking stoves that they were using. The
heads of households, their spouses, or their representatives
were available in nearly all cases. Repeat visits were made
as necessary to ensure that the appropriate individuals were

interviewed.

D I E l i I 'l' 1 E J
Computer coding and data entry was done after the
survey was completed. QUATTRO PRO 3.0 spreadsheets were
used for data entry (Borland International, 1991). PARADOX
3.5 was used as a database management tool (Borland
International, 1990). SYSTAT 5.0 was used for statistical
analysis (Wilkinson, 1990). SYSTAT Probit (Version 1.3) was
initially used in model formulation (Steinberg, 1988). At a
- later stage in the analysis, a switch was made to LIMDEP,
Version 6.0 ( Greene, 1992). The reason for the switch was

that LIMDEP was more effective in calculating partial

56
derivatives needed to estimate the elasticities at the mean
values of independent variables. LIMDEP also generated a
cross-tabulation of correct versus incorrect predictions as
well as other measures of goodness of fit. The likelihood
ratio test was used as the measure of the goodness of fit
for probit models. Cross-tabulation provided a useful
insight to the models’ predictive power. Results from
LIMDEP and SYSTAT were otherwise consistent.

The two villages were expected to be different in a
number of parameters, hence, the selection of independent
samples from each village. Prior to any model building,
tests were run to establish whether the two villages were
statistically different. The parametric t—test and the
Mann-Whitney U test were used: these are discussed in the
next section. For many of the important variables, the two
villages were statistically different. Therefore, they were
treated as different populations for the purpose of this

analysis.

W

The population under study was defined as the
households of small-scale agricultural landowners in the
high potential areas of central Kenya. Village 1 and
Village 2 were selected to capture the diversity of wood
energy sources (national forests versus privately grown

trees) and the different impacts of fuelwood shortages in

57
the region. Within the high agricultural potential zone of
central Kenya, Village 1 is similar to other villages which
have been settled since independence in 1963 whereas Village
2 represents an older, more established village.
Accessibility and expected support from local administration
officials were also considered. Random samples were
selected in the above context.

A concise definition of the sampling frame and the
selection of a random sample reduced the possibility of a
sampling bias. Subjective sampling leads to a sampling bias
and this makes it difficult to statistically interpret the
data (Hammond and McCullagh, 1978). A sampling bias is the
difference between the expected value of the population
estimator and the population quantity being estimated.

The questions used in the survey were clarified as
needed following the pretest to eliminate any biases that
may have resulted from instrument design. Another potential
source of bias is where a substantial difference between the
sampled and target population exists. Care was taken to
ensure that the lists used did not exclude certain sections
of the population that should have been a part of the
sampling frame. Non-land owners were not considered part of
the sampling frame for the purposes of this study. The
principal focus was on land owners who had more latitude in
making land-use decisions (e.g., planting trees). Moreover,

collecting information on landless residents would also have

58
immensely increased the cost of the study since no previous
lists existed. In determining the technologies used for
cooking and heating, the sampled households were asked to
answer some questions on technologies used by other
households living on their land. A future study designed to
evaluate technology adoption by households that do not own
land can take care of this limitation. Non-response can be
another source of bias if it is extensive. Face-to-face
interviews were used, and non-response was not a major

problem.

E ! . l H _ ! . I I

In testing the stated research hypotheses (see Table
3.1), a significance level of a = 0.05 was selected.
Appropriate statistical tests are selected to determine the
probability that the problem data may have been a chance
phenomenon under the null hypothesis. If the probability is
less than the selected a, the null hypothesis is rejected at
the specified significance level.

Hammond and McCullagh (1978) noted three important con-
siderations in the selection of statistical tests for
research hypotheses. They are: (1) data characteristics,
(2) relative cost-effectiveness of different tests, and (3)
assumptions made about the background populations from which
the data are derived. The data collected in this study

included nominal, ordinal, interval and ratio measurement

59

scales. Data on interval and ratio scales can be subjected
to more powerful tests (e.g., the parametric t-test) than
data on nominal and ordinal scales. Assumptions on the
background populations determine whether parametric or
non-parametric tests are used. The Mann-Whitney U-test is
the non-parametric equivalent of the t-test. It possesses
about 95 percent of the power efficiency of the t-test.

Parametric tests can only be applied to data that are
at least on an interval scale. The principal assumption in
these tests is that the background population is
approximately normally distributed. If a small sample size
(less than 30) is to be used, it is required that the
population be nearly normal. The t-test, which is the most
power-efficient test, is used in the hypotheses tests when
normal distribution can be safely assumed.

Normal distribution is completely determined by two
parameters: the mean and the standard deviation. The
reasons for its use include the following: (1) it is
extensively and accurately tabulated, (2) some variables are
approximately normal, (3) simple transformations may induce
normality for other variables, and (4) repeated samples
yield normally distributed values (Central Limit Theorem).
In this study, the SYSTAT Lilliefors chi-square goodness of
fit test was used to test for normality (Wilkinson, 1990).
It standardizes the variables and tests whether the

standardized versions are normally distributed and then

60
examines whether the observed distribution in the sample is
likely to be different from a normal distribution. The
variables' original means and standard deviations do not
affect the test. The distributions of research variables
were not significantly different from normal in either
village. Therefore, variable transformations were not
necessary.

Non-parametric tests are also referred to as
distribution- free tests. They make no assumptions on the
distribution of the background population and typically use
some simple aspects of the sample data such as signs of the
measurements, order relationships, and category frequencies.
When the sample size is small or moderate, it is best not to
assume normality unless overwhelming evidence to the
contrary exists. As the sample size increases, there is a
tendency towards normality. In a study on income
distribution in central Kenya, Kmietowicz and Webley (1975)
found that income had a log-normal distribution. A prior
assumption that income was normally distributed may have led
to misleading analysis if parametric tests only were used.
Using non-parametric tests or otherwise transforming such
variables improves the analysis. As mentioned above, data
transformation was unnecessary in the current study.

Some of the variables in this study are difficult to
measure on meaningful numerical or ratio scales (e.g.

attitude towards tree planting, preferences for specific

61
methods of cooking, etc.). Appropriate non-parametric tests
in this study include the binomial test for differences of
proportions or numbers, sign tests, and the Mann-Whitney U
test. More details on specific tests are provided in

Chapter 4.

MW

Econometric models provide an empirical content to
economic relationships. The model-building process seeks to
develop models that adequately characterize a given data
set. The first step is to assume a set of models which are
all equally probable (Grasa, 1989). The true model M is "a
member of a class of models M such that the distributional
characteristics prescribed by M on a given population P are
the same as those generated by P (Grasa, 1989)." The
econometric model selection process thus involves looking at
a wide range of models that may be applicable and then
narrowing down the choices based on research hypotheses and
population characteristics. Specification errors may result
from omitted variables and inclusion of irrelevant variables
(Greene, 1991).

The probable models for this analysis include ordinary
least squares regression, linear probability models, binary
choice models (probit and logit), and models with limited
dependent variables (Tobit) (Maddala, 1983: Greene, 1991).

Discriminant analysis may also be used but it is more

62

appropriate for generating rather than testing hypotheses
(Johnston, 1980). Factor analysis could also be used
(Rauniyar and Goode, 1992).

The ordinary least squares regression model can be
specified as follows:

Y = XB + e
where:

Y = an n*1 matrix of dependent variables,

X = an n*k matrix for n observations and k independent

variables,

B = a k*1 matrix of unknown parameters, and

e random disturbance term.

e is N(0,6’) and Var e = E [ee'] = 6’1.

The general linear model above assumes that: there is
no multi-collinearity, ea is white noise with mean zero, the
random part is independent of the random regressors, and the
vector e has an n-variate normal distribution. The
parameters of the model are calculated using the formulae
shown.

B = (x'xr1 X’y

6’ = e'e/n-k

where:
e = y-X(X'X)“X'y = (I-X(X'X)*X')y = My,
M = I—X(X'X)“X,

(M is a symmetric idempotent matrix which premultiplies

any vector y to produce the vector of least squares

63

residual in the regression of y on X).

I an identity matrix,

x a matrix of independent variables, and

y = a vector of dependent variables.

In a situation where the dependent variable can only
take two values, the models described later are superior to
the general linear model. The general linear model analyzes
one dependent variable at a time. For example, in a study
where the dependent variable is the adoption of a specific
technology, the general linear model can be misleading since
it is based only on individuals who had adopted. In so
doing, current non-adopters are excluded from the analysis.
Non-adopters should be viewed as displaying rational market
behavior which also needs to be explained. Leaving them out
could lead to an omitted variable problem.

The variables examined in this study relate to whether
an innovation is accepted or rejected. Judge et_alL (1988)
noted that "questions about whether to produce or consume
rather than how much to produce or consume" are of great
interest to the economist. This study is concerned with
such questions. The research interest is in modeling choice
behavior of small-scale farm households when they are faced
with the decision of whether or not to adopt a particular
wood-energy production or consumption strategy.

The strategies evaluated are the adoption or non-

adoption of: (1) fuelwood tree planting, (2) the "Kuni

64
Mbili" fireplace, (3) the Kenya Ceramic Jiko, (4) kerosene
as a wood-energy substitute, and (5) electricity as a wood
energy substitute. Seme farmers were also found to be using
innovative strategies such as liquid petroleum gas, solar
power, and sawdust stoves. The numbers were however too
small to allow adequate statistical analysis of the users.
A general discussion of such cases is presented in Chapter 4
and is based on the average characteristics of those
households.

The linear probability model is considered to be
computationally simple and has been used in a variety of
studies (Amemiya, 1981: Debertin et_al;, 1980). Like other
binary choice models, let Y32be a binary dependent variable
with Y = 1 for adoption (e.g., if the Kenya Ceramic Jiko
(KCJ) has been adopted), and Y = 0 for non-adoption (e.g.,
if the KCJ is not adopted). If there are 145 observations
(1 = 1,2, ..., 145) as in Village 1, the model is set up as
below.

Y1 = X’lB + e;
where:

(1) Y; is a Bernoulli random variable, and

E [v1] = Pr [Y1 = 1] = x33.

Since Xl’B is unbounded, the model can generate proba-

bilities outside the unitary range. Such probabilities

can not be logically interpreted.

(2) The disturbance term e1 can only take two values since

65
Y; can only take two values.
E [Y1] = XI’B,
eit = 1-X1’B with probability x,'B when Y1 = 1
(Y1 = 1 => 1 = xi'B + e1 => e1 = 1-X1’B), and
eil = -X1’B with probability 1-X1 when Y1 = 0
(Y‘ = 0 => 0 = xi'a + e1 => el = -X,’B).
(3) e,L is heteroscedastic since var (ey) = E [y,](1-E[y1]).
The variance of the error term varies systematically as

independent variable values change.

It has also been argued both theoretically and
empirically that the diffusion of innovations has a
curvilinear relationship (Akinola, 1987). The above
features of the linear probability model make it unsuitable
for this study. Models that can overcome its problems can
be developed.

Logit and probit models are alternatives to the linear
probability model. The logit model assumes a logistic
cumulative distribution function in the error term which is
specified as follows (Kmenta, 1986):

F(a + 3x1) = 1

1 + e"""
As in the case of linear probability models, the binary
dependent random variable is assumed. Multinomial cases
(more than two dependent variable categories) can be

developed for both logit and probit models. Logit models

66
have been used widely in biological sciences and in
econometrics. Some studies which use logit models include
Beggs §L_§11 (1981), Bagi (1983), Capps and Kramer (1985),
Hodges and Cubbage (1983), Lee and Stewart (1990), Teachman
and Polonko (1988), Park and Kerr (1990), and Mitchell and
Preissler (1991).

In binary choice models, it is assumed that the
attributes of a particular choice and the households
socioeconomic characteristics determine the average utility
derived from adopting particular strategies. For example, a
small scale farm household's decision to use firewood or
kerosene, may be influenced by the attributes of the fuel
(e.g. cost and perceived safety) and also by the attributes
of the individual (e.g. income, age, and educational level).

A binary-choice, probit model is used in this study.
Following Dennis (1990) and Skog (1986), it can be assumed
that households seek to maximize expected utility in
choosing different strategies. The utility derived from the
choice of a particular strategy was expressed as follows:

Um = Um + em = a0 + Zmb + Wico + e10

U11 = U11 + en = a1 + znb + wici + e11

where:
i = household = 1, 2, 3, ..., I,
1 = choice to adopt is made,
0 = choice to adopt is rejected,

,5:
0
ll

utility derived by household i from choice 0,

67

U“, = utility derived by household i from choice 1,

Zm = vector of characteristics of alternative 0 as
perceived by household 1,

Zn = vector of characteristics of alternative 1 as
perceived by household i,

W1 = a vector of socioeconomic characteristics of the
1“ household,

a, b, and c express the relationship between utility

and associated attributes,

random disturbance term,

em

e,1 = random disturbance term, and

Um and U11 are_random.
Household i makes choice 1 if U11 > U1° or if the
unobservable random variable Y; = U“,- U“,>t0. Y; is the
observable random variable, where:

Yy=1ifvf>0

Y£=Oifo50

Y1 (Zn - Zio)’b + W1 (C1 - co) + (en " elo)

' b
[(211 - 210)! W1] [(31 _ C°]+ e;

= Xi’B + e,‘

The probability distribution of the error term (ef)
may take different forms, and the assumptions made determine
the model choice. A probit model assumes standard normal
errors, and a logit model assumes logistic error terms.

Linear probability models, however, assume uniform

68
distribution of the error terms (Maddala, 1983).

The cumulative distribution function for the standard
normal distribution, which is used in the probit model, is
specified as follows (Kmenta, 1986):

a + BX

F (a + 8X1) = _L Ie"'dzi

i2n -m

where w = 23/2
Both probit and logit formulations have an S-shaped
distribution. Figure 3.3 illustrates their difference with
the straight line formulation used in ordinary least squares
regression.

Dennis (1990) carried out a study on forest harvest
decision making by non-industrial private forest owners in
New Hampshire and used a probit analysis to estimate the
probability of timber harvesting given the attributes of a
forest, the owner, and socioeconomic variables. Skog (1986)
carried out a study on factors influencing county-level
household fuelwood use in the United States. A probit model
was developed to explain the probability of households using
wood with certain socio-economic characteristics and located
in given environments.

Purnama (1990), used Tobit analysis to study household
energy consumption in Java, Indonesia. The model estimated
the probability of a household selecting a particular type
and amount of energy. His results indicated that it would

be feasible to encourage alternative fuels and reduce the

69

Figure 3.3 Comparison of ordinary least squares estimate
and "typical" S-shaped relationship with a
dichotomous dependent variable

 

 

Source: Aldrich and Cnudde, 1975.

70
use of fuelwood. Tobit models are useful in specifying
quantities and probabilities of various possible outcomes.
The current study confines itself to questions of whether
or not certain strategies are adopted and does not therefore
use the Tobit analytical framework.

The policy question for Kenya relates to increasing
fuelwood production while also complementing these efforts
with other strategies that lead to more efficient wood-
energy use. Various fuelwood substitutes are considered.

The choice between probit and logit models appears to ‘
be largely a matter of the researcher's preference. Kmenta
(1986) noted that the two models are similar when we are
considering dichotomous dependent variables. This is the
situation in the present case where we are interested in
whether a wood energy production or consumption strategy is
adopted or not. Results can however be significantly
different if we are considering more than two dependent
variables (Hausman and Wise, 1978). Such cases are referred
to as multinomial models. Parameter estimates may also
differ if a lot of observations are at extreme probability
values (Aldrich and Nelson, 1984).

Sellar et_al. (1986) used a logit model in a contingent
valuation study of the demand for recreational boating (a
non-market good) in Texas. They argued that the logit was
simpler to estimate and compared well with the probit. The

logit model assumes independence from irrelevant

71
alternatives in a multiple-choice situation.

Probit and logit models can provide useful policy
analysis information on the probabilities of adoption or
non-adoption of various strategies. After evaluating the
range of models that could be used in this study, probit
models appear to be best suited for the analysis. In
summary, they are appropriate for the following reasons:

(1) the dependent variable is conceptualized as a binary
choice, (2) the attributes of households and different wood
energy production and consumption strategies can be observed
or derived, (3) the calculated probabilities are bounded
between 0 and 1, (4) they compel the error term to be
homoscedastic, (5) they generally require fewer observations
than logit models, and (6) the probit model is effective
even when strategies are close substitutes.

The probabilities in probit models are commonly
evaluated at the mean values of the explanatory variables.
The effect of a unit change in an explanatory variable on
the probability of adoption depends on the specific value of
the explanatory variable under consideration (Gunderson,
1974).

Response elasticities for any changes in the
explanatory variables can also be calculated. They
represent the percentage change in the probability of
adoption of a particular strategy as a result of a one

percent change in the value of an explanatory variable.

72
The elasticity is usually evaluated at the mean value of the
explanatory variable.

The coefficients in probit models indicate the effects
of changes in the independent variables on F’1 (P1). P1 is
the probability of adopting strategy i. The coefficients
can not be interpreted directly like those in the OLS model
(Aldrich and Nelson, 1984). Instead, coefficient signs,
significance level, and elasticities are used. Initial
values of the independent variables and coefficients
determine the magnitude of change in the probabilities. The
following relationships illustrate this:

P1 = F(xl’B)

Xs’B = F“(Ps)

Q21 = F(Xh'B)
6X“,

where:
P1 = probability of adopting strategy i,
F = probability density function,
Xi' = vector of independent variables,
B = matrix of parameters to be estimated,
k = index of independent variables, and
QB; = response elasticity for independent
8X1k variable

An iterative process was used in determining the
variables that would remain in the final models reported.
The testing process began with an examination of the a

,pzigri expectations presented in Table 3.1 and correlation

73

analysis. The variables retained in the final models were
those that either had strong a_prigri expectations or
relatively high t-statistics for the independent variables.

The goodness of fit for the models developed may be
measured by the likelihood ratio test, MacFadden's R2, the
log-likelihood function, or the percentage of correct
predictions (Aldrich and Nelson, 1984: Polson and Spencer,
1990). If desired the coefficients of linear probability,
logit, and probit models can be transformed in order to make

them comparable (Kmenta, 1986).

Summary

This chapter has presented the conceptual and
theoretical foundations upon which this study was based.
Sampling, survey administration, model selection, and
variable selection were also discussed. The merits of using
probit analysis over other techniques was presented.
Different software packages that may be used to carry out

the analysis were also described.

IV. RESULTS AND DISCUSSION

This chapter presents the results of specific
parametric and non—parametric tests used to examine
differences between villages and between adopters and non-
adopters. Then, probit models based on factors hypothesized
to influence adoption or non-adoption of different
strategies are presented. The process of statistical
inference from LIMDEP output is also described. Finally,

the significance of the models is discussed.

W

To determine whether there were statistical differences
between Village 1 and Village 2 and between adopters and
non-adopters, the parametric t-test and the non-parametric
Mann-Whitney U-test were used. SYSTAT was used for this
part of the analysis.

The data being analyzed were unpaired measurements on
different scales. In comparing two sets of data, the t-test
for independent samples is appropriate for data on interval
and ratio scales. The Mann-Whitney test is more appropriate
for categorical data (Hammond and MacCullagh, 1978).-

The t-test is the most widely used and is the standard

74

75
against which the power efficiency of other tests is
measured. The t-test assumes that background populations
from which the samples are taken are normally distributed
and that the standard deviations are equal. To test the
normality assumption, the Lilliefors Chi-square test for
normality was applied to different variables (Wilkinson,
1990). The analysis indicated that the distribution of the
variables used for the t-test was not significantly
different from normal at a = 0.05.

The null hypothesis was that there was no difference
between the means of the populations of which Village 1 and
Village 2 were samples. The selected level of rejection was
a = 0.05. The two-tailed critical t-value for large samples
at that rejection level is 1.96. The null hypothesis of no
difference in the means is rejected for all values of t
above the critical value.

The Mann-Whitney U-test, on the other hand, is one of
the most powerful distribution-free tests and has close to
95 percent of the power of the parametric t-test. It also
works well with small, medium, and large samples. The test
makes no assumption about the characteristics of the
distribution. The test can also be used with categorical
data. It examines whether the differences in the means of
two independent samples are statistically significant. The
U-statistic is calculated by considering sample sizes and

scores from two samples as a single set and then ranking the

76
entire group from lowest to highest. If the null hypothesis
of no mean difference is true, the observations from the two
samples should be fairly well mixed. Conversely, if the two
samples are from different populations, observations from
one sample will be banded together. A rejection level of a

= 0.05 was also used in this case.

E' . E H E'EE E 1!
Was

In the initial selection of the two villages, it was
hypothesized that they were different in a variety of
important parameters such as income, wood energy
availability, development of rural infrastructure, and
duration of settlement. Before probit analysis for the two
villages could be done, it was necessary to establish
whether the villages were statistically different. Such a
difference would justify the running of separate probit
models for the two villages or including a village dummy
variable in combined data models.

Table 4.01 shows a comparison of mean values using the
parametric t-test. In the t-test, significant differences
were observed for area of land owned (AREA), total revenue
(TREV), crop revenue (CREV), age of the head of household
(AGE), extension agent visits to the farms (EVPY), household

size (HSZ), hours spent listening to the radio (RHPD),

77

Table 4.01 Comparison of mean values using the parametric

 

 

t-test
Variable' Village 1 Village 2 t-statistic
Means Means

AC91 2.80 2.47 1.94
AREA 9.63 4.65 5.37*
CREV 40033.03 23235.34 5.20*
FIC 15093.23 9155.19 4.61*
SAVINGS 5579.15 1261.44 0.99
TEXP 35295.32 28050.86 3.26*
THHX 20751.54 18895.67 1.15
TREV 41400.62 29312.29 3.22*
AGE 54.68 49.31 3.46*
EDH 5.95 4.86 -0.30
HSZ 5.16 6.24 -3.72*
TIC 3.08 1.51 7.50*
TIM 24.90 23.23 1.11
EVPY 4.86 9.66 -4.15*
RHPD 2.54 3.55 -4.00*
SH10 2.66 2.17 1.31
CCBM 0.95 0.41 6.25*
CGAM 0.04 0.08 -1.26
CRHM 0.86 10.20 -9.29*
FWHM 12.10 14.76 -6.15*
KELM 7.25 5.94 1.49

 

* Significant at a = 0.05
‘ See Appendix 1 for variable definitions.

78

Table 4.01 (continued)

The variables above are grouped as follows:
1. Farm Area--AC91, AREA
2. Financial Information--CREV, FIC, SAVINGS, TEXP,

THHX, TREV
3. Household Characteristics--AGE, EDH, HSZ, TIC,

TIM, DSHH
4. Exposure to Information--EVPY, RHPD, SH10
5. Fuel Quantities Used per Month--CCBM, CGAM, CRHM,

FWHM, KELM
headloads2 of fuelwood and crop residues (FWHM and CRHM) ,
bags of charcoal used monthly (CCBM), farm input costs
(FIC), and time spent in collecting fuelwood (TIC).
Variables not significantly different included area under
cultivation (AC91), level of savings (SAVINGS), total
household expenditure (THHX), years of formal education
received by head of household (EDH), number of years the
family has been settled on the land (TIM), attendance to
agricultural shows in the preceding ten years (SH10),
cylinders of gas used per month (CGAM), and liters of
kerosene used per month (KELM).

The average area of farm land owned (AREA) is 9.6 acres

in Village 1 and 4.7 acres in Village 2. The former is a
resettlement scheme created about thirty years ago while the

latter village has been settled for many generations. In

both villages, population has grown over the years and land

 

2A headload is the quantity of fuelwood that can be
conveniently carried on the back from the source to.the home
where it is eventually used. This task is almost always
performed by women. Average headload size may vary with age of
the person, as well as the species and condition of the fuelwood.

79
parcels are relatively small.

Crop revenues (CREV) are statistically different.
Coffee and tea are the main crops for Village 2, and the
world market prices for them have been fairly low during the
last few years. Village 1, on the other hand, depends on
horticultural and dairy products and the households appear
to be doing relatively well. Total revenues (TREV) are also
significantly different for the two villages at the
specified level. Farm input costs (FIC) are, however,
significantly different and this is probably a reflection of
the different input (land, labor, capital, and technology)
requirements for coffee, tea, horticulture, and dairy
production. The total household expenditures (THHX) are not
significantly different, and this appears to indicate that
consumption levels are fairly similar.

There are several differences related to fuel usage.
Village 1 has more physical access to fuelwood but uses
fewer headloads of fuelwood per month (FWHM) on average
compared to Village 2. However, it (Village 1) does use
more charcoal bags per month (CCBM) than Village 2. These
differences are examined further in probit analysis. More
headloads of crop residues per month (CRHM) are used in
Village 2: this is probably associated with the relative
Vcosts of alternative fuels. Tea and coffee twigs from
annual pruning are also used. Households in Village 1 spend

twice as much time in collecting (TIC) a headload of

80
fuelwood compared to Village 2. This probably reflects the
distances that have to be travelled to collect fuelwood from
national forests. For Village 2, access to fuelwood from
national forests is limited.

Results of the Mann-Whitney U-test are presented in
Table 4.02. The U-statistic revealed further significant
differences between the villages in the percentage of
households adopting the Kenya Ceramic Jiko (KCJU), use of
the traditional metal stove (TMSU), land tenure categories
‘(TN1, TN2, TN3 and TNP), use of the "Kuni Mbili"
stove/fireplace (KUMU), and use of kerosene stoves (KESU).
The Kenya Ceramic Jiko is used by 43 percent of the
households in Village 1 and only 4 percent of the households
in Village 2. Households in this village may be responding
to higher prices of charcoal compared to prices in Village
1. Village 2 appears to use the "Kuni Mbili" fireplace more
than Village 1. In relative terms, this probably reflects
the greater scarcity of fuelwood in Village 1.

Tree planting activities (TPHSIO) in the two villages
during the preceding ten years (1982-1991) were also
significantly different. Forty-five percent of the
households in Village 1, and 20 percent in Village 2
reported having planted multi-purpose trees during the
specified period. Village 1 was established as a

resettlement scheme and most of the households are first

81

Table 4.02 Comparison of mean values using the U-test

 

 

Variable‘ Village 1 Village 2 U-statistic
Means Means

FUKE 2.77 2.68 12741.0
FUGA 2.88 2.91 11421.0
FUFW 1.03 1.14 10893.0*
FUEL 3.00 2.98 11817.5
FUOT 2.94 2.98 11573.0
FUCR 2.08 2.34 8711.0*
FUCC 2.06 2.65 5316.0*
TPHSlO 0.45 0.20 14585.0*
KUMU 0.15 0.26 10390.5*
ELEU 0.00 0.01 11527.5
SDSU 0.00 0.04 11237.5*
OTHU 0.04 0.09 11140.5
TSFU 0.88 0.83 12181.0
KCJU 0.43 0.04 16236.5*
KESU 0.08 0.27 9521.0*
TMSU 0.63 0.37 14801.0*
DSHH 2.27 1.99 12875.5
LEAD 0.46 0.33 l3223.5*
TNl 0.88 0.32 18126.0*
TN2 0.02 0.36 7709.0*
TN3 0.10 0.32 9110.0*
TNP 0.08 0.14 11043.5

 

*

1.

2.

3.
4.

Significant at a = 0.05

See Appendix 1 for variable definitions

The variables above are grouped as follows:

Fuel Used by Households--FUKE, FUGA, FUFW, FUEL, FUOT,
FUCR, FUCC

Cooking and Heating Technologies used by Household--KUMU,
ELEU, SDSU, OTHU, TSFU, KCJU, KESU, TMSU

Household Characteristics--LEAD

Land Tenure Structure--TN1, TN2, TN3, TNP

82
generation farmers and have full private ownership rights
over their land. Full-land ownership rights are very
important in making long-term management decisions such as
tree planting. Only 10 percent of the households in Village
1 had restricted land ownership rights (TN3, inherited but
still shared) as compared to 33 percent in Village 2..
Households in Village 2 cited land shortage as a major
constraint to their tree planting.

Table 4.03 shows the adoption rates for different
strategies in the two villages. The adoption rate for the
Kenya Ceramic Jiko in Village 1 is 43 percent and 4 percent
in Village 2. Average charcoal prices are significantly
different in the two villages. At least one trader in the
main shopping center in Village 1 had a large stock of KCJ
for sale. Village 2 households used an average of 0.41 bags
of charcoal per month (CCBM) as compared to 0.95 bags per
month for Village 1. In Village 1, 88 percent of the
households reported using the three stone fireplace (TSFU)
and 63 percent the traditional metal stove (TMSU). The
corresponding figures for Village 2 were 83 percent for the
three stone fireplace (TSFU) and 37 percent for the
traditional metal stove (TMSU).

The mean difference tests have established that the two
villages are statistically different. The probit models for
the two villages are therefore run independently. An

aggregation of data from the two villages was nevertheless

83

Table 4.03 Adoption rates (percentages) for policy
supported strategies in the two villages

 

 

 

Strategy Village 1‘ Village 2b
We:

KCJU 43 4
ELEU N /A

KESU 8 27
KUMU 15 26
TPHSlO 45 20
Qld_Strategiss9

TSFU 88 83
TMSU 63 37
‘ n=145

” n=161

New strategies include the Kenya Ceramic Jiko, electric
cookers, kerosene stoves, the "Kuni Mbili"
stove/fireplace, and tree planting in the preceding ten
years.

The old strategies include the three-stone fireplace
(TSFU) and the traditional metal stove (TMSU) which are
supposed to be replaced by more efficient technologies.

84
done to evaluate whether stronger adoption models could be
generated with the combined data. Those results are
presented and discussed in a later section of this chapter.
The underlying assumption for this type of analysis was that
the two villages were subsamples representing two strata in
the population. A dummy variable was used to test for

village effect.

Was

Relative to adoption, the following independent
variables are examined: size of household, farm size, age
of head of household, educational level of head of '
household, income, land tenure structure, contacts with
different sources of information, knowledge of different
strategies, participation in agricultural shows and
meetings, leadership roles, gender, labor availability, and
area under cultivation (Voh,1982: Rogers,1983: Akinola,
1987: Dennis, 1990: Polson and Spencer, 1990: Hodges and
Cubbage, 1990). Results of these probit analyses provide
information on the household characteristics associated with
adoption or non-adoption of different strategies.

Table 4.04 presents the number of households using
different combinations of strategies. When all five
strategies are considered together, 30 households in Village
1 and 69 in Village 2 had not adopted any of them. A t-test

and a U-test were carried out to examine whether there were

85

Table 4.04 Number of households using different
combinations of strategies

 

 

Number of strategy Village 1 ‘Village 2
combinations‘ n=145 n=161
Ill] E1] 5! ! .
3 6 4
2 33 25
1 76 63
0 30 69
W
3 0 1
2 12 13
1 71 64
0 62 83

 

‘ Strategy combinations are KCJU, KUMU, KESU, ELEU, and
TPHSlO.
See Appendix 1 for variable definitions.

86
any statistical differences between the households of those
who adopted at least one consumption technology and those
who adopted none. Results are presented in Tables 4.05 (t-
test) and 4.06 (U-test). The four fuelwood- and charcoal-
saving or substituting technologieS‘were considered.

Most of the adopters had either 1 or 2 strategies.

When tree planting (TPHSlO) was excluded, the numbers
adopting none of the consumption-based technologies were
revealed to be fairly high. Based on correlation analysis,
many of the wood-energy saving technologies tended to be
substitutes to each other. Households tended to adopt only
one or two technologies at a time.

Farm input costs (FIC), total expenditures (TEXP), time
spent in fuelwood collection (TIC), cylinders of gas used
per month (CGAM), liters of kerosene used per month (KELM),
use of charcoal (FUCC), use of the three stone fireplace
(TSFU), use of the Kenya Ceramic Jiko (KCJU), and use of
kerosene stoves (KESU) were important distinguishing
variables for adopters and non-adopters in the two villages.
Other variables such as area under cultivation (AC91) and
extension visits per year (EVPY) were unique for adopters
and non-adopters in each village.

Adapters in Village 1 were characterized by less area
under cultivation (AC91), higher crop revenue (CREV), higher
education (EDH), lower duration of residence in the village

(TIM) for example. In village 2, they were characterized

87

Table 4.05 Comparison of adopters and non-adopters,
Independent t-test

 

 

 

 

Village 1 Village 2
Variable‘ Without TPHSIO Without TPHSlO
AC91 -2.97* 0.48
AREA -0.03 -0.28
CREV 3.47* 1.34
FIC 2.29* 2.02*
SAVINGS 1.80 -0.15
TEXP 3.71* 1.98*
THHX 4.25* 1.13
TREV 3.57* 1.15
AGEH -3.33* -0.74
EDH 3.17* 1.43
HSZ 0.67 2.16*
TIC 3.51* -2.15*
TIM -2.23* -0.40
EVPY 1.62 3.49*
RHPD 1.15 3.16*
SH10 3.15* 0.98
CCBM 3.93* 1.67
CGAM 2.27* 1.98*
CRHM -0.94 2.16*
FWHM -1.37 -2.53*
KELM 2.66* 3.30*
* Significant at a = 0.05

‘ Variable definitions are in Appendix 1.

88

Table 4.06 Comparison of adopters and non-adopters, Mann-
Whitney U—test

 

 

 

Village 1 Village 2
Variable‘ Without TPHSlO Without TPHSlO
FUKE 2428.5 4693.0*
FUGA 2747.0 3530.0*
FUEL 2562.0 3320.0
FUOT 2436.0* 3240.0
FUCR 2196.0* 3281.5
FUCC 3307.5* 3770.5*
KUMU 1921.5* 1535.5
ELEU 2562.0 3154.0
SDSU 2562.0 3229.5
OTHU 2596.5 3139.0
TSFU 3111.0* 3955.0*
KCJU 640.5* 2946.5*
KESU 2196.0* 1452.5*
TMSU 2873.5 2881.5

 

* Significant at a = 0.05
‘ Variable definitions are in Appendix 1.

89
by higher expenditures (TEXP, FIC), higher household size

(HSZ), and lower time in collecting fuelwood (TIC).

W

The average duration for which a technology has been in
use is also important in evaluating adoption.i Table 4.07
shows the average duration of use for different technologies
in the two villages. The four energy-saving or wood-
substituting technologies have been promoted for about ten
years. The Kenya Ceramic Jiko (KCJU) had been adopted for
an average of 3.3 years in Village 1 and only 1.4 years in
Village 2. Similarly, the "Kuni Mbili" stove/fireplace had
been adopted for an average of 3.3 years in Village 1 and
3.8 years in Village 2. The corresponding values for

electricity and kerosene are as shown.

E!! .1 ! E S! ! .
Another important dimension to adoption is the

attributes of the strategies. Table 4.08 summarizes the
desirable attributes cited by households as important to
adoption. Responses were not provided by some adopting
households and this limits the analyses of the role of
attributes that could potentially be carried out with more
complete data. In addition to the attribute data, some
concerns on some of the technologies were also raised.

Costs, suitability for multiple uses, and lack of

90

Table 4.07 Average duration of technology use in years

 

 

Technology‘ Village 1 Means Village 2 Means
TSFU 25.5 (n=130)b 23.6 (n=132)
TMSU 21.6 (n=95) 10.8 (n=132)
KCJU 3.3 (n=60) 1.4 (n=7)
KUMU 3.3 (n=18) 3.8 (n=41)
KESU 11.0 (n=12) 9.1 (n=41)
ELEU 0.0 (N/A) 9.5 (n=2)
SDSU N/A 1.7 (n=6)
OTHU N/A 8.3 (n=11)

 

‘ See Appendix 1 for variable definition.
” Number of households reporting duration of use.

91

Table 4.08 Attributes for different strategies

 

Technology

Attributes

 

Village 1

Village 2

 

KCJU

ELEU

Better than previously
used technology (TMSU)
Easy to use

Culturally acceptable in

village
Charcoal-saving
Clean
Trialability (can be
dropped easily if
found unsuitable)
Observability (had seen

it used and liked it)

Better than TSFU
Efficient
Acceptable in the
village

Easy to use

N/A

Better than TMSU

Easy to use
Acceptable in village
Trialability
Observability

Warmth

Safe

Convenient

Efficient

Not smoky

Concerns
Not available locally

Too small
Costly
No knowledge of it

Efficient
Safe
Improvement over

previous

technology
Easy to use
Acceptable
Trialability
Observability
Quick cooking
Cheap

Observability

Improvement over
previous
technology

QQDQEIDS
Costly

 

92

 

 

 

Table 4.08 (continued)
Attributes
Technology Village 1 Village 2
KESU Easy to use Better than
Improvement over current previously used
practice technology
Easy to use
Acceptable in village
Trialability
Observability
Fast
Convenient
Cheap
Safe
QQDQEID§3
Costly
TMSU Warmth Improvement over TSFU
Improves on TSFU Easy to use
Easy to use Trialability
Acceptable in village Observability
Clean Acceptability
Durable Cheap
Cheap Warmth
Fast
Clean
QQDQQID
Charcoal is costly
TSFU Easy to use Acceptable in village
Accepted in village Observability
Observability Warmth
Fuelwood abundant Easy to use
Cheap Improvement
Warmth Copes with green wood
Multipurpose Safe
Convenient Adequate fuelwood
Cheap
SDSU N/A Cheap
OTHU N/A Improvement
Trialability
Observability
Fast (gas)

Clean (gas)
Fast (gas)

 

93
information were cited as problems for the efficient or
wood- and charcoal-saving technologies considered in the
study. In general, these responses were consistent with
those noted by Rogers (1983) and which were discussed in

Chapter 3.

EIQDiL_E§§ElL§

This section discusses the twelve-probit model results
obtained for the two villages. The first eight models deal
with the adoption of the Kenya Ceramic Jiko (KCJU), the
"Kuni Mbili" stove/fireplace (KUMU), kerosene stoves (KESU),
and tree planting in the preceding ten years (TPHSlO) in
Villages l and 2. The next four models are based on
combined data. Electricity had such a low-use rate for
heating and cooking that no significant model could be
developed. The two households that were using electricity
are described using basic descriptive statistics such as
means. A discussion of models estimated, important policy

variables, and interpretation follows.

AssessinLEstimatieuesults...

The individual coefficients in probit models are the
parameter estimates for the independent variables. The
associated standard errors and t-statistics as generated by
LIMDEP probit are also presented. The estimates of

coefficients are asymptotically unbiased and efficient.

94
This means that coefficients are valid as approximations and
would be expected to improve as sample size was increased.

The t-statistic tests the null hypothesis that a
particular coefficient is zero. The estimated coefficients
are B/a, with a assumed to be equal to 1 (Latrielle, 1992).
Significant variables are identified as being associated
with the likelihood of adoption or non-adoption. Probit
estimates are maximum likelihood estimates.

In ordinary least squares regression analysis (OLS),
the F—statistic tests the hypothesis that all coefficients
except the constant are zero. For probit analysis, the
likelihood ratio test is used (Aldrich and Nelson, 1984).

The Likelihood ratio statistic (c) is calculated as follows:

c = -2 log (LO/L1) = -2(log L0 - Log L1)
where:
c = likelihood ratio statistic.

L1 = value of the likelihood function for the full
model as estimated. I
L0 = Maximum value of the likelihood function if all
coefficients except the intercept are zero.
The calculated xf'tests the null hypothesis that all
coefficients except the intercept are zero. For example, in
probit model KCJl (Table 4.09),
-2 LLR (calculated x3, 5 OF) = 61.674
x” TAB (5, 0.05) = 12.832

The critical value from Chi-square tables show the

95
probability that.x? calculated is the result of a chance
distribution. The larger the value of XZ'TAB (relative to
1’ calculated), the smaller is the probability that null
hypothesis is correct. The degrees of freedom are based on
the number of coefficients minus one. The above model is
significant based on these criteria. All the probit models
presented in this study are significant and are evaluated in
the same manner.

The percentage of adopters and non-adopters correctly
predicted can also be used as another measure of the
goodness of fit for the models developed. The predictive
power of the models based on this criterion is discussed
later.

Probit analysis does not have a statistic comparable in
interpretation to R? in regression analysis. Aldrich and
Nelson (1984) caution that even though analogous statistics
such as the pseudo-Ra and McFadden's R2 can be developed,
such summary measures should be used with "extreme caution,
if at all."

The single village probit models evaluated in this
study appear in Tables 4.09 through 4.14, 4.16, and 4.17.
Each model is for a specific wood-energy production or
consumption strategy. As indicated earlier, the only
production strategy considered in this study was the growing
of fuelwood or multipurpose trees. The models for each of

the two villages are evaluated together to provide a

96
contrast as to what differences existed, if any. The
factors associated with the adoption of different
technologies are then discussed.

For all probit models, elasticities for the independent
variables at the mean levels were calculated using the
procedure explained in Chapter 3. These values indicate the
percentage change in probability of adoption associated with
a one percent change in the mean value of the independent
variable. The change may be positive or negative signifying
that the probability of adoption either increases or
decreases. These values were calculated using LIMDEP
(Version 6.0). The values and signs of these elasticities
are useful in interpreting the results since they indicate
which policy has a greater impact on the probability of
adoption. The lumpiness of units such as extension visits
(EVPY) and age of the head of household (AGE), however, is a
limitation in the interpretation of the elasticities. For
example, the effect of a 1—percent increase in extension
visits on adoption at the mean may be extremely low.
Marginal changes (elasticities) nevertheless provide useful
information on important policy variables.

Significant variables in the models may also be ranked
by elasticities. This involves using mean value data and
the calculated elasticities to determine the percentage by
which the probability of adoption or non-adoption changes as

a result of a 1-percent change in the independent variable.

97
For the Kenya Ceramic Jiko models which appear in the next
two sections (KCJl and KCJ2), the top three variables in
rank from highest to lowest using the above procedure were:
liters of kerosene used per month (KELM), extension visits
per year (EVPY), and age of head of household (AGE). The
more commonly used procedure of explaining elasticity is to
present results for each probit model. Then, quantitative
results from different models are not inappropriately
combined.

All probit models are presented in tabular rather than
equation format so that standard errors, t-statistics, and
elasticities can be efficiently included. The equation
specification format was presented in Chapter 3. Model KCJl
which follows in the next section may, for example, be
specified as Y, = 0.5766 - 0.0393(AGE) + 0.0295(EVPY) +
0.0735(KELM) + 0.5E-04(THHX) - 0.3046(MEMFEOF). Y1 is a
binary dependent variable while the other variables are as
defined in Appendix 1. Variables for which strong a_prigri
expectations of significance existed are retained in the
final model presentation even when they are insignificant to
illustrate their effects (or lack of effects) on adoption or
non-adoption.

The model structure presented for different
technologies is based on prior hypotheses about the
variables considered to be associated with adoption or non-

adoption based on literature and personal observations (see

98
Table 3.01). In addition, correlation analyses results were
used to determine the variables to be included in a single
model. For example, in Village 2, farm input costs (FIC)
and crop revenues (CREV) had a correlation coefficient of
0.739 and could therefore not be used in the same model due
to potential multicollinearity problems. Similarly, total
revenue (TREV) and crop revenue (CREV) had a correlation

coefficient of 0.975 in Village 1.

WW (Model KCJ 1)

The adoption of the Kenya Ceramic Jiko in Village 1
presents a rather unique situation: forty-three percent of
households sampled have already adopted it. There were a
number of explanatory variables that were statistically
significant and associated with the adoption of this
charcoal-using stove. Results are presented in Table 4.09.
Age (AGE, negative coefficient), extension visits per year
(EVPY, positive coefficient), liters of kerosene used per
month (KELM, positive coefficient) and total household
expenditure (THHX, positive coefficient) were statistically
significant.

As expected, older heads of households (AGE) were found
to be less likely to adopt the Kenya Ceramic Jiko than the
younger heads of households. This is consistent with much
of the literature reviewed. Younger heads of households

tend to be more mobile, more educated, and more anxious to

99

Table 4.09 Probit results for the adoption of Kenya Ceramic
Jiko (KCJ1) in village 1.

 

 

Variable‘ Coefficient Std t-value Elasticityb
error

CONSTANT 0.5766 0.5325 1.083

AGE -0.0393 0.0102 -3.863* -0.015
EVPY 0.0295 0.0175 1.688 0.012
KELM 0.0735 0.0281 2.616* 0.029
THHX 0.5E-04 0.1E-04 4.416* 0.2E-04
MEMFEOF -O.3046 0.2894 -1.052 -0.120

 

Log-likelihood = -68.421

-2LLR (CHI-SQ, 5DF) = 61.674

*Significant at a = 0.05

CHI-SQ TAB = 12.832

‘ See appendix 1 for definition of variables

‘ Elasticity computed at mean values for independent
variables

100
try new methods than older heads of households. Although
these factors are important, resource endowments of the
individuals are also likely to be important and age alone is
certainly not sufficient in determining adoption.

Total household expenditures (THHX) were statistically
significant. In Village 1, most of the households relied on
agricultural revenue for their livelihood. Total household
expenditure tended to be higher than crop revenue (CREV) and
this possibly reflected the existence of non-farm income or
borrowing. The expected positive sign of the coefficient was
consistent with the a_prigri expectation. On average, it
was expected that high expenditure households were also
likely to have higher incomes. Higher household
expenditures were thus associated with the likelihood of
adopting the stove (positive coefficient).

The presence of members of households who were employed
off-farm (MEMFEOF) was statistically insignificant. The
financial contribution of such family members was initially
expected to be positive.

Kerosene was used for cooking and heating by 8 percent
of the households in Village 1. Based on this model, the
liters of kerosene used per month (KELM) had a positive and
significant coefficient at a = 0.05. The a_pzigri
expectation was that the coefficient sign could be either
positive or negative. Kerosene can be used as a substitute

(negative sign) for or to complement (positive sign)

101
charcoal in cooking and heating. In this case, it appears
that kerosene is more of a complement since its increased
use tends to be associated with more use of the charcoal-
based Kenya Ceramic Jiko. Kerosene stoves cook fast and
some charcoal-users may prefer them when less cooking time
is crucial.

The elasticities show the responsiveness of the
likelihood of adoption or non-adoption to changes in
different independent variables. A l-percent increase in
the number of extension visits (EVPY), for example, leads to
a 0.012 percent increase in the likelihood of the adoption
of the Kenya Ceramic Stove in Village 1. High elasticities
indicate that adoption can be influenced through appropriate
policy changes directed at those variables. Increased
support of extension services, for example, would be a
helpful policy in facilitating adoption. Programs that
increase household income are also likely to encourage
adoption of the Kenya Ceramic Jiko. All the elasticities
presented in subsequent models are conceptually interpreted

in the same way.

EWWZ (Model KCJ 2)
Table 4.10 presents model KCJ2 (probit analysis results

for the adoption of KCJ in Village 2). The statistically
significant variables at a = 0.05 were the liters of

kerosene used per month (KELM), the acreage owned by the

102

Table 4.10 Probit results for the adoption of Kenya Ceramic
Jiko (KCJ2) in Village 2.

 

Variable‘ Coefficient Std error t-value Elasticityb

 

CONSTANT -6.1252 1.716 -3.569

HSZ 0.0993 0.0976 1.017 0.6E-03
KELM 0.0889 0.0307 2.898* 0.6E-03
AREA 0.0669 0.0305 2.198* 0.4E-04
THHX 0.4E-04 0.2E-04 2.787* 0.3E-06
MEMFEOF 1.6130 1.012 1.594 0.010

 

Log-likelihood = -16.39

-2LLR(CHI-SQ, 5DF) = 24.797

*Significant at a =0.05

CHI-SQ TAB = 12.833

‘ See appendix 1 for definition of variables

” Elasticity measured at mean value of independent variable.

103
household (AREA), and the total household expenditure
(THHX). These variables were positively related to the
likelihood of adoption. Household size (H52) and the
presence of family members who were employed off—farm
(MEMFEOF) were insignificant. Large households in general
were considered likely to use more fuel and therefore have
an incentive to increase efficiency. Family members
employed off-farm were likely to contribute to household
income and to come across information on new technologies.

As indicated earlier, Village 2 faced a more acute
fuelwood and charcoal shortage compared to Village 1. Many
socioeconomic variables were also different for the two
villages as shown earlier in Tables 4.01 and 4.02. In
addition to liters of kerosene used by a household per month
(KELM), total household expenditure(THHX), and whether a
member of the household was employed off- farm (MEMFEOF)
which were significant in Village 1, farm area (AREA) was
found to be significant for Village 2.

Only 4 percent of the households surveyed in Village 2
were using the Kenya Ceramic Jiko. Twenty-seven percent of
the households, on the other hand, reported using kerosene
stoves for cooking and heating. Based on anecdotal
evidence, the low level of adoption for the KCJ was probably
due to lack of adequate information on this technology.
About 37 percent of the households surveyed indicated they

were using the traditional metal stove which the KCJ was

104
designed to replace. This can be compared to 43 percent in
Village 1 who reported using the KCJ and 63 per cent who
reported using the traditional metal stove (Table 4.02).

There is a rather high positive correlation coefficient
(0.538) between liters of kerosene (KELM) and bags of
charcoal used per month (CCBM). This indicates that there
was a tendency for households using more kerosene to use
more charcoal as well. This possibly reflects a situation
where the two fuels are used to complement each other.

Fuel price may affect adoption, but in some cases is
relatively equal for adopters and non-adopters. There were
a number of petrol (gas) stations in the small urban center
near Village 2. This may have eliminated the middlemen who
retail kerosene in Village 1. The middlemen buy it from the
petrol stations and then sell at higher prices in retail
outlets. At the same time, charcoal had to be transported
in trucks over relatively long distances to Village 2. The
average price per bag therefore tended to be higher in
Village 2 (Ksh 105.70 versus Ksh 60.20). Transportation
costs were lower in Village 1.

Farmers with large tracts of land were generally
expected to have higher crop revenues (CREV). Total
household expenditures (THHX)'were found to be significant
in determining the adoption of KCJ at a = 0.05. The poorest
households may continue to rely on fuelwood even when labor

(time spent in collection) and fuelwood costs are relatively

105
high due to relatively higher costs of substitutes.

The elasticities in this model are low, and relatively
high investments in the desirable policy variables would be
needed to increase adoption. Income increasing policies
would be helpful since households are more likely to spend
money on new technologies as incomes rise. A focus on

farmers with larger farms may also facilitate adoption.

Adoption of Kuni Mbili stayezfiizenlage in yillage 1 (Model
KUMl)

Table 4.11 presents the results for this technology.
Fifteen percent of the households in Village 1 had adopted
this technology. This can be contrasted with eighty-eight
percent of the households who reported using the three-stone
fireplace and were therefore potential adopters.

The statistically significant variables at a = 0.05
included acres of land owned (AREA, positive), headloads of
crop residues used (CRHM, negative), leadership role in the
village (LEAD, positive), attendance to agricultural shows
(SH10, positive), and total revenue (TREV, negative). The
coefficient signs for AREA, CRHM, LEAD, and SH10 were
consistent with a_nIini eXpectations. The existence of a
member of the household who worked off-farm (MEMFEOF) was
insignificant. Higher household incomes make other energy
options feasible and this may explain the negative total

revenue (TREV) coefficient. Households with leaders,

106

Table 4.11 Probit results for the adoption of "Kuni Mbili"
stove/fireplace (KUMl) in Village 1

 

Variable‘ Coefficient Std error t-value Elasticityb

CONSTANT -0.4125 0.4368 -0.944

CRHM -1.6995 0.5753 -2.954* -0.108
LEAD 1.2766 0.4599 2.776* 0.081
AREA 0.0886 0.0281 3.159* 0.006
SH10 0.1711 0.0531 3.221* 0.011
TREV -0.4E-04 0.1E-04 -2.719* -0.2E-05
MEMFEOF -0.5142 0.3894 -1.320 -0.033

 

Log-likelihood = -42.802

-2LLR (CHI-SQ, 6 DF) = 34.348

*Significant at 0 =0.05

CHI-SQ TAB = 14.449

‘ See appendix 1 for definition of variables

” Elasticity computed at mean values for independent
variables.

107
regular agricultural show attenders, larger tracts of land,
and high consumption of crop residues as a fuelwood
substitute were more likely to adopt this technology.
Identifying households with these characteristics is a
useful first step in facilitating more adoption of the ”Kuni
Mbili" stove/fireplace.

The elasticities for the independent variables are as
shown. Participation in agricultural shows and leadership
role have relatively high elasticities. Policy variables in
order of importance were LEAD, SH10, and AREA. Focusing
extension information on leaders, charging low gate fees to
encourage show attendance, and advising farmers with larger

tracts of land may enhance the adoption of this

technology.
Adoption of "Kuni Mbili" Stgyezfiireplage in Village 2 (Model
KUM2)

The "Kuni Mbili" stove/fireplace had been adopted by
twenty-six percent of the households surveyed. At the same
time, 83 percent of the households reported using the
traditional three-stone fireplace at least part of the time.
The adoption rate was higher than for Village 1.

Model results are presented in Table 4.12. The
quantity of crop residues used per month (CRHM) and the
number of extension visits per year (EVPY) were significant

in influencing adoption of the Kuni Mbili stove/fireplace.

108

Table 4.12 Probit results for the adoption of "Kuni Mbili"

stove/fireplace (KUM2) in Village 2

 

 

Variable‘ Coefficient Std error t-value Elasticity”
CONSTANT -1.6670 0.280 -5.986

CRHM 0.0315 0.009 3.383* 0.010
EVPY 0.0199 0.010 2.034* 0.006
TREV 0.6E-05 0.4E-05 1.528 0.2E-05
LEAD 0.2249 0.2574 0.874 0.069
MEMFEOF 0.3015 0.2365 1.275 0.092

 

Log-likelihood = -80.069
-2LLR (CHI-SQ, 5 DF) = 22.563
Significant at a = 0.05
CHI-SQ. TAB = 12.8325

‘ See appendix 1 for definition of variables.
Elasticity computed at mean value of independent

1)

variable.

109
Both had positive coefficients. As mentioned earlier, crop
residues were widely used in Village 2. This reflected the
relatively higher physical scarcity of fuelwood in this
area. Tea and coffee prunings were considered valuable
sources of fuel. More extension visits per year (EVPY) were
found to positively associated with adoption. Extension
contacts provide new information to farmers and their
households. It is likely that farmers who seek and receive
extension services are more likely to be adopters of the
"Kuni Mbili" stove/fireplace. As extension visits increase,
households are more likely to acquire knowledge on new
technologies. This is consistent with findings in
agricultural economics literature. Total household revenue
(TREV) and leadership role (LEAD) were not statistically
significant in influencing adoption.

Response elasticities were also relatively low for the
different variables. Significant gains in adoption can be
made by promoting extension activities and also
concentrating on communities that are using plenty of crop

residues for cooking and heating.

WWW (Model 10351)

The policy approach with regard to kerosene has been to
subsidize it so that it is more widely available at low
prices. Average prices of kerosene in Village 1 were KSh

9.80 per liter compared to KSh 9.03 per liter in Village 2

110
(Table 4.05). Only 8 percent of the households in Village 1
reported using kerosene for cooking and heating (Table 4.3).
As can be observed in Table 4.13, the significant variables
were age (AGE, negative coefficient), extension visits per
year (EVPY, positive coefficient) and the education level of
the head of household (EDH, negative coefficient). Initial
expectations were that kerosene stoves would be used by
younger, better educated, more wealthy, and more progressive
households. Leadership role (LEAD), member of household
working off-farm (MEMFEOF), and total revenue (TREV) had
statistically insignificant coefficients. Kerosene users in
this village are likely to be younger. The older heads of
households in the village were less likely to adopt kerosene
stoves as expected. The negative coefficient sign on
educational level, however, was inconsistent with
expectations. It is possible that younger and more educated
heads of households preferred other energy alternatives.
More educated heads of households were considered more
likely to be involved in both on-farm and off—farm
activities which would encourage them to use fast cooking
technologies such as the kerosene stove. The older but less
welléeducated heads of households in this village could also
have had similar pressures on their time due to involvement
in non-farm economic activities. Extension visits play an
important positive role in promoting adoption. Households

that have interest in extension information are more likely

111

 

 

Table 4.13 Probit results for the adoption of Kerosene
Stove (KESl) in Village 1
Variable‘ Coefficient Std error' t-value Elasticityb
CONSTANT -2.2314 30.85 -0.072
AGE -l.0935 0.0197 -2.791* 0.6E-03
_ EVPY 0.0585 0.0246 2.381* 0.6E-03
LEAD 0.6133 0.4008 1.530 0.007
EDH -0.2243 0.1003 -2.236* -0.002
MEMFEOF 4.4580 30.83 0.145 0.049
TREV 0.2E-05 0.6E-05 0.409 0.3E-07

 

Log-likelihood = -27.878
-2LLR(CHI-SQ, 6DF) = 27.027
*Significant at a = 0.05
CHI-SQ TAB = 14.449

1)

variables.

See appendix 1 for definition of variables.
Elasticity computed at mean value of independent

112
to be aware of different technologies and their relative
merits or demerits.

The elasticity for the educational level of the
household (EDH) identifies it as the most useful policy
variable in the model. In the short run, it is difficult to
change average education levels. However, long-term
investments in formal and informal education could
contribute significantly to the adoption of energy-saving
technologies. A deliberate effort to focus extension
information on younger heads of households would also be

appropriate.

W2 (Model KESZ)

In the case of Village 2, twenty-seven percent of the
households reported using kerosene for cooking and heating.
Probit results for this strategy are presented in Table
4.14. Extension visits per year (EVPY) were found to
positively and significantly influence the adoption of
kerosene stoves at
a = 0.05. Age (AGE) was insignificant only at a = 0.05.
Extension visits (EVPY) and age (AGE) were also significant
for kerosene stove adoption in Village 1. Individuals who
receive extension support on a regular basis appear to be
more willing to try out new technologies. There is
inadequate evidence that younger heads of households are

more likely to be associated with adoption at the specified

113

Table 4.14 Probit results for the adoption of Kerosene
Stoves (KES2) in Village 2

 

Variable‘ Coefficient Std error t-value Elasticity”

 

CONSTANT -0.1065 0.5258 -0.203

AGE -0.0150 0.0086 -1.737 -0.005
EVPY 0.0238 0.0104 2.277* 0.007
LEAD -0.4690 0.2877 -1.630 -0.145
TREV 0.3E-05 0.4E-05 0.743 0.1E-05
RHPD 0.0651 0.0519 1.254 0.020
TIC -0.1756 0.1163 -l.510 0.054
Log-likelihood = - 83.835

-2LLR(CHI-SQ, 6 DF) = 19.198

*Significant at a = 0.05

CHI-SQ TAB., 6 DP = 14.449

‘ See appendix 1 for definition of variables.

” Elasticity computed at mean value of independent
variable.

114

significance level. Time spent in fuelwood collection
(FIC), and leadership role (LEAD), total revenue (TREV), and
exposure to radio information (RHPD) were statistically
insignificant. Kerosene subsidies are likely to favor those
individuals who actively seek and receive extension advice.
These are usually the more progressive members of the
community.

The elasticities for the independent variables indicate
that increased extension services would facilitate faster

adoption of this technology.

it Ii E E] ! . °! . M']] 2
As shown in Table 4.03, only 1 percent of the

households in Village 2 reported using electricity for
heating and cooking. Due to the small number of electricity
users in the sample, no significant models could be
generated at the specified significance level. Installation
costs of electricity, purchase costs of appliances, and high
monthly bills may be hindrances to wider adoption of
electricity for cooking and heating especially for low-
income rural households.

A comparison of mean values for some variables is
presented in Table 4.15. Electricity users appear to have
above average revenues (CREV, TREV), expenditures (TEXP,
THHX), education (EDH), age (AGE) and frequency of visits

from extension agents (EVPY). On the other hand, their use

115

Table 4.15 Comparison of electricity user means against
Village 2 Means

 

 

Variable‘ Village 2 Electricity-Users
Means Means

AC91 2.47 2.05
AREA 4.65 3.95
CREV 23235.34 26100.00
TEXP 28050.86 47390.00
THHX 18895.67 40190.00
AGE 49.31 54.0
EDH 4.86 5.5
H82 6.24 8.5
TIM 23.23 28.5
EVPY 9.66 18.0
SH10 2.17 2.0
CCBM 0.41 2.0
CRHM 10.20 6.0
FWHM 14.76 6.0

 

‘ See Appendix 1 for variable definitions.

116
of crop residues (CRHM) and fuelwood (FWHM) is below the
village average.

People who are leaders in their communities tend to be
better off than the average village resident and may also
view the adoption of new technologies as being consistent
with their social status. They therefore tend to become
early adopters. The two households adopting electricity for
cooking and heating had members with leadership roles in the
community. The small group of households using electricity
implies that other energy alternatives are used more widely.

A more comprehensive study of electricity users in
Village 2 would involve obtaining a list of all users from
the Kenya Power and Lighting Company and treating them as a
unique stratum. A sub-sample could then be drawn and used
to evaluate the adoption of electricity for cooking and
heating. Characteristics of those households could then be

analyzed and compared with those of the village as a whole.

1000 o! e *- ' g. '0 ' g- ‘ - -op.o‘ O -;_ 1
11113519.]. (Model rpm)

The reported adoption of tree planting in Village 1 was
45 percent as compared to 20 percent in Village 2 (Table
4.03). Village 1 is located in a region that was
historically well-wooded. The stock of wood on private
farms however appears to have declined during the past three

decades (personal communication with farmers). A government

117
tree nursery was located close to this village, and this may
have encouraged all categories of small-scale farmers to
participate in tree planting. Seedling prices were also
low. Their counterparts in Village 2 often cited lack of
land as the main obstacle to tree planting while Village 1
residents cited game damage, insect damage, and lack of
adequate water supplies as their main obstacles to planting.

Table 4.16 presents the model results for tree planting
in Village 1. Larger, higher income households (HSZ, TREV)
appear to be the ones more likely to plant trees. Larger
families may increase the pool of resources available to the
household which may in turn enable them to plant trees. The
pool of resources may include finances, labor and ideas.
Farmers with large tracts of land (AREA) are less likely to
adopt tree planting. Such farmers probably do not perceive
lack of wood for future use as a risk since they may already
own adequate reserves. Farmers with higher incomes can
aIso afford more energy alternatives and hence the positive
and statistically significant coefficient for total revenue
(TREV).

The elasticities in this model indicate that a focus on
farmers with smaller than average farms might yield faster
adoption results. Such farmers are likely to be aware of
the need to plant more trees and effort can be focused on
providing logistic support to them. That may, for example,

involve making sure seedlings are available at planting time

118

 

 

Table 4.16 Probit results for the adoption of tree planting
(TPHl) in the preceding 10 years in Village 1

Variable‘ Coefficient Std error t-value Elasticity”

CONSTANT -0.8687 0.3969 -2.188

HSZ 0.1594 0.054 2.951* 0.062

TREV 0.1E-04 0.4E-05 2.320* .0.4E-05

AREA -0.0655 0.023 -2.833* -0.025

SH10 -0.0698 0.040 -1.730 -0.027

TNl 0.3408 0.3845 0.886 0.1323

TN3 -4.0110 45.21 0.089 1.556

 

Log-likelihood = -86.216
-2LLR (CHI-SQ,
*Significant at a = 0.05

CHI-SQ TAB = 14.449

‘ See Appendix 1 for definition of variables

6 DF) = 27.027

” Elasticity evaluated at mean value of independent

variables.

119
and providing extension support as needed. Policies that
increase household incomes would also facilitate adoption of
tree planting. Information on family sizes may also be used
to identify households that would be more likely to adopt
tree planting.

0 O 0 0 O
A -- . - . O .
...-.00 or 0, so 1.0 o r 0. 1° 0 -. 1

21113ng (Model TPH2)

Twenty percent of the households surveyed in Village 2
reported having planted fuelwood or multipurpose trees in
the preceding ten years. Model results are presented in
Table 4.17.

The statistically significant variables related to tree
planting in the preceding ten years were: headloads of
fuelwood used per month (FWHM, positive), and total
household expenditure (THHX, negative) at a = 0.05.
Household size (HSZ) was insignificant at a = 0.05. Land
tenure coefficients (TNl and TN2) were also statistically
insignificant.

Initially, it was expected that individuals without
restrictions in land ownership rights were more likely to
adopt tree planting. The current results do not provide
evidence in support of this hypothesis. The three tenure
structures conceptualized for the two villages do not
capture major differences in the likelihood of households to

adopt tree planting. Even though lack of adequate land was

120

Table 4.17 Probit results for the adoption of tree planting
(TPHZ) in the preceding 10 years in Village 2

 

 

Variable‘ Coefficient Std error t-value Elasticity”
CONSTANT 2.3791 0.5091 -4.673

FWHM 0.0456 0.0154 2.976* 0.011
THHX -0.3E-04 0.1E-04 ~2.106* -0.7E-05
HSZ 0.1136 0.0587 1.935 0.027
RHPD 0.0674 0.0540 1.248 0.016
EVPY 0.0084 0.0110 0.763 0.002
TNl 0.2708 0.3167 0.855 0.065
TN2 0.2665 0.3030 0.880 0.064

 

Log-likelihood = -69.246

*Significant at a = 0.05

-2LLR (CHI-SQ, 7 DF) = 22.082

CHI-SQ TAB = 16.013

‘ See Appendix 1 for definition of variables

” Elasticity computed at mean value of independent
variables.

121
often cited as a major reason for not planting trees, farm
area (AREA) and the net area (NAREA) variable were not
statistically significant and are not included in the final
model presented here. This seemed to indicate that other
factors were more important. If labor availability was a
crucial concern, the households using more fuelwood were
likely to place a higher value on cost and time savings as
a result of growing their own fuelwood. Generally, labor
availability did not appear to be a cruCial problem. As
household expenditures increased, the headloads of fuelwood
consumed tended to decrease. Higher income and expenditure
households are more likely to adopt different technologies
and fuels. The elasticities for total household expenditure
(THHX) and headloads of fuelwood used per month (FWHM)

indicate their sensitivity to policy interventions.

nghined_Dnt§_MOd§l§

Models KCJC, KUMC, KESC, and TPHC (for the Kenya
Ceramic Jiko, ”Kuni Mbili" stove/fireplace, kerosene stoves,
and tree planting respectively) were estimated using
combined data for the two villages. The full results are
presented in Appendices 6, 7, 8 and 9. A short discussion
of the results follows.

For model KCJC (Appendix 6), the significant variables
at o = 0.05 were: age of head of household (AGE, negative

asign), time spent in fuelwood collection (TIC, positive

122
sign), liters of kerosene used per month (KELM, positive
sign), total household expenditure (THHX, positive sign),
farm area (AREA, negative sign), and village dummy variable
(DUMMY, positive sign). Younger heads of households,
increase in distance from the source of fuelwood, more
reliance on kerosene, relatively small farm sizes, high
household expenses and village dummy variable (DUMMY) are
associated with the adoption of the Kenya Ceramic Jiko in
the two villages. Policies directed towards increasing
incomes and subsequently the consumption levels of
households facilitate adoptions. Focus on younger heads of
households and those with relatively small tracts of land
may also help increase adoption.

Results for the adoption of the "Kuni Mbili"
stove/fireplace are presented in Appendix 7. The
statistically significant variables at a = 0.05 were
leadership role (LEAD) and the headloads of crop residues
used per month (CRHM). Both had positive coefficients.
Individuals faced with acute fuelwood shortages tend to use
more crop residues, and a deliberate policy focus on this
group would encourage them to use the "Kuni Mbili”
stove/fireplace to save whatever fuelwood may be available.
.As observed in literature and some of the other models,
leaders are likely to be early adopters of technology.
IGetting appropriate information to them would therefore be

very helpful .

 

123

In the kerosene stove model (Appendix 8), extension
visits per year (EVPY) and hours of radio exposure per day
(RHPD) were significant at a = 0.05 with positive signs
indicating that the two sources of information were
associated with adoption. Age of head of household (AGE)
and education level (EDH) were significant with negative
coefficient signs at a = 0.05. Younger and less educated
heads of households were thus more likely to be associated
with adoption. The village dummy variable was significant
indicating differences in the two villages play a role in
adoption.

Tree planting adoption (Appendix 9) was different for
the two villages. Acres of farmland owned (AREA) and
household size (HSZ) were statistically significant factors
associated with adoption at a = 0.05. Farmers with large
tracts of land also tended to be less inclined to plant
while those with large households (HSZ) were more interested

possibly due to their greater need for wood products.

E' i E I . . M . l]

The variables used in probit model building are briefly
defined in Appendix 1. The importance of the variables is
highlighted in this section under the following categories:
farm area, financial information, household characteristics,
exposure to information on different technologies,

tquantities per month of different fuels used, specific fuels

124
chosen by households, specific cooking and heating

technologies used by households, and land tenure structures.

Eazm_A;ea. The farm area variable was conceptualized in two
ways: the total amount of land owned (AREA) and area under
cultivation in 1991 (AC91). Village 1 had an average
ownership of 9.63 acres of which an average of 2.80 acres
were reported to be under cultivation in 1991 (Table 4.).
Village 2 on the other hand, had an average ownership of
4.65 acres with an average of 2.47 acres under cultivation
during the same year. Many of the households that had not
planted trees in Village 2 stated that they were limited by
the amount of land available. Village 1 had a net land area
(NAREA) of 6.83 acres which, theoretically, could be at
least partially be planted with trees. Village 2 on the
other hand had an average net area (NAREA) of 2.18 acres.

It was initially hypothesized that households that
owned large tracts of land were likely to be better off
economically and hence more likely to adopt new wood-energy
technologies (positive probit coefficients). It was also
hypothesized that such households would on average have
larger acreages under cultivation compared to smaller farms.
Studies by Dennis (1988), Purnama (1990), Polson and Spencer
(1990), and Akinola (1987) support the case for positive
probit coefficients for farm size. Farmers with relatively

small tracts of land were viewed as less likely to adopt new

125
technologies due to their limited resource base. The probit
analyses results however indicated that farm area and net
area were not significant factors in many of the strategy

choices.

Financial_lnfgrmatign. The collection of financial

information for rural households was somewhat difficult. In
Village 1, horticultural crops were marketed throughout the
year and the prices varied with supply and demand. The
farmers were asked to give what they considered an average
price for their produce during the preceding year as well as
the quantities sold. In the case of products such as milk,
where there was only one major buyer (i.e., Kenya
Cooperative Creameries), average prices were easier to
obtain. The generated list of product prices and quantities
was used to calculate crop revenues (CREV). Farmers were
also asked to recollect the various farm input costs they
had incurred during the previous year. Costs and quantities
were used to estimate the total farm input costs (FIC).
Household expenses were also listed and added up to obtain
total household expenditure (THHX). In some cases total
household expenditures appeared to capture non-farm income
better. Other financial variables such as total revenue
(TREV), total expenditures (TEXP), and savings (SAVINGS)
‘were created from mathematical transformations of the

financial information collected.

126

In Village 2, coffee and tea production information was
more readily available. Tea payments to farmers were made
on a monthly basis while those for coffee were made on a
quarterly basis and records were better kept. The situation
was otherwise similar to that in Village 1 with regard to
other financial information. Financial status was important
in the adoption of some strategies as the probit results

have demonstrated.

HQusehgld_§haragteri§tics. These characteristics were
mainly observed and recorded for the head of the household
and the spouse. The household variables included age of the
head of household (AGE), years of formal education received
(EDH), household size (HSZ), duration of stay in the village
(TIM), and leadership roles held by members of the household
(LEAD). These characteristics were expected to provide some
insights on household characteristics that may influence the
adoption of different strategies. Probit models developed
indicate that certain household characteristics are

associated with the adoption of different strategies.

WW- Exposure to information was

evaluated in terms of access to sources of knowledge on
different wood-energy production and consumption policies.
The different sources examined included visits by extension

agents from different government agencies (EVPY), attendance

 

127
at agricultural shows (SH10), public meeting attendance
(MTATT), and hours spent listening to the radio (RHPD).
Other sources of information such as television (TVHPD) and
newspapers (NPPM) were also considered but there were very

few farmers who had access to them.

W The quantities of
different fuels used by households were provided by the head

 

of household, spouse, or his/her representative. The
quantities were measured in terms of bags of charcoal
(CCBM), cylinders of liquified petroleum gas (CGAM),
headloads of crop residues (CRHM), head-loads of fuelwood
(FWHM), and liters of kerosene (KELM). For electricity,
monthly payments to the Kenya Power and Lighting Company
were used as a surrogate measure of the quantity consumed.
This decision was made because printed electricity bills
were not always readily available.

Fuel prices varied between thetwo villages as shown in
Table 4.18. Fuelwood from national forests in Village 1 had
an average price of Ksh 0.60 per headload. Village 2 did
not have this alternative. The cheapest option for these
without adequate fuelwood supplies of their own was buying
from neighbors at an average price of Ksh 32.50 per headload
(Ksh 279.50 per stack) The market price per stack for
Village 2 was Ksh 207.60 per stack. This was lower than

the price of Ksh 279.50 charged for the same quantity from a

 

128

Table 4.18 Average prices for different fuels and
strategies for Village 1 and Village 2

 

Fuel or Strategy

Village 1
average price
(KSh‘)

Village 2 average
price (Ksh)

 

Kerosene 9.80/liter 9.03/liter
Charcoal 60.20/bag 105.70/bag
Basic electric N/A 122.73/unit
cooker
Electricity bill N/A 176.60/month
KCJ 102.60/unit 107.00/unit
Kerosene stove 232.50 251.75/unit
Market fuelwood N/A 207.60/stack
Neighbor fuelwood 22.70/headload 279.50/stack
(32.50/headload)
Own fuelwood 15.00/headload 237.17/stack
(27.58/headload)
Traditional metal 54.20/unit 59.71/unit
stove
National forest 0.60/headload N/A

fuelwood

 

‘ 1 US$ = Ksh 26.00 (approximately) at the prevailing
exchange rate between July-September, 1991. One shilling is
equal to one hundred cents.

129
neighbor's land. Market prices were expected to be higher
than prices for fuelwood from a neighbor’s farm, and this is
an aspect that may require further study to explain why they
were not. Fuelwood merchants may have had access to low
cost fuelwood sources. Kerosene, charcoal, and stove prices
also varied. Some of the price differences are difficult to
interpret since they may have been influenced by historical
factors that could not be captured by a cross-sectional
study such as this. Kerosene stoves in Village 1, for
example, may appear cheaper on average because relatively
few households had adopted them and traders were probably
selling their old stock of equipment and accessories. They
could therefore keep the prices low to encourage households
to buy them. Village 2, on the other hand, shows a
relatively higher adoption rate for kerosene stoves. If
stoves are fast-moving products in the stores, new stock is
likely to be purchased at higher prices. As expected, the
Kenya Ceramic Jiko and the traditional metal stove were
clear substitutes in Village 1. A small positive
correlation was observed in Village 2. With the low level
of adoption, individuals probably still use both stoves

interchangeably.

W- This study
dealt with factors associated with the adoption of different

wood-saving or wood-substituting technologies. The choices

130
were clearly explained to the respondents to ensure that
answers were related only to the strategies of interest. As
noted previously, the strategies included the Kuni Mbili
stove/fireplace (KUMU), the Kenya Ceramic Jiko (KCJU),
electric cookers (ELEU), and kerosene stoves (KESU). The
incidence of traditional metal stoves (TMSU), three stone
fire place (TSFU), sawdust stoves (SDSU), and any other
technologies (OTHU) was also noted.

The percentages of households using improved cooking
and heating technologies were summarized earlier in Table
4.03. The old technologies were still widely used at the
time of the study. Table 4.19 shows the correlation
coefficients for old and new technologies for Village 1.
Table 4.20 shows the corresponding results for Village 2.
Overall, most of the technologies tended to be substitutes
for each other as indicated by the many negative correlation
coefficients in both villages. The ”Kuni Mbili"
stove/fireplace and the three-stone fireplace, for example,
are reasonably good substitutes as evidenced by a relatively
high negative correlation coefficient (a high -0.915 for
Village 1 and -0.425 for Village 2). The correlation
coefficient for KCJU and TMSU in Village 1 was 0.259.
Similarly, electric cookers (ELEU) and the Kenya Ceramic
Jiko (KCJU) are substitutes for the traditional metal stoves
(TMSU) with correlation coefficients of -0.250 and -0.230,

respectively in Village 2. Electric cookers and Kenya

 

131

 

 

 

 

 

Table 4.19 Pearson’s correlation coefficients for important
old and new technologies used for cooking and
heating in Village 1
KCJU KESU TMSU TSFU KUMU

KCJU 0.000

KESU -0.112 1.000

TMSU -0.259 -0.084 1.000

TSFU -0.077 -0.115 -0.025 1.000

KUMU -0.124 0.090 0.068 -0.915 1.000

Table 4.20 Pearson's correlation coefficients for important
old and new technologies used for cooking and
heating in Village 2

ELEU KCJU KESU KUMU SDSU TMSU TSFU

ELEU l

KCJU 0.251 1

KESU 0.059 -0.060 1

KUMU -0.066 0.085 -0.031 1

SDSU -0.022 -0.042 -0.045 0.036 1

TMSU 0.031 0.027 0.124 -0.001 -0.014 1

TSFU -0.250 -0.230 -0.067 -0.425 0.088-0.004 1

 

132
Ceramic Jiko have a small positive correlation coefficient.
The other technologies had very low correlation

coefficients.

Hgnsehgld_£uel_§hgige. Fuel choice by households was an

important concern in this study. As a result, explicit
questions on the choices were asked. The fuel choices
considered included kerosene, gas, fuelwood, electricity,
crop residues, and charcoal. An "others” category was also
included to accommodate unanticipated fuels or fuels used by
very few individuals in the population. Some unique options
such as solar energy did show up. The current study does
not include a probit analysis of such options.

An important preliminary question was whether the
choices of different fuels by a household were correlated.
Tables 4.21 and 4.22 show the correlation coefficients
associated with the use of different quantities of fuel per
month for Village 1 and Village 2 respectively. Most of the
Village 2 correlation coefficients were negative indicating
the substitutability of different fuels. The bags of
charcoal (CCBM) and the headloads of fuelwood (FWHM) used in
Village 2 for example had a correlation coefficient of -
0.157. This means that as the quantity of charcoal used by
a household increased, the quantity of fuelwood used
decreased by a small amount. The corresponding value for

Village 1 was 0.181. The quantity of liquified petroleum

133

Table 4.21 Pearson Correlation Coefficients for Quantities
of Fuels Used per Month in Village 1

 

 

CCBM CGAM CRHM FWHM KELM
CCBM 1.000
CGAM 0.652 1.000
CRHM 0.132 -0.059 1.000
FWHM 0.181 0.050 0.087 1.000
KELM 0.538 0.698 -0.094 .0.114 1.000

 

Table 4.22 Pearson's correlation coefficients for
quantities of fuels used per month in Village 2

 

 

CCBM CGAM CRHM KELM FWHM
CCBM 1
CGAM 0.462 1
CRHM -0.041 -0.159 1
KELM -0.005 0.113 0.080 1

FWHM -0.157 -0.242 -0.021 0.162 1

 

134

gas (CGAM) was, on the other hand, positively correlated

 

with the bags of charcoal (CCBM) and liters kerosene (KELM).
The correlation coefficient for kerosene and gas is very
small. Gas is generally expensive and its users probably
use charcoal as a substitute some of the time. Since there
is no policy incentive supporting the use of liquified
petroleum gas: it is not examined further in this study.

The results in Table 4.22 revealed a small negative

 

correlation (-0.005) between the quantity of charcoal used
and the liters of kerosene used per month in Village 2.
This sign is consistent with expectations, but the
correlation is very small. Kerosene and kerosene stoves
tend to be more expensive than traditional or improved

charcoal stoves such as the KCJ.

Land_Tenure_Strugtures. Land tenure was hypothesized to be
a crucial factor in determining the adoption of tree
planting practices in the preceding ten years. Three
categories of tenure were conceptualized. They were:
wholly privately owned (TNl), inherited and shared (TN2),
and inherited but not shared (TN3). The tree planting
strategy (TPHSlO) deals with the wood-energy supply side.
Most of the farmers who had planted trees in the preceding
ten years consistently cited fuelwood need as an important
consideration in the decision to plant. In all the model

results however, tenure was insignificant. It is likely

135
that the three tenure systems identified in central Kenya
confer similar rights to land use and households, therefore,

respond to tree planting similarly.

E 2'” “.1.! EH Ill]

Table 4.23 presents the predictive ability of the
models developed. Predictive ability results are generated
by LIMDEP Version 6.0 in terms of the number of cases
correctly or incorrectly predicted. The percentages of
correct predictions can be calculated from those values.

Specific models from tables 4.8 through 4.16 are listed
in column 1. Column 2 shows the values of the dependent
variable that were actually 0 (non-adopters) and were also
predicted by the model as 0. Column 3 shows the cases where
the dependent the dependent variable was actually zero but
was incorrectly predicted as 1 (Adopter). Column 4 shows
the number of cases where the dependent variable was
actually 1 but was incorrectly predicted as a zero.

Finally, column 5 shows the situations where the model
correctly predicted the number of adopters. This provides a
helpful presentation of predictive capability.

Model KCJ1 illustrates that 76 out of 82 non-adopters
were correctly predicted as non-adopters. Similarly, 60 out
of 63 adopters were correctly predicted. Model KCJ2 did
well in predicting non-adopters but only 3 out of 7 adopters

were correctly predicted. Model KUMl predicted 121 out of

136

Table 4.23 Predictive capability of the models

 

 

 

Actual 0 Actual 1
Model‘ Predicted 0 Predicted Predicted 0 Predicted 1

KCJ1 67 15 13 50
KCJ2 153 1 4 3
KUMl 121 3 18 3
KUM2 117 3 31 10
KESl 130 3 12

KESZ 114 4 36

TPHl 58 22 27 38
TPH2 126 3 27 5
KCJC 222 14 22 48
KUMC 239 5 54

KESC 243 8 49

TPHC 182 26 64 34

 

‘ Model KCJC, KUMC, KESC, and TPHC are based on the combined
data for the two villages (n=306) and reflect adoption of
KCJU, KUMU, KESU, and TPHSlO respectively.

137
124 non-adopters correctly and 3 out of 21 adopters
correctly. Model KUMZ shows that 117 out of 120 non-
adopters were correctly predicted as non-adopters while 10
out of 41 adopters correctly predicted. Model KESl
predicted 130 out 133 non-adopters correctly. The 12
adopters were incorrectly predicted as non-adopters. Model
KESZ illustrates that 114 out of 118 non adopters were
predicted correctly as non-adopters while 7 out of 43
adopters were correctly predicted. In model TPHl, 58 out of
90 non-adopters were predicted correctly while 38 out of 65
adopters were predicted correctly as adopters. Finally,
Model TPH2 illustrates that 126 out of 129 non-adopters were
predicted as non-adopters and 5 out of 32 adopters were
predicted as adopters.

The percentage of correctly predicted values is
obtained by adding the actual number of correctly predicted
zeros (non-adopters) to the actual number of correctly
predicted ones (adopters), dividing that value by the total
(sample size), and then multiplying it by one hundred.
These values were calculated for the twelve models and are
shown in Table 4.24.

The combined data models do better in predicting non-
adopters than single village models. There were a lot more
non-adopters than adopters of the technologies in the two
villages and that is the main reason why non-adopter based

percentages are more robust. The percentages for adopters

138

Table 4.24 Percentages correctly predicted for all models

 

 

Model‘ Adopters and Non-adopters Adopters
non-adopters

KCJ1 80.7 81.7 79.4
KCJ2 95.6 99.4 42.9
KUMl 85.5 97.6 14.3
KUM2 78.9 97.5 24.4
KESl 96.6 97.7 0
KESZ 75.2 96.6 16.3
TPHl 66.2 72.5 58.5
TPH2 81.4 97.7 15.6
KCJC 88.2 94.1 68.6
KUMC 80.7 98.0 12.9
KESC 81.4 96.8 10.9
TPHC 70.6 87.5 34.7

 

‘ The model acronyms are for Kenya Ceramic Jiko (KCJ1, KCJ2,
KCJC), "Kuni Mbili" stove/fireplace (KUMl, KUM2, KUMC),
kerosene stoves (KESl, KE82, KESC), and tree planting (TPHl,
TPH2, TPHC) for Village 1, Village 2, and combined data
respectively.

139
only were generally low. From a policy perspective, being
able to correctly predict a non-adopter is also as good as
being able to predict an adopter. Once the two groups are

properly identified, relevant policies can be put in place.

W

Table 4.25 presents a summary of the overall
performance of the models in terms of the expected and
observed signs on the probit coefficients for the
independent variables. The possible reason for differences
between signs were explained under separate models in this
chapter. Some of the variables expected to be significant
were not. This summary and the predictive power of the
models summary (Table 4.23 above) provide useful evaluation
criteria. Extension visits and household size had positive
coefficients in most models. Other variables produced
mixed results. An evaluation of each model shows where the

opportunity for improving adoption for each technology lies.

SHEEBIY

This chapter has presented results of the parametric
and non-parametric tests and the probit analyses. The mean
difference tests suggested that the two villages were
significantly different in a number of important variables.

The primary emphasis was therefore placed on analyzing the

140

Table 4.25 Summary of coefficient signs

 

Observed coefficient signs for

 

 

models‘
Vari abl e” 5.2111915. KCJ1 KCJ 2 KCJ C KUMl KUMZ KUMC
Sign
AGE - - -
EVPY + + +
THHX + or - -+ + +
MEMFEOF + NS° NS NS NS NS
HSZ + NS
KELM + or - -+ + +
AREA + - +
CRHM + - +
LEAD + + NS +
$111 0 + + NS
TREV + - NS
EDH +
TIC + +
RHPD +
TN1 +
TN2 + or -
TN3 + or -
FWHM +
DUMMY + or- + NS
CGAM + or - NS

 

‘ Model column headings represent technology specific model
results for Village 1, Village 2, and combined data
respectively.

” Variable descriptions appear in Appendix 1.

° Not significant (NS) at a = 0.05.

141

Table 4.25 (continued)

 

Observed coefficient signs for

 

 

modelsc

Variable‘ A_Erigzi, KESl KESZ KESC TPHl TPH2 TPHC

Slgn
AGE - - NSb -
EVPY ‘ + + + + NS NS
THHX + or - -
MEMFEOF + N3
HSZ + + N8 -+
KELM + or -
AREA + - -
CRHM +
LEAD + NS NS
SH10 + NS NS
TREV + NS NS + NS
EDH + — -
TIC + NS
RHPD + NS -+ NS NS
TN1 + NS NS
TN2 or - NS NS
TN3 or -
FWHM + +
DUMMY + or - - +
CGAM + or -

 

‘ Model column headings represent technology specific
results for Village 1, Village 2, and combined data
respectively.

” Variable descriptions appear in Appendix 1.

° Not signifiCant (NS) at a = 0.05.

142
villages as two separate entities in the central Kenya
region. To examine whether more general statements could be
made for the two villages, four more probit models (KCJC,
KUMC, KESC, TPHC) were run with combined data using a
village dummy variable. Electricity use was excluded since
it was not available in Village 1 and also because of its
very limited use for cooking and heating in Village 2.
There were only two electricity users. It was therefore
difficult to examine the factors that influenced its
adoption and discussion was based on the mean
characteristics of electricity-using households compared to
the overall village means.

A comparison of mean differences between adopters of at
least one technology and non-adopters indicated that the two
groups had different characteristics. In Village 1, they
differed in parameters such as area under cultivation
(AC91), crop revenues (CREV), household expenditures (THHX),
age (AGE) and education (EDH) among others. For Village 2,
differences included farm input costs (FIC), household size
(HSZ), extension visit per year (EVPY) and headloads of
fuelwood used per month (FWHM). Those differences are
useful indications of where policy interventions may be
made.

The probit results indicated that some of the important
factors influencing adoption of wood-energy production and

consumption strategies included age (AGE), crop revenue

143
(CREV), time spent in fuelwood collection (TIC), extension
visits per year (EVPY), education level of head of household
(EDH), area of farm land owned (AREA), and access to
different sources of information (NPPM, SH10, TVHPD, RHPD)
among others. These variables were examined and evaluated
with regard to how they affected the adoption of different
policy-supported strategies. The elasticities presented

serve a useful function in that regard.

 

 

V. SUMMARY AND CONCLUSIONS

The specific objectives of this study were (1) to
determine how households have responded to specific wood-
energy policies which were put in place or strengthened
during the 19808, and (2) to identify the factors associated
with household adoption or non-adoption of the strategies.
The intended end-use of these results is to make
recommendations on suitable implementation approaches for
wood-energy policies.

A review of literature in this area indicated that
important data were lacking in Kenya and other countries
faced with similar fuelwood problems. As a result, some
policy decisions have been based on experiences in other
sectors or regions that may sometimes be significantly
different. For example, quantitative analysis including
information on response elasticities has been lacking. The
models estimated in this study provide adoption-related
information for the two villages studied.

The more efficient wood- and charcoal-burning
technologies showed highly variable adoption rates in the
two villages. The Kenya Ceramic Jiko, for example, had been

adopted by 43 percent of the households in Village 1 and

144

145
only 4 percent in Village 2. Kerosene stoves had been
adopted by 8 and 27 percent, respectively, in Villages 1 and
2. This study has shed some light on the adoption process
and the different factors associated with adoption or non-
adoption of these technologies. Table 4.25 summarizes the
important variables in the adoption process for the two
villages. Households have generally responded positively to
the wood-saving and supply-enhancing technologies.

The technologies that appear to have a good potential
in two the villages and for the central Kenya region in
general are the Kenya Ceramic Jiko and the "Kuni Mbili”
stove/fireplace. The two technologies are improved
adaptations of widely used technologies and are therefore
easily appreciated. Their costs relative to other improved
technologies are also low. The Kenya Ceramic Jiko had an
adoption rate of 43 percent in Village 1 where average
household charcoal use per month (CCBM) was twice that of
Village 2. Although charcoal prices were low in Village 1
relative to prices faced by Village 2 households, the cost
of charcoal may have provided an incentive towards using
charcoal-efficient stoves. On the other hand, Village 2 had
an adoption rate of 27 percent for the "Kuni Mbili"
stove/fireplace compared to 15 percent in Village 1.
Households tended to respond by adopting technologies that
saved frequently-used fuels that were relatively expensive

or scarce. The adoption rates are encouraging considering

146

that the average adoption period is rather short. The Kenya
Ceramic Jiko had been in use for an average of 3.3 and 1.4
years for Villages 1 and 2, respectively. Similarly, the
"Kuni Mbili" stove/fireplace had been in use for an average
of 3.3 and 3.8 years in Villages 1 and 2, respectively.

Electricity does not appear to hold a great promise in
substituting fuelwood for cooking and heating. Distribution
and use costs are likely to be major limitations to its
adoption. In the development of rural infrastructure, the
government policy of accelerated rural electrification may
still have a positive contribution. Industries and
institutions wishing to locate in rural areas often need
electricity. Such facilities positively contribute to rural
income and development. The adoption of new technologies is
likely to improve as household resource endowments increase.

Kerosene was used for cooking and heating by 27 percent
of the households in Village 2 and 8 percent of the
households in Village 1. Existing and proposed subsidies
may therefore benefit some households. It is necessary to
understand how regularly kerosene is used for cooking and
heating before it can be justified as a valuable
alternative. As indicated elsewhere in this study, kerosene
subsidies pose serious equity questions. Forest depletion,
soil degradation and watershed destruction due to excessive

cutting for fuelwood and other forest products are

147
nevertheless serious problems for Kenya. It is therefore

important to explore all possibilities.

E J' I J' !'

The technologies discussed in this study have been
supported through different government policies with varying
degrees of success. Research results indicate that
important policy variables include: extension services,
income growth, improved access to different energy
alternatives, area of farm-land owned, leadership role,
household size, years of formal education received by head
of household, and locational characteristics. Policy
choices include: improving information flows for new
technologies, improving linkages between private and public
sector efforts, improving the efficiency of supply- and
demand-side interventions, developing better feed-back
systems, and encouraging household to change energy
consumption patterns.

The probit models evaluated indicate that the
availability of technology information through various
channels is crucial. This means that extension services,
meetings, agricultural shows, and mass media information
flows need to be improved. Local meetings and radio are
good avenues for conveying information on efficient wood-
energy strategies since they appear to be accessible to most

small-scale farmers. Agricultural shows and field visits to

 

148
farmers’ training centers are likely to have greater
potential if they were accessible to more farmers. These
two information sources can directly address the wood-energy
question and provide demonstrations of more efficient wood-
energy technologies. Subsidized field trips and lower gate
fees for small-scale farmers may improve awareness of these
technologies. Efforts can also be made to ensure that
extension agents are well prepared with this knowledge so
that they can disseminate it.

Though not examined quantitatively in this study,
another helpful intervention would be for the Ministry of
Energy and private producers of efficient stoves to work
together in supporting and training stove producers and
retailers in rural retail outlets. Such producers and
traders can easily advise fellow village residents on the
merits of these technologies and also sell them. This would
complement the work of extension agents. The on-going
program of training artisans from rural village polytechnics
(vocational schools) should also be intensified. The stove
prices can be lowered if they are produced locally thereby
reducing transport costs considerably. As more artisans
become available, more of these technologies would be
produced and disseminated. The costs of the Kenya Ceramic
Jiko and the "Kuni Mbili" stove/fireplace do not appear

prohibitive.

149

Demand- and supply-side policies need to be pursued
concurrently in addressing wood-energy problems in the rural
areas of Kenya. Literature indicates that much effort has
been invested in increasing wood-energy supplies. With
adequate public and private sector investment levels, new
fuelwood and charcoal saving or substituting technologies
can make significant contributions in the future. As the
country develops, more energy alternatives will become
available. Large-scale adoption of efficient technologies
can save forests and delay certain environmental
degradation. I

Agroforestry approaches to land use are important in
increasing fuelwood supply. Small-scale farmers can
conveniently adopt agroforestry. Farmers faced with serious
wood shortages tend to be more interested in tree planting.
Agroforestry enables them to enjoy the benefits derived from
multi-purpose tree crops grown in association with other
agricultural products. In this study, results indicate
there is good agroforestry potential among small-scale
farmers. Well-funded tree nurseries and extension agents
are likely to play an important role in the adoption of
agroforestry practices.

The Ministry of Energy or another agency should develop
an appropriate system of collecting feedback from new
technology users. Such a system would provide useful

information for future planning and improvements in new

150
technology deSigns. An interesting observation made by some
Kenya Ceramic Jiko users in Village 1, for example, was that
the stove tended to lead to quick deterioration of the
bottom part of aluminum cooking pots due to the intense and
focused heat that pots are exposed to during cooking.
Better information flow between researchers, stove
manufacturers and users would be very helpful in design and
use of such stoves. In that process, problems which may
hinder adoption can be detected and rectified in the early
stages.

Changing the energy use patterns of households can be
improved through better public awareness programs. Better
marketing programs for new technologies is very important.
In Village 2, some households reported that stoves such as
the Kenya Ceramic Jiko were not available locally. A better
distribution network may significantly improve adoption.

In general, low-cost strategies that directly address
household needs in specific regions of the country are
likely to be more widely adopted than more expensive ones.
Programs directed towards economic development and the
reduction of population growth rates in the country are also
useful in facilitating adoption. Higher incomes increase
the ability of households to purchase new technologies.

The population growth rate in Kenya during the last
three decades has averaged 3.3-4.1 percent. This high

population growth rate has continuously reduced the impact

151

of gains in economic growth. A higher population puts a
strain on the nation’s resources and also exerts greater
pressure on the land. Increasing rural energy needs under
these circumstances can lead to rapid cutting of woody
vegetation. The end result is environmental degradation,
reduced productivity and poverty. Effective wood-energy
production and consumption strategies can alleviate the
pressure on the natural resources of Kenya while a search
for long-term solutions continues. The different policy
variables discussed can be manipulated to improve wood-

energy use, supply, and availability of substitutes.

BMW

A lot of research work is still needed to effectively

deal with the wood-energy shortages in Kenya. Policies have

to be examined within the context of rapidly changing socio-

economic situations. Land parcels will continue to decrease
as population grows. There is need for a broader study
linking the use of different technologies to supply and

demand situations in different regions of the country.

Other research needs include: regular studies to capture the

temporal dimensions of adoption, developing effective
extension techniques, better definition of research
variables, evaluating energy needs and adoption strategies
of poorer villages, and evaluating potential adoption of

improved charcoal conversion kilns.

152

Comprehensive studies of selected villages at regular
time intervals would provide useful longitudinal information
which would be very helpful in understanding the diffusion
process for the strategies discussed. The average duration
of use for wood-saving technologies was low. Regular
follow-up over time would therefore supply new and useful
information on adoption profiles and trends.

A focus group type of study dealing with opinion
leaders in different villages in Kenya would be useful in
evaluating the technology knowledge base for larger parts of
central Kenya or similar regions in other parts of the
country. Such a study can show what the general production
and consumption trends are. A comparison of such studies
with village specific results can provide additional
insights on the wood-energy problem and relevant technology
for rural households in Kenya

Providing better extension services emerged as an
efficient way of increasing the likelihood of adoption for
many of the technologies. Extension agents in the field
tend to focus their efforts on revenue generating
activities. It is necessary to ensure that the most
effective communication techniques are developed and used.
Farmers can adapt to wood-energy shortages more effectively
if they know all their options.

The level of education among the heads of households in

the study areas is low and extension information must be

 

 

153
simplified to capture their attention. Posters and other
literature should be written in local languages or Swahili
to enable as many farmers as possible to acquire information
about these technologies. Research on how to effectively do
this would be very helpful.

In examining attributes of different strategies (e.g.
costs, perceived safety), information was found to be
inadequate for comparative analyses. Different technologies
were sometimes preferred for reasons such as tradition. In
other cases, farmers did not give any indication as to why
they used certain technologies. After discarding samples
without adequate data, not enough cases were left for
comparative analyses. A future study can incorporate these
information concerns in the objectives and in the related
questionnaire. To address model specification problems,
additional variables such as perceived affordability and
relative advantages of specific regions need to be examined.

A relatively high proportion of the households surveyed
reported that they were still using (at least some of the
time) the same old strategies (e.g., the traditional metal
stove and the three-stone fireplace that were supposed to be
replaced). An evaluation of the extent of use for various
technologies is important. It was apparent in many cases
that new strategies were replacing older ones. It is needed
to establish whether new wood-energy technologies completely

replace older technologies or whether older ones are

154
retained for occasional use. For example, the traditional
metal stove may be retained for warming the house during
cold days while the Kenya Ceramic Jiko is used for cooking
most of the time. This information may help bring about
improving designs.

Local geographic and economic differences exist in the
study areas and they should be given due consideration in
applying results of this analysis. The two villages studied
generally represent the more productive parts of the high
agricultural potential region of central Kenya. At least
one village in the less productive parts of the region needs
to be studied. The less productive areas tend to have fewer
trees and are therefore likely to have unique wood-energy
adoption profiles. Many post-independence settlement
schemes for previously land-less households were established
in such areas.

An extension of the current study may look at the
factors associated with the adoption of different
combinations of technologies. A number of households had
adopted more than one technology. A study could be carried
out to establish whether the relative use rates for
technologies in use at any one time. The relative merits of
different technologies can then be better understood.

A study of the factors associated with the of adoption
of more efficient charcoal-making technology would also

complement the current study. Significant energy losses

155
occur in the charcoal-making process. The diffusion of
efficient charcoal-making kilns would complement the
diffusion of strategies discussed in this study.

In addition to filling some important information gaps,
this study has provided information that can be used to
improve adoption rates for desirable wood-energy production
and consumption strategies in high agricultural potential
areas of Kenya. The full potential of new technologies,
however efficient, will never be realized if adoption rates
remain low. A good understanding of the factors associated
with the adoption of these technologies is useful for Kenya,
and also for governments in Africa and elsewhere which are
interested in developing and implementing workable wood-

energy policies.

APPENDICES

156

Appendix 1. Variable Description

AC91

AGE

CCBM

CGAM

CRHM

DSHH

DUMMY

EDH

ELEU

EVPY

FIC

FUCC

FUCR

Area under cultivation in 1991 in acres

Age of the head of household in years

Area of land owned and managed by the household
Charcoal bags used by household per month
Cylinders of gas used by household per month
Crop revenue per year in Kenya Shillings
Headloads of crop residues used by household per
month

Dwelling structure occupied by household
(1sPermanent stone or brick house, 2=Timber house
with tin roof, 3=Mud walls with tin roof, 4=Mud
walls and grass thatched)

Dummy variable for village effect (1 or 0)
Education level of the head of household in years
Use of electricity in the household for cooking
and heating (1=Yes, 0=No)

Extension agent visits per year

Farm input costs per year in Kenya Shillings

Use of charcoal for cooking and heating (1=Yes,
0=No)

Use of crop residues for cooking and heating
(1=Yes, 0=No)

Use of electricity for cooking and heating (1=Yes,

0=No)

 

 

157
FUFW Use of firewood for cooking and heating (1=Yes,
0=No)
FUGA Use of gas for cooking and heating (1=Yes, 0=No)
FUKE Use of kerosene for cooking and heating (1=Yes,
0=No)
FUOT Use of other fuel for cooking and heating (1=Yes,
0=No)
FWHM Headloads of fuelwood used by household per month
GNDH Gender of the head of household (1=Male, 0=Female)
HSZ Household size (Includes persons generally bound

by ties of kinship who normally reside together
under a single roof or under several roofs within
a single compound and who share the community of
life in that they are answerable to the same head

and share a common source of food (CBS, 1981

P-22))
KCJU Use of the Kenya Ceramic Jiko (KCJ) (1=Yes, 0=No)
-KELM Liters of kerosene used per month
KESU Use of kerosene stove (1=Yes, 0=No)
KUMU Use of Kuni mbili fire place (1=Yes)
LEAD Leadership position in the village (1=Yes, 0=No)
LBAV Labor availability measured in hours
MTATT Attendance to at least one public meeting during

the previous year (1=Yes, 0=No)
MEMFEOF Member of household employed off-farm (1=Yes,

0=No)

NPPM

OTHM

OTHU

RHPD

SAVINGS

SDSU

SH10

TEXP

THHX

TIC

TIM

TMSU

158
Number of issues of newspapers read per month
Amount per month of other fuel used in specified
units
Other method used for cooking and heating (1=Yes,
0=No)
Hours per day spent by head of household listening
to the radio
Difference between total revenue and total
expenditure
Use of sawdust stove for cooking and heating
(1=Yes, 0=No)
Number of visits to agricultural shows during the
last ten years
Total expenditure in Kenya shillings
Total household expenditure in Kenya shillings
Time spent collecting one headload (or other
specified unit) of fuelwood in hours
Time the household has settled on the-land in
years
Use of the traditional metal stove for cooking and
heating (1=Yes, 0=No)
Land bought and privately owned (1=Yes, 0=No)
Land inherited but shared (1=Yes, 0=No)
Land inherited but not shared (1=Yes, 0=No)
Coded categories of land tenure structure (1=TN1,

2=TN2, 3=TN3)

TNP

TPHSlO

TSFU

159
Existence of tenure problems (land disputes)
(1=Yes, 0=No)
Household has planted trees on their farm during
the last ten years (1=Yes, 0=No)
Total revenue for the household in Kenya shillings
Use of three stone fireplace for cooking and

heating (1=Yes, 0=No)

 

APE

115C)

Appendix 2. UCRHS Approval Letter

MICHIGAN STATE UNIVERSITY

 

omaorvnmronm' MWOWDWI“
WMOIMGIAWATIM

July 8, 1991

Albert M. Nuangi
1647 J Spartan Village
East Lansing, MI 48823

RE: ANALYSIS OF 8000 ENERGY PRODUCTION AND CONSUMPTION STRATEGIES AMONG SMALL
SCALE FARMERS IN CENTRAL KENYA, IRB #91-328

Dear Mr. Mmangi:

The above project is exempt from full UCRIHS review. The proposed research
protocol has been reviewed by another committee member. The rights and welfare
of human subjects appear to be protected and you have approval to conduct the
research.

You are reminded that UCRIHS approval is valid for one calendar year. If you
plan to continue this project beyond one year, please make provisions for
obtaining appropriate UCRIHS approval one month prior to July 2, 1992.

Any changes in procedures involving human subjects must be reviewed by UCRIHS
prior to initiation of the change. UCRIHS must also be notifed promptly of any
problems (unexpected side effects, complaints, etc.) involving human subjects
during the course of the work.

Thank you for bringing this project to my attention. If I can be of any future
help, please do not hesitate to let me know.
Sincerely,

0 id E. Hright, Ph.D., Chat
University Committee on Resea
Human Subjects (UCRIHS)

DER/doe

 
  

cc: Dr. Larry Loafers

"51] 'm - Affine-saw Ann-15...! Op'mfimmuy Imuflmamm

161

Appendix 3. Domestic Energy Production and Consumption
Strategies Survey

Interview No. : ... ...... ......
Village : ............ ...
Interview date : ...............
Interviewer’s Name: ..... ..........

Checked by PI On : ...............

SQQIQEQQNQMI§_DAIA1
l. a. How many people live in this household? ..........
No. of children : ........ ..
Ages of the children: ....................... .....
No. of adult males : ..........
Ages of the adult males: ............... ..........
No. of adult females: ..........
Ages of the adult females: .......................
b. How many other independent households are
represented on your farm? ..............
2. a. Is the person answering this question male or
female? ......
[Enter 1 if Male, and 0 if Female]
b. What is the relationship of the person interviewed
to the head of the household? ................( e.g.
wife)
c. Does the head of household live on the farm most of

the year?............. [Enter 1 if Yes, and 0 if No]

5.

162
What type of dwelling structure is occupied by the
head of household? .............
[Enter 1 if Permanent (i.e. stone or brick house), 2
if Semi-permanent (i.e. timber with tin roof), 3 if
Temporaryl (i.e. mud walls with tin roof, and 4 if
Temporaryz (i.e. mud walls and grass thatched) ]
How many years of education do you have? ..........

How many years of education does your spouse have?

What is your age? ......Years

Does anyone in your household have any leadership

position in the village (e.g. a cooperative official, a

school committee member, women group leader, etc.)?

........... [Enter 1 if Yes, and 0 if No]

How much total land do you own? ...... acres or

...... hectares

(note: 1 ha = 2.47 ac. )

a.

Is farming land split into smaller tracts for other
members of the family? ............

[Enter 1 if Yes, and 0 if No]

How much land do they farm? ..... ......acres /
hectares

How many acres/hectares do you farm? .........
acres/ha.

How many acres/hectares are under cultivation this

year?...

 

 

9.

10.

163
Food crops = ........ ac./ha.
Cash crops = ........ ac./ha.
c. If food crops also serve as cash crops, what

percentage of the crops produced is sold?........

d. How many acres/hectares are under
Pastures? . ...... . ac./ha.
Woodlots? ........ ac./ha.
(Be certain to circle acres or hectares)
How long has your household lived on this land? .....
Years
How did you gain access to your land?
(Enter 1 if Yes, and 0 if No)
Bought and privately owned : .....
Inherited but still shared

Inherited but not shared : .....
Other : .....

SpSCify other: 00......00.000.000.000...

11. Have you had any disputes over your right to use the

12.

land recently? .........

[Enter 1 if Yes, and 0 if No]

Could you please give an estimate of your cash crop
and/or livestock yields and revenues since this time
last year( or for the most recent production year for

your crops and/or livestock)?

13.

14.

164
PRODUCT YIELDS REVENUES REVENUE

(Regular) Bonuses)

How much money (Kenya shillings) did you spend on these
farm
inputs since this time last year?
Agricultural labor : ....... . ......
Agricultural seeds : ..............
Fertilizers : ..............
Pesticides : ............ ..
Animal feed : ... ........ ...
Tree seedlings : ..............
Others : ...... . .......
How much money (KShs.) does your household spend on the
following items in a month (or year) ?
Food : ..............
School fees for your children : ....... . ......
(specify whether per month,
term, or year)
Other educational costs : .... ....... ..
(e.g. uniforms, books, etc.)
Medical expenses : .............

Transportation

 

15.

16.

17.

165
Others : .............

Specify important others: ...........................

Do your children help you work on the farm? .........
[Enter 1 if Yes, and 0 if No l

b. How many?‘ ...........

c. How much time (hours per day) during the school
term? ....... .

d. How much time (hours per day) during school
holidays? ........

e. How many paid workers did you use on your farm last
year? ........ ....

f. For how long did each of the laborers work on your
farm?

(i) ....... days per week/month/year

(ii) ...... "
(iii). ..... "
etc.

a. Do you sometimes work outside your farm for pay?
[Enter 1 if Yes, and 0 if No]
b. If yes, how often?
Days per week : .......... or
Days per month: ..........
What are the current wage rates for farm labor in your

area? Ksh/per day: .... ...... ....

 

 

18.

19.

20.

166
a. Are any family members employed off-farm? .........
[Enter 1 if Yes, and 0 if No]
b. If yes, how many? .........

c. What kind of professions are they engaged in?

d. How long have they been engaged in their

professions?

a. Are any family members involved in a trade other
than agriculture e.g. backyard carpentry, basket
weaving, shopkeeping etc.? ...........

[Enter 1 if Yes, and 0 if No ]
b. What is the name of their trade: ......... .......
c. What is their monthly income from the trade?

She. ..... O O

DQHESIIQ_ENEBG1_2BQDHQIIQN_AND_QQN$UHEIIQH_DAIAL
What fuel do you usually use for daily cooking?

[Place a 1 in the appropriate spaces, Enter 0 for all

others]

Always Sometimes Never
Fuelwood ....... ......... .......
Kerosene ....... ......... .. .....

Charcoal 0...... 00.0.0... 0......

 

 

21.

22.

23.

Electricity ....... ......... .......
Gas (LPG) . ..... . ... .............
Crop residues ....... ......... .......
Others ....... ......... .......

Was the fuelwood availability better or worse for your

household
one year ago? ...... .....
five years ago? ..... ......
ten years ago? ...........

[Enter 1 if Better, and 0 if Worse]

What quantities of the following fuels are used in your
household per month?

Fuelwood ...... ..... ......headloads/stacks/other unit
Kerosene ................. litres

Charcoal ................. bags

Gas ................. cylinders

Crop residues............. headloads

Electricity ............. kilowatt hrs/last months
bill

Others .............
a. Do you use any of the following for cooking or

heating?
[Interviewer- Enter a 1 if Yes, or a o if No in the

appropriate spaces below]

168
Currently Duration of Used in

use in years. other

household
Three stone fireplace : .... .......... ....
Traditional metal stove: .... ... ....... ....
Improved Jiko e.g.KCJ : .... ...... .... ....
Electricity : .... .......... ....
Kerosene stove : .... .......... ....
"Kuni Mbili" Fireplace : .... .......... ....
Saw-dust stove : .... .......... ....
Other : .... .......... ....

b. If there are other independent households on your
farm, please indicate whether they use any of the
above cooking and heating methods to the best of
your knowledge.

[Enter 1 if Yes, and 0 if No in column 3 above ]
24. The following are possible reasons for choosing a

particular energy consumption strategy:

(1) it would be an improvement over my current
practice,

(2) it is easy to use,

(3) it is an acceptable method in the village,

(4) it can be easily tried and dropped if found
unsuitable, and

(5) I have seen it being used and believe it would

serve me well.

169
Are any of the above reasons important to you in
deciding whether or not to use any of the following?
[Interviewer- Circle the appropriate reason code(s)

next to the alternative methods]

Kuni Mbili Fireplace : .1. .2. .3. .4. .5. .6.

 

Improved stove (KCJ) : .l. .2. .3. .4. .5. .6. r-
Kerosene stove : .1. .2. .3. .4. .5. .6.
Electric cookers : .1 .2. .3. .4. .5. .6. 5%

Three-Stone Fireplace : .1. .2. .3. .4. .5. .6.

Traditional metal stove: .1. .2. .3. .4. .5. .6.

Please specify any other reason(s) that are important to
you. [Interviewer- Please note down the reasons in the

space below]:

25. To the best of your knowledge, what are the current
costs in KShs. of:
[Interviewer- Please indicate "Don't Know" for
situations where respondents can not make reasonable
guesses]
Traditional metal stoves: .......
Improved Jiko e.g. KCJ : .......

Kerosene stoves : .......

 

 

170
Charcoal (e.g. per bag): .......
Electric cooker (1 plate) : .....
Electricity (e.g. accesories plus average monthly bill):
[Basic accesories: ........; Last month's bill:.......]
Kerosene (per litre) 1: ......
[Secondary data sources will be used to corroborate the
answers given here]
26. How much fuelwood do you use per
Week? : .........
Other specified period? : ....... ..
e.g. per month
(Interviewer-specify the units of measure e.g.

headloads, stacks, etc)

27. Who collects the fuelwood needed in the household?

[Interviewer-note down 1 if yes, and 0 if no]

Always Sometimes Never Don't Know
Wife: 0.0... 00.... 0.... 00....
Husband : ...... ...... ..... ......
Children: ...... ...... ..... ......

Other : ...... ...... ..... ......

171

28. Where do you collect your fuelwood?

29.

30.

[Interviewer- note down 1 if yes, and 0 if no]

Always Sometimes Never
Own land : ..................
Neighbor’s land: ...... ...... ......
Market : ..................
Public land : ..................

(Govt. Owned)

Others

#1

What are the current prices (in K.Shs.) for the
fuelwood you use?
[Interviewer-Please indicate the sources and units on

which cost is based?]

Fuelwood from own land : .....................
Fuelwood from neighbor's land: .....................
Fuelwood from government land: ..... ............. ...
Fuelwood from the market : ........ ........ .....

Fuelwood from other sources : .....................

How satisfied are you with the fuelwood supply from
your current sources?

[Interviewer- note (1) for Very Satisfied, (2) for
Satisfied, and (3) for Not Satisfied]

Own Land : .......

Neighbor's land: .......

 

172
Market : .......

Other Sources : .......

31. a. How much time do you use to gather and bring home a
headload (or other specified unit) of fuelwood?
b. If a cart, truck, etc. is used, how much time does
it take to

collect, load, and deliver the fuelwood? : .........

32. Do you have more or fewer trees on your farm than you
had
[Interviewer- note down (1) if more, and (0) if fewer]
1 year ago? : .........
5 years ago? : .........

10 years ago? : .........

33. a. What species of trees have you planted on your land
during the last 10 years?
[Interviewer- Please note down the species

mentioned. Local tree names are acceptable ]

b. Where do you do most of your tree planting?

Enter 1 if in woodlots, 2 if along borders/hedges, 3 if

34.

35.

173
among crops, and 4 if others.

a. If you have planted trees during the last ten years,
how important were the following needs to your
decision to plant?

[Interviewer- please note (1) for Very Important,

(2) for Important, and (3) for Not Important]

Fuelwood : ....
Poles : ....
Posts : ....
Charcoal : ....
Rafters : ....
Timber _ : ....
Soil conservation : ....
Fodder : ....
Others : ....

b. Please rank the above uses of trees from the most

important one first to the least important one last.

If you have not planted trees during the last ten years,
were any of the following reasons important in that
decision?

[Interviewer- note down (1) if yes, and (0) if no]
Already have enough trees : ....

Seedlings are not available : ....

 

 

36.

37.

38.

174
Tree nurseries too far away : ....

Dont know where to get seedlings

Suitable species not available

Not enough land

Not enough labor : ....
Other ( Please specify) : ....
a. Is it necessary to consult other members of your

household before you can decide where to plant
trees? .......
[Please note down (1) if YES, and (0) if NO]

b. If the above answer is YES, which members of the

household are consulted? ........

[Please note down (1) if Male, and (0) if Female]
c. Do you do the same consultations for tree cutting
decisions?.....
[Please note down (1) if Yes, and (0) if No]
Are you required to seek permission from public
officials before you can cut your trees? ......
[Please note down (1) if Yes, and (0) if No]
From around this time last year, have you attended a
chief's (or any other) meeting in which tree planting
activities were discussed? ......

[Please note down (1) if Yes, and (0) if No]

175

39. How many agricultural shows (fairs) have you attended
during
The last 1 year : ....
The last 5 years : ....

The last 10 years: ....

40. Do you have access to any of the following sources of
information?

[Please note down (1) if Yes, and (0) if No]

Radio : ....
Television : ....
Newspapers : ....

Extension agents: ....

41. How often do you listen, watch, read or come into

contact with the following?

Radio (e.g. hours per day) : ..........
Television (e.g. hours per day) : ..........
Newspapers(e.g. Issues per week or month) : ..... .....
Extension agents (e.g. visits per month) : ... .......

42. a. If you have planted trees recently, have any of the
following sources of information positively
influenced your decision?

[Please note down (1) if Yes, and (0) if No]

 

 

43.

176
Fellow Farmers : ......
Extension workers : ......

Agricultural shows, meetings, field days etc: ......

Radio : ......
Television : ......
Newspapers : ......

b. Have any of the above sources of information
negatively influenced your decision to plant trees?
[Enter 1 if Yes, and 0 if No]

c. If the answer to (b) above is yes, please state

which one(s).

How do you think economic well being in this village

will change in the next ten years? ...........
[Enter 1, 2, or 3 for the answers below]

1. Things will remain the same.

2. Things will be better.

3. Things will be worse.

 

 

177

Appendix 4. Variables with High Correlation Coefficients in

 

 

Village 1.
Variables‘ Correlation Coefficient
CREV, SAVINGS 0.758
CREV, THHX 0.559
TREV, CREV 0.975
TREV, THHX 0.562
TREV, SAVINGS 0.792
TREV, CCBM 0.669

CREV, CCBM 0.711

 

 

 

 

 

178
Appendix 5. Variables with High Correlation Coefficients in
Village 2.

Variables‘ Correlation coefficient
FIC, CREV 0.739
FUGA, CREV -0.740
KELM, CREV ,. 0.686
SAVINGS, CREV 0.756
TEXP, CREV 0.724
THHX, CREV 0.547
TREV, CREV 0.973
FUCR, CRHM -0.602
NPPM, EDH 0.536
TEXP, FIC ' 0.876
TIC, FIC 0.561
TREV, FIC 0.742
TREV, FIC -0.679
TREV, KELM 0.707
SH10, KELM 0.642
TVHPD, NPPM 0.607
TREV, SAVINGS 0.792
THHX, TEXP 0.891
TREV, TEXP 0.728
TVHPD, SH10 0.510
TREV, THHX 0.551
TN1, TIM 0.681
TN2, TIN ‘ -0.537
TN3, TN1 -0.902
TN3, TIM -0.627

 

‘Variables with relatively high correlation (above + or -
0.50). These variable combinations were not used together
as independent variables in a single probit model.

179

Appendix 6. Coefficients for Model KCJC

 

Variable‘ Coefficient Std error t-value Elasticity”

 

 

CONSTANT -2.3195 0.5356 -4.331

AGE -0.0345 0.010 -3.630* -0.006
THHX 0.5E-04 0.9E-05 -3.630* 0.8E-05
MEMFEOF -0.0812 0.2458 5.668* -0.013
KELM 0.0854 0.0187 4.555* 0.014
AREA -0.0309 0.0130 -2.372* -0.005
TIC 0.2065 0.518 3.985*, 0.033
DUMMY 2.1023 0.2981 7.052* 0.335

 

Log-likelihood = 84.419

*Significant at a = 0.05

-2LLR (CHI SQ 7 DP) =160.28

CHI-SQ TAB = 16.013

‘ See appendix 1 for definitions of variables

” Elasticity computed at mean value of independent
variable.

 

180

Appendix 7. Coefficients for Model KUMC

 

Variable‘ Coefficient Std error t-value Elasticity"

 

CONSTANT -1.1744 0.1796 -6.54

CRHM 0.0274 0.009 3.051* 0.007
LEAD 0.4034 ’0.1735 2.325* 0.108
SH10 0.0393 0.0252 1.562 0.011
CGAM -0.3882 0.4334 -0.896 -0.104
DUMMY -0.2181 0.1986 -1.098 -0.058

 

Log-likelihood = 142.23

*Significant at a = 0.05

~2LLR (CHI SQ 5DF) = 23.984

CHI-SQ TAB = 12.833

‘ See appendix 1 for definitions of variables

” Elasticity computed at mean value of independent
variable.

 

 

181

Appendix 8. Coefficients for Model KESC

 

Variable‘ Coefficient Std error t-value Elasticityb

 

CONSTANT 0.0741 0.008 0.159

AGE -0.0200 0.008 -2.581* -0.005
EVPY 0.0227 0.465 2.694* 0.005
EDH -0.0578 0.026 -2.226* -0.013
RHPD 0.0828 0.040 2.059* 0.019
DUMMY -0.5230 0.200 -2.623* -0.119

 

Log-likelihood = 123.17

*Significant at a = 0.05

-2LLR (CHI SQ 5 DF) = 41.917

CHI-SQ TAB = 12.833

‘ See appendix 1 for definitions of variables

” Elasticity computed at mean value of independent
variable.

 

 

182

Appendix 9. Coefficients for Model TPHC

 

 

Variable‘ Coefficient Std error t-value Elasticityb
CONSTANT -l.672 0.2803 -5.966

EVPY 0.0112 0.0081 1.388 0.003
HSZ 0.1091 0.0328 3.322* 0.132
AREA 0.0333 0.0150 -2.215* -0.017
SH10 -0.0353 0.0261 -1.350 -0.006
TREV 0.3E-05 0.3E-05 1.136 0.112
RHPD 0.0364 0.0376 0.969 0.006
DUMMY 1.0845 0.1854 5.850* 0.212

 

 

Log-likelihood = 168.90

*Significant at a = 0.05

-2LLR (CHI SQ 7 DF) = 45.958

CHI-SQ TAB = 16.013

‘ See appendix 1 for definitions of variables

‘ Elasticity computed at mean value of independent
variable.

 

183

REEERENQES

Anonymous. 1989. "The Kenya Power and Lighting Company".
South_Magazine1 February 1989-’

Akinola, A. A. 1987. "An Application of Probit Analysis to
the Adoption of Tractor Hiring Service Scheme in

Nigeria" Oxford_Agrarian_Studies (16): 70- 82-

Aldrich, J.A. and C.F. Cnudde. 1975. "Probing the Bounds
of Conventional Wisdom: a comparison of regression,

probit, and discriminant analysis”. American_§gnrna1_gfi
Bolitical_Science XIX(3)=571-608-

Aldrich, J.A. and F.D. Nelson. 1984. Linear Probability,
Logit, and Probit Models. A Sage University Paper No.
45. Sage Publications, Inc. 91 p.

Allen, J.C. and D.F. Barnes. 1985. ”The Causes of
Deforestation in Developing Countries". Annals_gf_the

American_Aasociation_ef_seegranhera 75(2)=163-194-

Amemiya, T. 1981. "Qualitative Response Models: a
survey”- lonrna1_Qf_Echomic_Literature (19): 1483- 1536-

Avila, M. 1990. "Agroforestry Research Goes to the Farm".
Aaroforeatrx_moday 2(3):1o-12.

Baker, R. 1986. "Environmental Degradation in Kenya: two

conflicting explanations". African_£nyirgnmgnt
V(1-2-3):l8-43.

Bagi, F.S. 1983. "A Logit Model of Farmers' Decision About

Credit". Southern_I9urna1_of_Agricultura1_Ecenomica
(15):13-19.

Beggs, S., S. Cardell, and J. Hausman. 1981. "Assessing
the Potential Demand for Electric Cars". ignrna1_gf
Econometrics (16)=1-19-

Bhatia, R. 1987. ”Energy Demand Analysis in Developing

Countries: a review". Ih§_Energy_1gnzna1 Special L.D.C.
Issue. (8):1-33.

Borland International. 1990. Paradox Version 3.5. Scotts
Valley, CA: Borland International, Inc.

Borland International. 1991. Quattro Pro Version 3.0.
Scotts Valley, CA: Borland International, Inc.

 

 

184

Bradley, P.N.; N. Chavangi, and A. Van Gelder. 1985.
”Development Research and Energy Planning in Kenya”.

Ambig 14(4-5):228-236.

Bradley, P. N. 1988. "Methodology of Woodfuel Development
Planning in the Kenyan Highlands". Jon:na1_gfi

Biggeggznnhx (15):157-164.

Buck, L.E. 1980. Aspects of Woodfuel Supply and Demand in
Kenya. Discussion Document. Stockholm , Sweden:- The
Beijer Institute. 11 p.

Capps, Jr. 0. and R. Kramer. 1985. "Analysis of Food Stamp
Participation Using Qualitative Choice Models".

American_J9urnal_of_Agricultural_Ecgnomica (67): 49- -59.

Cochran, W.G. 1977. Sampling Techniques. Third Edition.
New York, NY: John Wiley and Sons. 428 p.

Cochran, W.G. 1983. Planning and Analysis of Observational
Studies. New York, NY: John Wiley and Sons. 145 p.

Debertin, D.L.: A. Pagoulatos, and E.D. Smith. 1980.
"Estimating Linear Probability Functions: a comparison
of approaches".

Economics 12(1):65-69.

Dennis, D.F. 1990. "A Probit Analysis of the Harvest
Decision Using Pooled Time series and Cross-Sectional
data". Journal__of_znx1ronmental_Economics_and
Management (18):176-187-

Dillman, D.A. 1978. Mail and Telephone Surveys: the total
design method. New York, NY: John Wiley and Sons.
325 p.

Dixon, J.A. 1986. "Selected Policy Options for Fuelwood
Production and Use in the Phillipines". in: Dixon,
J.A. and M.M. Hufschmidt (eds.) 1986. Economic
Valuation Techniques for the Environment: a case study
workbook. Baltimore and London: The Johns Hopkins
University Press. pp:163-l77.

Downs, G.W. and L.B. Mohr. 1976. ”Conceptual Issues in the
Study of Innovation".
(21):?00-714.

Engelhard, R.J. 1992. "Wood Energy Policy Development:
lessons from Kenya". in: Nemetz, P.N. (ed.) 1992.
Emerging Issues in Forest Policy. British Columbia,
Canada: University of British Columbia Press. pp:497-
504.

 

 

185

FZKL 1989. Forestry and Food Security. Forestry Paper No.
90. Rome, Italy: FAO. 128 p.

JFAKL 1990. "The Impact of Fuelwood Scarcity on Dietary
Patterns: hypotheses for research”. Unasylya (41):29-
34.

FESD Newsletter. 1990. (Forestry Extension Services
Division, Ministry of Environment and Natural
Resources). Nairobi, Kenya.

Finsterbusch, K., L.G. Llewellyn and C.P. Wolf, (eds.).
1983. Social Impact Assessment Methods. Beverly
Hills, CA: Sage Publications. 318 p.

IFoley, G. 1986. "The Economics of Fuelwood Substitutes: an
assessment of the impact of domestic fuelwood demand of
replacing fuelwood with conventional fuels". Unasylya
38(151):12-20.

Foley, G. 1992. "Renewable Energy in Third World
Development Assistance: learning from experience".

Enermx_zolicx (April)=355-364.

French, D. 1986. "Confronting an Unsolvable Problem:

Deforestation in Malawi". flgz1§_ngy§19pn§nt
14(4):531-540.

G.K. (Government of Kenya). 1979. Kenya Population Census
Vol.1. Nairobi, Kenya: Government Printer.

G.K. (Government of Kenya). 1979. anpgndium_tg_yglnm§_1
1222.292u1ation_9ensus. Nairobi. Kenya: Government

Printer.

G.K. (Government of Kenya). 1980. The Rural/Urban
Household Energy Consumption Survey: October-November,
1978. Nairobi, Kenya: Government Printer.

G.K. (Government of Kenya). 1986. Economic Management for
Growth: Sessional Paper No. 1 of 1986. Nairobi,
Kenya: Government Printer.

G.K. (Government of Kenya). 1989. National Development
Plan 1989-93. Nairobi, Kenya: Government Printer.

G.K. (Government of Kenya). 1991. Economic Survey 1991.
Nairobi, Kenya: Government Printer.

 

 

186

Gottschalk, R.J. 1982. Innovation Adoption Decisions in
Organizations: an empirical investigation. Unpublished
M.A. Thesis. Department of Psychology. Michigan State
University. East Lansing, MI. _138 p.

Grasa, A.A. 1989. Econometric Model Selection: a new
approach. Dordrecht: Kluwer Academic Publishers.
260 p. .

Greene, W.H. 1990. Econometric Analysis. New York, New
York: Macmillan Publishing Company. 783 p.

Greene, W.H. 1992. LIMDEP User's Manual and Reference
Guide Version 6.0. Bellport, New York: Econometric
Software, Inc. 473 p.

Gunderson, M. 1974. "Retention of Trainees: a study with
dichotomous dependent variables". lgnzngl_gfi

Econometrics (2):79-93.

Hall, D.O., F. Rosillo-Calle and P. de Groot. 1992.
"Biomass Energy: lessons from case studies in

developing countries". Eng;gy_£gligy (January):62-73.

Hammond, R. and P. McCullagh. 1978. Quantitative
Techniques in Geography: an introduction. Oxford:
Oxford University Press. 364 p.

Hausman, J.A. and D.A. Wise. 1978. "A conditional Probit
Model for Qualitative Choice: discrete decisions
recognizing interdependence and heterogeneous

preference". Econometric; 46(2):403-426.

Hodges, 0.6. and F.W. Cubbage. 1990. "Adoption Behavior of
Technical Assistance Foresters in the Southern Pine

Region". Eg:§§;_591engg 36(3):516-530.

Hyman, E. L. 1986. ”The Strategy of Decentralized
Production and Distribution of Improved Charcoal Stoves
in Kenya". in: Southgate, D.D. and J.F. Disinger.
(eds.) 1986. Sustainable Resource Development in the
Third World. Boulder, CO: Westview Press. pp:151-157.

IBRD (International Bank of Reconstruction and Development).
1980. Growth and Economic Development in Kenya.
Washington, D.C.: The World Bank. 213 p.

ICRAF (International Council for Research in Agroforestry).
1990. Strategy to the Year 2000. Nairobi, Kenya:
ICRAF. 76 p.

 

 

187

Johnston, R.J. 1980. Multivariate Statistical Analysis in
Geography: a primer on the general linear model. New
York, New York: Longman Inc.

Jones, D.W. 1988. ”Some Simple Economics of Improved
Cookstove Programs in Developing Countries". Resources

ang_Enezgy (10):247-264.

Joseph, S. and M. Kinyanjui. 1982. "Kenya's Wood/Charcoal
Cook-Stove Program". in:

Brocesdinga_of_the_negional
Workshon_on_Hood_Energx_1n_East_Africa Nairobi Kenya:
October 5-8.

Judge, G,G.: R.C. Hill, W.E. Griffiths, H. Lutkepohl, and
T.C. Lee. 1988. Introduction to the theory and
Practice of Econometrics. New York, New York: John
Wiley and Sons. 1024 p.

Juma, L. and S. Namuye. 1987. "The Evolution of Improved
Cookstoves in Kenya: a study in technological

articulation" Afr1can_Enxironmenta1_Studies_and
£1ann1n9_Bulletin V.VI(1-2)=207-220-

Kimuyu, A. P. 1990. "Urbanization, Economic Structures and

Demand for Fuels in Kenya" Eastern_Africa_Eoonomic
Sexism 6(2): 111-115.

Kish, L. 1965. Survey Sampling. New York, New York: John
Wiley and Sons. 643 p.

Kmenta, J. 1986. Elements of Econometrics. New York:
Macmillan Publishing Company. 786 p.

Kmietowicz, T. and T. Webley. 1975. "Statistical Analysis
of Income Distribution in the Central Province of

Kenya"- Eaatern_Africa_Economic_Bexiex 7(1)=1-25-

Kosmo, M. 1989. "Commercial Energy Subsidies in Developing

Countries: opportunity for reform”. Energy_2911gy
(June): 1989.

Latreille, P.L. 1992. "Unions and the Inter-establishment
Adoption of New Microelectronic Technologies in the
British Private Manufacturing Sector". Qxfigzd_fln11§tin

of_Economics.and.§tatistica 54(1):31-51.

Layard, P.R.G. and A.A. Walters. 1978. Migrggggngm1g_
Theory. New York: McGraw-Hill Book Company. 498 p.

Lee, L.K. and W.H. Stewart. 1983. "Land Ownership and the
Adoption of Minimum Tillage". American_Jgurna1_gf
Amricultural.£conomics (65):256-264

188

Macklin, B. 1984. "

". Unpublished Ministry of Energy
(Kenya) Manuscript. 19 p.

Maddala, 6.8. 1983. Limited-Dependent and Qualitative
Variables in Econometrics. New York: Cambridge
University Press. 401 p.

Magrath, W. 1984. "Microeconomics of Agroforestry". in:
K.H. Shapiro (ed.), Agroforestry in Developing
Countries. Center for Research on Economic
Development. University of Michigan. Ann Arbor, MI.
pp:77-143.

Mercer, D.E. and J. Soussan. 1992. "Fuelwood Problems and
Solutions". in: N. P. Sharma (ed.), Managing the
World's Forests: looking for balance between
conservation and development. Dubuque, Iowa:
Kendall/Hunt Publishing Company. pp:177-213.

Mitchell, R.G. and H.K. Preissler. 1991. "Analysis of
Spatial Patterns of Lodgepole Pine Attacked by the
Outbreak Populations of the Mountain Pine Beetle".
Forest_acience 37(5):1390-1408-

MOERD (Ministry of Energy and Regional Development). 1986.
"Agroforestry in Kenya, Institutional Development
Strategy, 1986-1988". Working Draft No. 1.
Unpublished Manuscript. 31 p.

MOERD. n.d. Kenya Renewable Energy Development Project
Booklet.

Ngugi, A.W. 1988. "Cultural Aspects of Fuelwood Shortage

in the Kenyan Highland8”- Journa1_of_819959granhx
(15):165-170.

Olson, K.P. 1982. Michigan's Statewide Forest Resources
Plan: a case study in innovation. Unpublished Ph.D
Dissertation. Department of Forestry. Michigan State
University. East Lansing, MI. 145 p.

Odingo, R.S. 1988. "Sustainable Agricultural Development
in Kenya and Competing Needs for Agricultural Land to
the Year 2000 AD”. in: Ominde, S.H. 1988 (ed.).
Kenya’s Population Growth and Development to the Year
2000 AD. Nairobi, Kenya: Heinemann. pp:66-86.

Onyebuchi, E.I. 1989. "Alternate Energy Strategies for the
Developing WOrld's Domestic Use: a case study of
Nigerian households' fuel use patterns and

preferences". Ing_zn§zqugnrnnl. 10(3):121-138.

189

Owino, F. 1982. "An assessment of Forestry Programs in
Kenya". in: °
. Nairobi, Kenya: October
5-8, 1982. pp:75-80.

Park, K.H. and P.M. Kerr. 1990. "Determinants of Academic
Performance: A Multinomial Logit Approach". Jon:na1_gfi

Economic_Education- (Spring Issue):101-111-

Peacock, D.L. 1972. The Adoption of New Agricultural
Practices in Northeast Brazil: an examination of farmer
decision making. Unpublished Ph.D Thesis. Department
of Agricultural Economics. Michigan State University.
East Lansing, MI. 340 p.

Pharo, J.A. 1982. Wood as an Energy Substitute.
Unpublished Ph.D Dissertation. Department of
Forestry. Michigan State University. East Lansing,
MI. 117 p.

Polson, R.A. and D.S.C. Spencer. 1990. "The Technology
Adoption Process in Subsistence Agriculture: the case
of cassava in southwestern Nigeria". Ibadan, Nigeria
(IITA): Eaner_2resentedIat_the_American_Aaricultural

egeu : ,ece ; e, ,1: go .. e_ 1 I; 1 °l__~

4:51

Poulsen, G. 1978. Man and Trees in Africa: three essays
on the role of trees in the African environment.
Ottawa, Canada: International Development and Research
Center (IDRC). 31 p.

Purnama, B.M. 1990. Rural Households Energy Consumption in
Central Java, Indonesia. Unpublished Ph.D
Dissertation. Department of Forestry. Michigan State
University. East Lansing, MI. 162 p.

Randall, A. 1987. Resource Economics: an economic approach
to natural resource and environmental policy. New
York, New York: John Wiley and Sons. 434 p.

Rauniyar, G.P. and F.M. Goode. 1992. "Technology Adoption
on Small Farms”. fig;1d_ngyglgpngnt. 20(2):275-282.

Rogers, E.M. 1983. Diffusion of Innovations. Third
Edition. New York, New York: The Free Press. 453 p.

Salem, B.B. and T. Van Nao. 1981. "Fuelwood Production in
Traditional Farming Systems". Unasylxa (33):13-18.

190

Scherr, S.J. 1989. The Legislative Context for
Agroforestry Development in Kenya. Reprint No. 56.
Nairobi, Kenya: ICRAF.

Scherr, S.J. 1989. "What Happens in Agroforestry
Development Projects?". Agrgfgr§§§:y_19day, 1(4):8-13.

Scherr, S.J. and E.U. Muller. 1990. "Evaluating
agroforestry interventions in extension projects".

Agroforestrx_$¥stema (11)=259-280-

Schultz, T.W. 1964. Transforming Traditional Agriculture.
New Haven, Connecticut: Yale University Press. 212 p.

Sellar, C.: J. P. Chavas, and J. R. Stoll. 1986.
"Specification of the Logit Model: the case of
valuation of nonmarket goods".

Economics_and_nanaeement. (13):382-390-

Shah, N.; P. Girard, C. Mezerette, and A.M. Vergnet. 1992.
"Wood-to-charcoal conversion in a partial-combustion
kiln: an experimental study to understand and upgrade
the process". -

. Butterworth-Heinemann Publishers.
(71):955-962.

Shaikh, A. and P. Larson. 1981. The Economics of Village
Level Forestry: a methodological framework. Washington
D.C.: USAID. 77 p.

Skog, K.E. 1986. Factors Influencing County Level
Household Fuelwood Use. Unpublished Ph.D Dissertation.
Department of Forestry. Michigan State University.
East Lansing, MI. 128 p.

Skutsch, M.M. 1983. Why People Don’t Plant Trees: the
socioeconomic impacts of existing woodfuel
programs-village case studies, Tanzania. Washington,
D.C.: Resources for the Future. Discussion Paper No.
D-73P. 58 p.

Spector, L.G. and M. Mazzeo. 1980. "Probit Analysis and
Economic Education". The_Iourna1_of_Economic_Education.
(Spring Issue):37-44.

Steinberg, D. 1986. PROBIT: A Supplementary Mbdule for
SYSTAT and SYSGRAPH. Evanston, IL: SYSTAT, Inc.

Survey of Kenya. n.d. Settlement Schemes' Map for Central
Province. Nairobi, Kenya.

191

Swinkels, R.A. and S.J. Scherr. 1991. Economic Analysis of
Agroforestry Technologies: an annotated bibliography.
Nairobi, Kenya: ICRAF. 216 p.

Teachman, J. D. and K.A. Polonko. 1988. "Marriage,
Parenthood, and the College Enrollment of Men and

Women".$gg1a1_£gzgg§. 62(2): 512- 523.

Teel, W. 1984. A Pocket Directory of Trees and Seeds in
Kenya. Nairobi, Kenya: Published by KENGO. 151 p.

Voh, J. P. 1982. "A Study of Factors Associated with the
Adoption of Recommended Farm Practices in a Nigerian

Village" A9ricultura1_Adm1nistration (9): 17- 29-

World Resources Institute. 1992. World Resources (1992-
1993): a guide to the global environment. New York:
Oxford University Press.- 385 p.

Wilkinson, L. 1990. SYSTAT: The System for Statistics.
Evanston, IL: SYSTAT, Inc.

Yambo, M. 1982. "Socioeconomic Aspects in the Development

and Diffusion of Fuelwood Technology: some theoretical
and methodological considerations". in: Proceedings_g£

Qctober.5:81_12821 Nairobi. Kenya- pp=81-92-

"‘uzumum