Content uploaded by Kidanemariam Gebregziabher Gebrehiwot
Author content
All content in this area was uploaded by Kidanemariam Gebregziabher Gebrehiwot on Aug 13, 2015
Content may be subject to copyright.
Extension participation, household income and income diversification: a system
equations approach
Kidanemariam G. Egziabher
1
*
a
, Erik Mathijs
b
, Miet Maertens
b
, Jozef Deckers
b
, and Hans
Bauer
a
a
Department of Economics, Mekelle University, P.O.Box Mekelle 451, Ethiopia
a
Department of Earth Science and Environmental , KU Leuven, Belgium
1.
PhD Candidate
Abstract
In contrast to substantial declines in budgetary support for extension services in Africa over the
last twenty years, the Ethiopian government invests heavily on agricultural extension services.
Despite this huge public investment, the attention given to evaluate the impact of the program is
very limited. Our study examines whether household access to agricultural extension affects
households‟ income diversification and income level. The study also examines the determinants
of farm households‟ agricultural extension participation decision. The data for this study derived
from 734 participant and non-participants sample households from eight villages, drawn from the
three agro-climatic zones of the Geba catchment, in northern Ethiopia. To analyze the survey
results, an instrumental variable (two stage least square model) and three stages least square
estimations were employed. These models take into account the sample selection bias associated
with access to agricultural extension service. The robustness of the models was validated by
various tests. Accordingly, both models resulted in significantly and positive impacts on income
(14.7%) and diversification level (12.7 points) for participant households. In addition to the
endogenous variable, exogenous variables such as the human, natural, physical and locational
capitals were found to be significant and important factors affecting the decision to participate in
an extension program, and increase household income and income diversification level. Despite
the positive impact of the program, the relative as well the absolute income level in the research
site is very low. Hence, introduction of more diversified farm technologies, non-farm income
generating activities and result oriented service delivery should be the direction of future
interventions.
Key words: Agricultural extension* income diversification* system equations * Ethiopia
I. Introduction
The Tigray region, in northern Ethiopia, is a predominantly agricultural economy where
agriculture contributes more than 50 percent to GDP and 85 percent to employment
(Tigray Bureau of Agriculture and Natural Resource Development-TBoANRD, 2003).
Due to population pressure and continuous land degradation and given the current
production technology, agriculture is reaching the limits of available natural resources.
Land-holdings per households have declined from 3.8 to 0.65 hectares over the last 30
years (ibid). Future increases in agricultural production and rural incomes will have to
come from agricultural intensification and diversification out of agriculture, rather than
area expansion or exploitation of additional natural resources.
Extension service is a process that facilitates the transfer of knowledge (technology) from
research centers to farmers and advises farmers in their decision making processes. The
idea behind providing extension service to farmers is that agricultural extension service
1
brings information, technology and improved farm practices into the farm household.
This will enable producers to manage their resources in the most rational and efficient
way so as to bring an overall farm productivity and ultimately improve their wellbeing
(Feder and Zilberman, 1986).
Since 1950s, various public extension programs have been introduced in Ethiopia.
Unfortunately, the impacts of all of these development interventions were not significant.
There were only a few changes such as the introduction of artificial fertilizer and the use
of improved seeds and animals. But in general productivity per unit area and the overall
welfare of the farming households participating in extension programs was not
significantly improved over the traditional and non-participant households (Abate, 2007).
The focus of the intervention was mainly on inefficient big commercial farms (1950-
1970) and state and collective farms (1970-1990), neglecting the small individual farms
which contribute to more than 90 percent of the cereal production. Whatever effort done,
was focused only on staple crops, marginalizing the other agricultural commodities, let
alone to consider non-farm income sources (Aredo, 1990, in Pausewang et al., 1990).
Previous extension approaches were top-down, non-tailored to the different agro-
ecological conditions and households‟ resource endowment, non-integrated (e.g.,
technology supply without credit; production without marketing etc). A new extension
program called Integrated Household Extension Program (IHEP) was introduced in 2003
(TBoANRD, 2003), at least by design, which tried to address the shortcomings of the
previous extension approaches. IHEP is a variant of the national agricultural extension
system called Participatory Demonstration, Extension and Training System (PADETES)
which is adapted to the regional conditions. The extension program has two main
objectives: increasing income and income diversification. Accordingly, the extension
program for the first time has covered services ranging from crop production, livestock
development, non-farm income generating interventions, marketing to infrastructural
development interventions and credit (Belay, 2003).
The objective of the extension program, as described above, in addition to increase in
income level, is to improve the income diversification level by increasing the sources of
income at the household level. In the literature of income diversification some try to
approach the concept in terms of motives, and others in terms of efficiency. The motives,
depending on the households‟ demographic and farm characteristics, could be survival
(push factors” driven by constraints and undertaken to manage risk, cope with shock) or
voluntary diversification, some times termed “pull factors” (driven by capability and
undertaken to manage risk) (Batterbury, 2001; Barrett, et al., 2001; Dercon.S and
Krishnan, 1996). Sometimes the cause for diversification can be said to maximize profit
(equalize the marginal rate of product substitution and thereby efficiency gains) or
minimizing the variance of income and risk reduction (Heady, 1952).
The term “income diversification” has been used by different authors to describe four
distinct but related concepts (Minot et al., 2006). The first definition of income
diversification refers to an increase in the number of sources of income or the balance
among the different sources (Joshi et al., 2003; Minot et al., 2006; Dercon, 1998). A
2
second definition of diversification concerns the switch from subsistence food production
to commercial agriculture (Delgado and Siamwalla, 1997). Third, income diversification
is used to describe expansion in the importance of non-crop or non-farm income
(Reardon, 1997) Finally, income diversification can be defined as the process of
switching from low-value crop production to high-value crops, livestock, and non-farm
activities (Minot et al., 2006).
The study has adopted the first income diversification definition and accordingly, six
broad income categories are established: crop, livestock, wage, business, migration, and
transfer incomes. Crop income includes all income coming from cereals, oil seeds, and
fruits and vegetables. Livestock income refers to any income generated by the household
from livestock by-product, livestock sells, or rental services (such as ox or pack animals
rental services). Wage income refers to all types of income derived from both agricultural
and non-agricultural wage employment, around someone‟s homestead. Migration income
consists of send back and bring-back income by migrant labor, but nationally. Any
income received by the household in the form of transfer from governments, non-
governments (food aid, cash or in kind) and friends/relatives is classified as transfer
income. Business income includes any income generated from a business activity run and
owned by the family located in their home or nearby towns. Finally, in this study
household is conceived as the social group which resides in the same place, shares the
same meals, and makes joint or coordinated decisions over resource allocation and
income pooling (Ellis, 1998).
A number of papers examined the concept of income diversification; mainly focusing on
characterization, entry barriers, determinants; farm-nonfarm linkages, poverty analysis
and different policy recommendations (Reardon, 1997; Ellis, 1998; Minot et al., 2006;
Woldenhanna, 2001). With the exception of two welfare studies (income growth, poverty
reduction and household expenditure) (Dercon et al., 2009 and Mogues et al., 2007),
there has been little careful analysis to study the impact of public policy intervention in
general and agricultural extension service in particular, on household income
diversification or income level in Ethiopia. However, the stated studies were not without
shortcomings.
Dercon et al. (2009) used an instrumental variables estimator to address the endogeneity
problem and extension in a continuous variable (frequency of visit by extension agent),
which is believed to show the intensity of extension effort, still it is not free from
biasness. Extension agent visit per se does not capture the intensity of extension service
provided to a household/ farm. Our extension participation variable is more accurate for
the following reasons. A household is said to be “participant” first, if the household
agrees to participate to adopt a given technology (package)-registered one. Second, she
has to participate in the training program designed/organized for new adopters of the
technology. Third, it must commit some resources and take credit to own the package and
Four, agree to host follow-up visits by the extension agent or other experts, which is
similar to frequency of visit by extension agent variable.
3
Hence, extension variable measured in terms of extension agent visits, which could
include in the visit non–fully committed farmers; the estimates of extension impact on
farmers‟ performance under such condition is likely to be biased downward. Moreover,
the extension visit extension variable fails to control the instantaneous farmer-to-farmer
information diffusion and the estimate of extension impact based on this approach may,
once again end-up with zero extension effect. But our approach minimizes the problem.
According to a review done by Dercon et al (2009), examining the impact of access to
agricultural extension, there is no study with the evidence of the direct impact of
extension on poverty in a developing country context. Whatever studies done to assess
the impact of agricultural extension focused mainly on farm productivity, level of farmer
knowledge and return to extension service. Some studied technology adoption and
income level and diversification (Sumberg et al, 2004; Jansen et al 2006). First, most of
the studies done are on the experience of Training and Visit (T&V) system and yet, the
impact of extension on farm performance was rather mixed and the results are too old
(Birkhaeuser,1991). Second, over the twenty years most developing countries budgetary
support for extension services is in dramatic decline, Ethiopia is of particular interest in
part because the government rejecting this trend is budgeting substantially on agricultural
extension (Spielman et al., 2010; Dercon et al., 2009). Despite this huge public
investment, the attention given to evaluate the impact of the program is very limited.
Thus, our study fills a gap to empirically examine the impact of agricultural extension on
income and diversification on marginal and semi-arid farming households. Given the
above theoretical discussions of household extension participation, and income
diversification the paper will try to answer the following questions. First, who are the
participants of the agricultural extension program? Extension participation is believed to
improve the income and overall welfare of the participant households. Hence, it is good
to understand whether the program is benefiting the relatively better-off households
measured in terms of access to land, livestock, main markets and sometimes active labor
force owners or not. Second, how are extension participant households performing in
terms of income and diversification compared to non-participants? This is important to
show the progress made towards improving the welfare of program participants and
assess policy efficiency. And it is important input for policy makers and other donor
agencies.
2. Background
According to Belay (2003) public agricultural extension work in Ethiopia is said to have
started since 1950s. In line with the development theories of that time, which accentuated
only the instrumental role of agriculture to supply resources for industrial development,
Ethiopia‟s agricultural sector in general and the small farms in particular was denied the
attention it deserves during the past (Aredo, in Pausewang et al., 1990). The small farms
which contribute to more than 90 percent of the national production had suffered from the
following problems. First, the subsistence farm was completely neglected in favor of
large-scale commercial farms (1950 to 1970s) and State and Collective farms (1970s to
1990s). Second, it was constrained by technology supplies due to the limited technology
generating capacity of research institutions and poor-research–extension linkages. Third,
4
the extension system was top-down, non-participatory, supply-driven and cereal
production dominated. These problems have caused the sector to perform poorly. As a
result, cereal production lagged behind the growth of urban and rural population and the
country, for the first time, become a net food importer (ibid).
In 1991 the T & V extension approach was adopted as a national extension system until
its replacement by the Participatory Demonstration and Training Extension System in
1995. The latter was adopted from the Sasakawa Global (SG 2000) extension strategy,
initiated in Ethiopia in 1993 by the Sasakawa African Association and Global 2000 of the
Carter centre (Gebremedhin et al., 2009). Based on the critical evaluation of the past
extension approaches and the experience of SG 2000 an extension approach called
PADETES was developed in 1995. Initially, PADETES promoted cereal production
using Extension Management and Training Plots (EMTP) usually half hectare on-farm
demonstration plots which were managed by farmers and used to train farmers and
extension workers on appropriate agro climatic and farm management practices (Belay,
2003). The beneficiaries were mainly those farmers who live in high rainfall areas of the
country (Gebremedhin et al., 2009).
The need for further decentralization in the extension service delivery system so as to
accommodate the needs of the different agro climatic condition forced PADETES to
undergo over the years. Given the general framework, regions were allowed to adapt the
extension system to fit to their own situations. Accordingly, Tigray region developed an
extension system called Integrated Household Extension Program (IHEP) and thence,
the content of the previous packages have been diversified to address the needs of
farmers who live in different agro-ecological zones of the country and landless farmers.
The IHEP includes service packages such as, improved seeds (for cereals, vegetables,
and fruits); livestock (dairy, fattening and poultry); water harvesting technologies;
beekeeping, and petty trade.
The technologies to be supplied were identified by farmers themselves through
participatory need assessment. Following the decision to participate in the extension
program, farmers will be provided a list of technology options to choose from. Farmers
are then familiarized with the technology of their choice through training at Farmers
Training Centers (FCT). Credit service is arranged to finance the cost of the technology.
Finally, to monitor the implementation and management of the new technology at the
farm level, regular visits (to give technical and market advice) are made by extension
agents or Woreda level experts.
3. Area description and data collection methodology
The Tigray Regional
a
National State is one of the nine states in the Federal Democratic
republic of Ethiopia, with an estimated area of 51560 km
2
(TBORAD, 2008). The
a
Region is an administration territory equivalent to Province. Woreda is the next administration
layer/stratum and equivalent to district. Tabia is the lowest government unit; sub-district.
5
altitudes in the region lie between 300 meters above sea level (masl) in the east to above
3000 masl in the north and central part. Hence, it covers three agro-climatic zones:
lowland (kolla) which falls below 1500 m.a.s.l; medium highland (woina dega) 1500 to
2300 m.a.s.l and upper highland (douga) 2300 to 3200 m.as.l. (ibid)
The regional population, according to 2008 census, is 4.6 million with an average annual
growth rate of above 2.6%. (CSA, 2007). The mean annual rainfall in the region varies
from 200 mm in the extreme east bordering Danakil Depression to over 1900 in the South
western part of the region. The typical farming system is mixed farming, subsistence
oxen plough single cropping cereal crop dominated (productivity below 1 ton per ha)
combined with livestock rearing. According to the agricultural sample census report, the
average farm size of holder in the region ranges from 0.6 in the Eastern part to 1.2 in the
western part of the region (with average holding of 0.65 ha.), and generally average land
holding decreases with altitude (TBORAD, 2008). As a result, the farm produce (mainly
cereal produce) is not in a position to support for more than 6 months for a household
with family size of 5-6 members. In drought years the people become fully dependent on
food aid. It is against this rationale the government is trying to diversify farm household
income both within and outside agriculture.
The research site is the Geba catchment , one of the catchments in the region, it covers
4600 km
2
area, 10 Woredas and 168 Tabias. To ensure representativeness the woredas
were clustered based on their agro-climatic condition. To reflect the contrasting agro-
climatic zones of the catchment we selected two Woredas from the mid-highlands, one
each from highland and lowland.Two Tabias were again randomly selected from each
Woreda.
Households were selected from Tabia based on population size and farmers‟ participation
status in the agricultural extension service. Accordingly, 390 participant and 344 non-
participant households were selected. Questionnaires were tested and validated before
survey work. However, like most empirical work in the social sciences which suffer from
the problem of missing data for different reasons, this study can not be an exception.
Simple statistical description, and instrumental variable and/or two-stage least square are
used to answer the research questions, test hypothesis and derive some conclusion on the
income diversification strategies of the rural communities in the research areas. The
variables used for the analysis are indicated in Table.1
4. Descriptive statistics
4.1 Household and Farm characteristics
Before moving to estimate the effects of agricultural extension participation on income
and diversification level, it is instructive first to consider simple descriptive statistics of
the two target groups in terms of the basic explanatory variables: household and farm
characteristics listed in Table 1. Comparing the two groups in terms of the household
characteristics, sometimes called human capital, the participant households consistently
do reveal some superiority over the non-participants. For instance participant households
6
are mainly male headed (80% for participants compared to 66% for non-participants),
have larger family size (6 versus 4.6 respectively), and marginally higher level of farm
experience (approximated by age of the head of the household). This indicates that
households with better human capital are more likely to adopt and participate in the
extension program.
Likewise, comparing participants and non-participant households in terms of their basic
asset holdings it is similar with the human capital endowments situations. They do have
34% higher land holding size (though land is a public property) and 51% TLU than non-
participants respectively. The same is true with access to irrigation. These natural
physical asset endowments coupled with the better human capital position is believed to
induce households to participate in the extension program.
Comparing participant and non-participant households in terms of access to off-farm
generating activities, the results are mixed. While participant households show higher
participation rate in wage employment and migration, on the other hand non-participants
show some dominance over participants in business and transfer income access. The high
participation rate of participant households in wage and migration is closely linked with
better access opportunities to extension participant households in the public work called
safety net employment program and high family size of participant households.
When we look at business and participation rate very closely, though non–participants
show high participation rate which is approximately 27% versus 22%; comparing the
two groups in terms of income shows the other way. The average business income for
participants is Birr 473 compared to Birr 367 (28.5% higher). This implies the non-
participant household which are relatively resource poor, do face entry barriers to better
earning business opportunities and as a result are mostly engaged in less remunerative
business activities. In terms of transfer income, as expected, the non-participants show
higher level of access to transfer income compared to participant households. This is
mainly that the transfer income which is constituted of food aid from government to poor
households mainly favors non-participants, who are relatively resource poor.
The dominance of participant households is also reflected in Edir membership, a type of
social capital. This shows that the better off households are relatively with better social
networking and access to information. The average distance to markets (both Woreda and
Mekelle) is greater for participants compared to non-participants. This shows, the strong
relationship between distance and basic assets holdings (land and livestock) and
extension participation.
4.2 Farm and non-farm income for participant and non-participant households
Comparing participants and non-participants in terms of the different incomes sources
and their respective contribution to the household income; participant households‟ do
show clear superiority over non-participants, Table 2. Participant and non-participant
household earn average of Birr 5387 and 4232 farm-income respectively. This indicates
that participants do earn 27.3% higher than non-participants and this is consistent with
7
resource endowments difference discussed above. Similarly, the amount of non-farm
income is Birr 2409 for participant and Birr 1979 for non-participant households.
4.3 Total household income and income diversification index quintiles.
Before we put our data into the empirical analysis and try to make some inference, it is
essential to classify or group, in a way that support the model estimation. Among the
many statistical methods for classification, the quintiles steps method which divides the
data set into predetermined number of classes with equal intervals is employed.
Accordingly, income and diversification index quintile is computed. The results are
indicated in Table 3.
When we compare the average share of farm income of the sampled household using
income diversification index quintiles the less diversified (first quintile) earns the highest
(almost 91 percent) the share keep falling continuously and reaches 48.5 percent for the
last quintiles. This income distribution seems strongly dictated by the high income
concentration coming from the crop income (7765 Birr). On the other hand the share of
non-farm income moves the opposite direction from 9.2% for the first category reaching
51.5% for the last quintile. In terms of mean income by income quintiles classification
the relatively richest quintile earns 9.1 times more than the first category. In terms of the
participants and non-participants distribution using income diversification index quintiles
the first quintiles is dominated by the non-participant and the last quintiles by the
participant group, showing relatively high level of diversification by participant
households, and vise versa.
5. Empirical Model
To examine the impact of agricultural extension service on household income and level
of income diversification (SID)
b
, using the variables described in Table 1, we use the
following model.
(1)
(2)
(3)
Based on economic reasoning and the extant literature on extension participation, income
and diversification (Dercon et al., 2009; Maertens, 2009; Sumberg et al., 2004;
Woldenhanna and Oskam, 2001; Rogers, 1984); variables are classified as either
exogenous or endogenous. Accordingly, Extenpart, total income and level of income
diversification index are endogenous and the remaining variables listed in Table 1 are
b
Income diversification is measured by the Simpson index of diversity. It is defined as:
Where
.
is the proportion of income coming from source . The value of always falls between 0
and 1. If there is just one source of income, =1, so . As the number of sources increases, the
shares declines, as does the sum of the squared shares, so that approaches 1
8
exogenous. We hypothesize that; participation in the agricultural extension program by
providing cash and relaxing liquidity constraints in agricultural activities: credit service
both in cash and in kind positively influences household income. Secondly, given the
objective and intervention of the extension program it is hypothesized that participation
in the extension program has a positive effect on income diversification. Hence, the
coefficient of interest are β and α. However, since the selection process into the extension
program is not random, the participant (treated) households are non-randomly selected
sample, thence giving biased results.
The sample selection problem may arise from (1) self selection where the households
themselves decide whether or not to participate in extension program, due to differential
resource endowments and/or (2) endogenous program placement where those who
administer extension program (such as village extension agent, Tabia Administrators etc)
select households with specific characteristics (relatively poor or reasonably wealthy). As
a result, extension participation is not normally distributed and our household survey data
clearly showed differences in terms of their basic covariates. To address the sample
selection problem in our estimation processes a large set of observable control covariates
are included; and used an appropriate estimation technique (Maertens, 2009; Imbens and
Angrist, 1995; Imai et al., 2010).
To see how data set fits to the different specifications, we first estimate the two models
using least squares (OLS). In the second specification, we included Tabia fixed effects to
account for potential differences in natural and physical characteristics among the
sampled Tabias. We employ the most commonly used technique, two-stage least squares
(2SLS), in the class of instrumental variables (IV) and to compare model results, three-
stage least squares estimators (Imbens and Angrist, 1995). The three Stage Least Squares
(3SLS) model adds a correction for heteroscedasticity to 2SLS by utilizing Generalized
Least Squares (GLS) methods. Moreover, since 3SLS uses all the information in the
system of equations to estimate parameters in each individual equation while 2SLS uses
only the information in the specific individual equation to estimate the parameters from
the corresponding equation, 3SLS is more efficient than 2SLS (Ahn, 2002). To get the
part of the treatment (participation) variable that is independent of the other unobserved
and/or uncaptured variables affecting the outcome, we first regressed the treatment
variable on the two instruments (number of extension agents, and lnlaggedTLU), the
other covariates in equation (2) and a disturbance term μ. The two instrument variables
are selected based on economic theory. The number of extension agents is instrumental
to increase the outreach areas or target beneficiaries and bring potential farmers into the
extension program and thereby affecting the instrumented variable. At the same time
lagged livestock asset holding does also influence participation decision but have less
influence over the current income and diversification level. In addition to the economic
reasoning, the instruments have qualified the relevant statistical tests (see Appendix A).
In the second stage, we estimate the two outcome variables of interest, log transformed
total income (lnTincome) and Simpson index of diversity (SID) by including the
predicted treatment variable (Extenpart) as an additional repressor, in addition to the
other covariates specified for the respective equation. In the three-stage least square
9
estimation, equation (2) and (3) are estimated. The control variables (X) include
household characteristics, asset holdings, distance and rural infrastructure variables. The
household characteristics are captured using household head age (Householdheadage),
sex of the household head (HHHeadgender), adult labor force supply (Adultlaborforce),
number of dependents (No.dependents), and household head schooling (Headschooling)
variables. The asset holdings of the household is represented by current log transformed
land size (lnlandsize) and log transformed livestock measured in tropical livestock unit
(lntlucurrent). A dummy variable for access to irrigation system (Accessirrigation),
distance to local (DistanceWoreda) and main market (DistanceMekelle) are part of the
control variables. Edir membership variable (Edirmemberships) which belongs to the
group of social capital is also included. Finally, we also included interaction terms
between land size and distance to main market and land size and TLU (landsize*tlu).
As reviewed by Birkhauser et al (1991) and Evenson (2001), a number of studies have
approached the impact of extension at household level (using extension agent visit as
participation variable). These studies have suffered from two basic estimation problems.
The first is the problem of endogeneity in the extension-farmer interactions. It is more
likely that the better off farmers desire and acquire information about new technology.
Extension agents themselves may also seek out contacts with better farmers who would
be good performers even in the absence of extension contacts (Everson, 2001). Under
such conditions, the extension participation variable is endogenous and the estimates of
extension impact on farmers‟ performance are likely to be biased upward.
The second source of potential bias is the problem of indirect or secondary information
flows, benefiting non-participant farmers. Once again, in such case, there may be no
difference in performance between participating and non-participating farmers leads to
erroneously lower extension effects.
The problem highlighted above can be handled or at least reduced econometrically using
two-stage procedures or simultaneous equation approaches and by specifying the
extension variable at a village level. However, this has rarely been done (Birkhaeuser et
al., 1991). In our context, with universal village coverage but partial coverage of
individual households, the second approach (extension variable at a village level) is not
feasible. Hence, to deal with the endogeneity problem in our context instrumental
variables two-stage least square and 3SLS equations have been employed.
In our case, the endogeneity problem is particularly relevant for some of the included
variables, such as household wealth status- livestock ownership, is likely to induce
extension participation and extension participation is like-wise expected to influence
household livestock holdings. To deal with the endogeneity problems arising from such
contemporaneous correlation among the error terms, Easterly and Levine, (1997) applied
the system equations approaches: instrumental variables/2SLS and 3SLS; to predict the
different parameters they want to examine.
Finally, in addition to the stated models to estimate the outcome variables, we used a
probit regression model which is an appropriate statistical tool to determine the influence
10
of independent variables on dependent variables when the dependent variable has only
two groups (dichotomous), e.g., participants and non-participants, and the explanatory
variables are continuous, categorical or dummy. Accordingly, a probit model has been
employed to explore the degree and direction of the relationship between dependent and
independent variables in the extension participation analysis.
6. Results and Discussions
6.1 Model Results
Following the identification of instruments, to verify the validity of the instruments, the
instruments are tested using Sargan test under the null hypothesis that the instruments
used are valid. The Sargan test resulted in a p-value of different from zero confirming
that the instruments used are valid. Further more our instruments are checked by a
quantitative test also known as tests of over identifying restrictions to see whether our
instruments pass test for weak instruments. Using the null hypothesis that all the
instrumental variables are uncorrelated with the residuals is accepted.
A likelihood ratio test of the hypothesis that all the slope coefficients were equal to zero
in the 2SLS regression was rejected. The chi-square statistic of 452.67 was greater than
27.68 critical value of the chi-square distribution, with degree of freedom equal to the
number of variable coefficients in the system at 1%. The Jarque-Bera normality test
indicates that the residuals of the models are normally distributed, implying that the
estimates obtained are not due to outliers in the data. The different test results are
indicated in appendix A.
Consistent with our prior hypothesis, most of the human capital variables (adult labor
force, dependant and house hold head schooling) show a positive and significant
influence on extension participation. The basic assets, land size (lnlandsize) and log
transformed lagged livestock ownership (lntlubefore1) went in line with our hypothesis.
Extenpart variable has a positive and significant influence on lnTincome and SID. The
results are consistent with our expectation.
6.2 Discussions
The first endogenous variable Extenpart in equation (1), is a dependent variable which is
a dummy variable with a value of 1 for participants and 0 otherwise. The second
endogenous variable is lnTincome, the natural logarithm of total income of the
household; constituted of crop, livestock, and non-farm income (wage earnings, business,
migrant labor earnings and transfer income) which is a continuous variable. The third
dependent variable is the Simpson index of diversity which measures the diversity in
terms of the number of income sources that a household has (Minot et al., 2006). The
other exogenous variables are as defined above.
Extension participation: The Probit model results for equation (1), characterizing
farmers‟ decision to participate in the extension program or not are given in Table 4. The
model was statistically significant at the 1% level. Correct classification of farmers as
either participating or not participating in the extension program, based on the
explanatory variable estimation is an important measure of goodness of fit. The model
11
correctly predicted 69% of the participants and non-participants. The (118.64) value
which tests the null hypothesis that all the explanatory variables are in fact non relevant,
is higher than the critical value for rejection (33.4087 for a 99% confidence level),
leading to a rejection of the null hypothesis. Most of the explanatory variables included in
the model, except for distance to main market, have the expected sign. Six variables at
the 1% of level, three at 5% of level and two at 10%, out of the sixteen initial
independent variables in the model contributed significantly to the decision to participate
or not in the extension program, and are discussed below.
The significant household labor endowment variables can be classified into household
family size (Adultlaborforce, No.dependent); and household head schooling. The
household size variables are positive and significant. Although, not as strong as the adult
labor (6.96%) force variable, dependents (4.47%) are still productive enough contributing
labor to the farming activities of the household. This suggests, that households without
labor constraints are more likely to participate in the extension program. This is
consistent with the labor intensive nature of the extension program, such as water
harvesting which requires adult labor and livestock packages which demands child labor
(herding). This supports previous results by Abdulai et al., (2008); Akobundu et al.,
(2004).
The positive and significant impact of household head‟s level of schooling on the
probability of joining the extension program is consistent with the notion that farmers
with better human capital are among the early adopters. The results also reveal a life-
cycle effect. First the probability of joining the extension program increase as age
increases (each additional year increases the probability at 1.95%), until it reaches its
maximum at 44 years, then starting to fall (-0.01%). These results, corroborate the
findings of previous works (Tiwari et al., 2008; Mendola, 2007; Feder et al., 1985 Feder
and Slade, 1984). However, our findings contradict the findings of Genius et al., (2006)
and others. Farming experience is expected to affect participation positively, but equally
younger farmers with longer planning horizons may be more likely to invest than older
farmer in new technologies and thereby inversely related with participation. Hence, the
impact of the farmers‟ age on extension participation and/or technological adoption is
less clear.
Ownership of land (wealth indicator), is commonly hypothesized to be related to
willingness to invest in new technology adoption. As expected, land size was found to be
positive and significant at 5% level of significance. Each additional unit of land was
found to increase the probability of participation in the extension program by 11.22%.
These results seem to affirm the important role of resource endowment in observed
adoption behavior (Adesina and Zinnah,1993). Certainly, farmers with large farms are
more likely to have more opportunities to learn about new farming practices. They are
also likely to have more incentives to adopt new technologies and are more able to bear
risks associated with the adoption of improved technologies (Adesina and Zinnah,1993;
Tiwari et al., 2008); Zepeda, 1990).
12
Distance to local market (Woreda) was positively related to technology adoption. The
positive influence of distance to local market on technology adoption decision,
contradicts the findings of most studies in the area; (Mendola, 2007; Gebremedhin et al,
2009; Barrett (2008). However, similar result was also found by Genius, et al (2006). On
one hand, remote areas are endowed with relatively larger land size and livestock
resources, which positively influence extension participation, through its wealth effect.
Moreover, remote areas mainly depend on agricultural extension service for their
information on agricultural technologies, hence increasing the probability of participation
(by 1%). On the other hand, remote areas do experience high transportation costs to reach
main markets. Hence, the positive relationship between participation and distance to local
market; implies the impact of wealth effect and the need for information is dominant over
the high cost of access to market, in the decision to participate or not.
Edir membership
c
is a variable type which belongs to the broader social capital group. Its
main activities are limited to self-help service to members during death of family
members. This self-help community group gives opportunity to incidentally discuss and
observe the practice of neighbors at no cost and time intensity, some times it is called
“passive information gathering” (Abdulai et al., 2008). The admission to such a group, in
the context of our rural research sites condition, does not depend on wealth. The payment
ranges from simple attendance to nominally low rate payment (Eth 0.25 -1.00
d
birr per
month) . The only driving force to join or not is, how household members value such
service. But it is instrumental for sharing information. Accordingly, Edir membership has
increased the probability of participation by 16.5% compared to non-participants. Similar
results were also documented by Tiwari et al (2008), and Zepeda (1990).
Household income level: The above discussion indicated that asset holdings and village
characteristics are the key among other things influencing whether the household is
able/willing to participate or not in the extension program. Extension participation
through access to credit and technical backup of the village level extension worker to
participating households is expected to increase income level. To determine the effects of
the different exogenous variables specified before and, predicted value of extension
participation (Extenpart) on household income (lnTincome) equation (2) the instrumental
variable two-stage least square (2SLS), and three stage-least square models are estimated.
Since Extenpart variable is endogenous, it has been instrumented using number of
extension agents (extenagents), and lagged tropical livestock unit (lntlubefore). Since in
the presence of endogeneity 3SLS, a full system estimator, is likely to have an efficiency
advantage over the single equation method result discussion will be based on 3SLS
estimates (Ahn, 2002; Wooldridge, 2002).
The results indicate that endogenous variable „extension participation‟, positively and
significantly influenced total income. Participant households were found to earn 14.7%
c
Edir in the context of rural and specifically in the research sites is a community–based institution
established on mutual interest of members and its primary objective is to support members during the time
of crises, such at death of family members.
d
1 ETB=0.0606 USD (http://www.xe.com/ucc/convert.cgi?Amount=1&From=USD&To=ETB). accessed
December 2, 2010.
13
more compared to non-participant households. The result, supports previous literature of
positive contribution of agricultural extension to productivity and income (Dercon et al.,
2009; Everson,2001). Our results thus contradicts the negative or non-significant impact
of extension service on productivity and income level findings of Gautam and Anderson
(1999) and mixed results of Owens et al. (2003).
The human capital resource endowment variables, household head age negatively; and
male household headship positively affected household income level. Male household
headship brings an earning 17.2 percent higher than female headed households; age is
inversely related though not strong enough (0.7 %). Adult labor force (adultlforce) and
number of dependents (dependents), though not significant, are having the expected sign.
Adult labor force is positively and number of dependents are negatively affecting
household income level. The insignificant effect of adult labor on income seems to
support Lewis‟s unlimited supplies of labor (Lewis, 1954). The inverse relationship
between lnTincome and number of dependents indicates the high level of dependency
ratio which is a typical feature of traditional societies; and detrimental effect of high
population growth (fertility rate) on saving and income.
Location variable (DistanceMekelle), as expected a priori was found to be highly
significant and negatively affecting the dependent variable. The reason for negative
relationship between the location variable and lnTincome is that the positive and strong
impact of access to market (both labor and output) to lnTincome of the household.
Despite the positive impact of the location variable to land holding size and TLU
ownership and thereby to farm income; the negative influence of location variable
through market access and non-farm income job opportunity effect is more dominant than
location variable‟s contribution to income through land holding size and TLU ownership.
Income diversification index (SID): In the SID equation (3), one endogenous and seven
exogenous variables are statistically significant affecting the dependent variable.
Extenpart are the instrumented endogenous variable affecting SID positively. The
positive sign for Extenpart are expected. On one hand, agricultural extension service are
expected to increase labor productivity and returns to labor and therefore, leads to an
increase in the reservation wage. This in turn are expected to reduce labor supply for non-
farm income activities, thereby reduce diversification or encourage specialization. On the
other hand, as the main objective of the extension program is working to diversify the
income sources of the households it induced income diversification at the household
level. Hence, the positive coefficient of Extenpart showed the dominance of the latter
(extension service contribution to diversification compared to specialization). The results
should be seen in the light of what the extension program are doing: to provide
diversified service to farmers including non-farm income generating (petty trade)
extension services. And the results are consistent with other related works (Chaplin et al.,
2004).
Interestingly, the variables adult labor force (AdultlForce) and number of dependents
(proxy for labor supply), did not show significant effects in the diversification model, but
land holding size and access to irrigation on other hand are inversely related with income
14
diversification. The inverse relationship between the labor supply and income
diversification index contradicts the widely accepted notion of surplus rural labor force as
the main driving force for income diversification so as to complement the meager farm
income. However, the inverse relationship between land and access to irrigation and SID
supports that the resource poor (fragmented and small holding) households should
diversify their income to bridge the income constraint they face (Ellis, 1998).
On the other hand, we found gender bias in the rural or neighboring small towns‟ labor
market of the research sites. Holding other things equal, income diversification level in
male-headed households were 5.1 points higher than in female-headed households.
Distance to main market is negatively (though too weak) related with SID. Access to
market improves nonfarm earning opportunities through reduced transportation cost and
easy access to information. This is consistent with the findings of Smith et al. (2001),
and Lanjouw, et al.(2001).
7. Conclusion and Recommendations
Several studies have been conducted to examine and test the determinants of extension
participation, income diversification and income level. However, most have failed to take
into account the endogeneity and contemporaneous correlation which might be expected
to exist between the different equations, which leads to inefficient results. Hence,
instrumental variable two-stage least square and three-stage least square estimation
models, which accounts the stated problems, were used to examine the effect of the
different household capitals on extension participation, and the effect of extension
participation on household income diversification and household income level.
Accordingly, the relatively wealthy households, with better land and livestock ownership,
were the ones participating in the extension program. Extension participation in turn are
becoming instrumental to influence positively income diversification, and household
income. Moreover, while the relationship between income level and basic wealth
indicators (land size, livestock ownership and access to irrigation) were positive; these
variables relationship with income diversification went the opposite direction.
Based on the regression models and descriptive statistics results, the relatively resource
poor households were the ones engaged in income diversification generating activities
mainly to supplement their farm income. Even though the extension program is positively
contributing improving household income and diversification, given the reported income
level on one hand, and large family sizes (adult and dependent) on the other hand, still
the level of per capita income were very low by any standard.
The availability of improved technologies were restricted to only few technology choices:
improved wheat maize, dairy, modern fertilizers (urea and dap). Therefore, alternative
and diversified technology options need to be sought to satisfy the need of rural
households. The non-farm income source in the research site was dominated by public
work, called safety net program and minor petty-trade activities. Innovative and
sustainable non-farm income generating activities were missing. Hence, to mitigate the
existing high dependence on land for livelihoods, improving road infrastructure,
15
information access, skill levels of farmers and streamlining the credit delivery system
could facilitate farmers‟ access to major labor markets and increase opportunities for
rural households.
References
Abate. H., 2007. Review of Extension Systems Applied in Ethiopia with Special
emphasis to the Participatory Demonstration and Training Extension System, Food and
Agriculture Organization of the United Nations, (unpublished), Addis Ababa, Ethiopia.
Abdulai A., Monnin P. and Gerber J. (2008). Joint Estimation of Information Acquisition
and Adoption of New Technologies under Uncertainty, Journal of International
Development, 20: 437-451.
Adesina, A.A. and Zinnah, M. (1993). Technology characteristics, farmers‟ perceptions
and adoption decision: A Tobit Model Application in Sierra Leone. Agricultural
Economics, 9: 297-311.
Ahn J.(2002). Beyond Single Equation Regression Analysis: Path Analysis and Multi-
Stage Regression Analysis. American Journal of Pharmaceutical Education, 66: 37-42.
Akobundu E., Alwang J., Essel A., Norton G. W., and Tegene A. (2004) Does Extension
Work? Impacts of a Program to Assist Limited-Resource Farmers in Virginia, Review of
Agricultural Economics, 26(3): 361-372.
Aredo D. (1990). The Evolution of Rural Development Policies, in Pausewang S.,
Cheru F., Brune S., and Chole E. (eds), Ethiopia-Rural Development Options, Zed Books
Ltd, London,UK.
Barrett C.A., Bezuneh M. and Aboud A. (2001). Income Diversification, poverty traps
and policy shocks in Cote d‟Ivoire and Kenya. Food Policy 26:367-384.
Belay. K. (2003). Agricultural extension in Ethiopia: the case of Participatory
demonstration and training extension system , Journal of social Development in Africa,
,18(1): 49-83.
Birkhaeuser D. and Everson. R.E. (1991). The Economic Impact of Agricultural
Extension: A Review. Economic Development and Cultural Change, 39(3): 607-650.
Birner R., Davis K., Pender J., Nkonya E., Anandajayasekeram P., Ekboir J., Mbabu A. ,
Spielman D., Horna D., Benin S., and Cohen M. (2006). From “Best Practice to Best Fit”
:A Framework for Analysing Pluralistic Agricultural Advisory Services Worldwide,
IFPRI, August 2006, Washington D.C.
Byerlee D. (1988). Knowledge Intensive Crop Management Technologies: Concepts,
Impacts, and Prospects in Asia Agriculture, In P. Pingali (ed), Impacts of Rice Research ,
16
Thailand Development Research Institute, Bangkok, and Internation Rice Research
Institute, Los Banos, Philippines.
Central Statistical Agency (2007). The 2007 National Statistics, Addis Ababa, Ethiopia.
Chaplin H., Davidova S. and Gorton M. (2004). Agricultural adjustment and the
diversification of farm households and corporate farms in Central Europe, Journal of
Rural Studies, 20: 61-77.
Delgado C.L and Siamwalla A. (1997). Rural Economy and Farm Income Diversification
in Developing Coutries, Discussion paper No. 20, Markets and Structural Studies
Division, Washington, D.C.IFPRI.
Dercon, S. (1998). Wealth, risk and activity choice: cattle in western Tanzania. Journal of
Development Economic 55(1): 1-42.
Dercon, S., Gilligan, D. O., Hoddinott, J. and Woldenhanna, T. (2009). The Impact of
Agricultural Extension and Roads on Poverty and Consumption Growth in Fifteen
Ethiopian Villages, American Journal Agricultural Economics,91(4): 1007-1021.
Easterly W. and Levine R. (1997). Africa‟s Growth Tragedy: Policies and Ethnic
Divisions, The Quarterly Journal of Economics; 112(4):203-1250.
Ellis, F., (1998). Household Strategies and Rural Livelihood Diversification: Survey
article, The Journal of Development Studies, 35(1): 1-38.
Everson R.E.(2000). Economic Impacts of Agricultural Research and Extension, A
Chapter for Handbook of Agricultural Economics vol.3: Agricultural Developments:
Farmers, Farm Production and Farm Markets, Evenson R. and Pingahi P, eds., published
by Elsevier.
Feder, G., Just, R.E., Zilberman. (1985). Adoption of agricultural Innovations in
Developing Countries : A Survey, Economic Development and Cultural Change, 33(2):
,255-298.
Feder, G., Slade, R. (1984). The acquisition of Information and the Adoption of New
Technology, American Agricultural Economics Association,.312-320.
Gautam M. and Anderson J. R. (1999). Reconsidering the Evidence on Returns to T&V
Extension in Kenya, Policy Research Working paper, The World Bank, Washington D.C.
Gebremedhin B. Jaleta M. and Hoekstra D. (2009). Smallholders, institutions services,
and commercial transformation in Ethiopia. Agricultural Economic 40:773-787.
Genius M.G., Pantzios C.J., and Tzouvelekas V. (2006). Information Acquisition and
Adoption of Organic Farming Practices. Journal of Agricultural and Resource
Economics, 31(1): 93-113.
17
Heady.E.O (1952). Diversification in Resource Allocation and Minimization of Income
Variability, Journal of Farm Economics, 34(4): 482-496.
Imai K.S., Arun T. and Annim S. K. (2010). Microfinance and Household Poverty
Reduction: New Evidence from India, World Development, 38(12): 1760-1774.
Imbens G.W. and Angrist J.D. (1995). Two-Stage Least Squares Estimation of Average
Causal Effects in Models With Variable Treatment Intensity, Journal of the American
Statistical Association, 90(430):431-442.
Jansen H.G., Pender J., Damon A., Wielemaker W. and Schipper R. (2006). Policies for
sustainable development in the hillside areas of Honduras: a quantitative livelihoods
approach. Agricultural Economics, 34: 141-153.
Lewis W. A. (1954). Economic Development with Unlimited Spplies of Labor, The
Manchester School, reprinted by permission of The Manchester School and the author.
Maertens M. (2009). Horticulture Exports, Agro-industrialization and Farm-nonfarm
Linkages with Smallholder Farm Sector: Evidence from Senegal, Agricultural
Economics, 40: 219-229.
Mendola M. (2007). Agricultural Technology Adoption and Poverty Reduction: A
propensity –score matching analysis for rural Bangladesh, Food Policy 32, 372-393.
Minot, N., Epprecht, M., Anh, T., Trung, L.Q. (2006). Income Diversification and
Poverty in the Northern Uplands of Vietnam, Research Report 145, International Food
Policy Research Institute, Washington, DC.
Mogues T., Ayele G., and Paulos Z.(2008). The Bang for the Birr: Public Expenditures
and Rural Welfare in Ethiopia, International Food Policy Research Institute , research
Report 160.
Reardon T., Taylor J.E., Stamoulis K. Lanjouw P., Balisacan A. (2000). Determinants
and effects of income diversification amongst farm households in Burkina Faso. Journal
of Development Studies, 28(1) : 264-296.
Reardon, T. (1997). Using evidence of household income diversification to inform study
of the rural non-farm labor market in Africa, World Development 25(5): 735-48.
Rogers, E.M. (1984). Diffusion of Innovations, third edition, The Free Press, Collier
MacMillan Publishers, London, UK.
Smith D.R., Gordon A. and Meadows K. (2001). Livelihood diversification in Uganda:
patterns and determinants of change across two rural districts, Food Policy; 26: 421-435.
18
Spielman D.J., Byerlee D., Alemu D. and Kelemework D. (2010). Policies to promote
cereal intensification in Ethiopia: The search for appropriate public and private roles,
Food Policy, 35: 185-194.
Sumberg J. Gilbert E. and Blackie M. (2004). Income Diversity, Technology Choice and
Agricultural Research Policy in Sub-Saharan Africa. Development Policy Review, 22(2):
131-146.
Tigray Bureau of Agriculture and Natural Resource Development-TBoANRD (2003)
Integrated Household Extension Intervention Program in Tigray vol. 1, Mekelle, Tigray,
Ethiopia.
Tigray Bureau of Rural and Agricultural development, 2008. Various years Annual
Report (In Tigrigna,unpublished).
Tiwari, K.R., Sitaula, B.K., Nyborg, I.L.P., Paudel,G.S. (2008). Determinants of
Farmers‟ Adoption of Improved Soil Conservation Technology in a Middle Mountain
Watershed of Central Nepal. Environmental Management, 42, 210-222.
Woldehana.T., Oskam.A., 2001. Income Diversification and entry barriers: evidence
from the Tigray region of northern Ethiopia, Food Policy .26:351-365.
Wooldridge, J.M. 2002. Econometric Analysis of Cross Section and Panel Data, The MIT
Press, Combridge, Massachusetts, London, England.
World Bank (2007). World Development Report: Agriculture for Development, The
world Bank , The international Bank for reconstruction and Development , Washington,
DC.
Zepeda.L., 1990. Predicting Bovine Somatotropine Use by California Dairy Farmers,
Western Journal of Agricultural Economics, 15(1):55-62.
19
Table 1 Description statistics of the explanatory variables
Participants (obs=390)
Non-parti (obs=344)
t- value
Variable
Description
Mean
Mean
Endogeneous variables
Tincome
Total household income
7053
6211
-3.4747***
lnTincome
Ln Total household income
8.8638
8.6638
-4.1941***
SID
Simpson index of diversity
0.4310
0.4130
-1.1599*
Extenpart
=1 if participant hh in agricultural extension 0 otherwise
1
0
--
household characteristics
Householdheadage
Age of the household head (continuous)
45
42.7
-2.1126**
Householdagesqr
Age of the household head squared
HHHeadgender
=1 if Sex of the head of the household is male and 0 otherwise
.8
.66
-4.3377***
Adultlaborforce
Number of adult labor force (between the of 15- 60 years)
3.08
2.35
-7.1234***
No.dependent
Number of household members outside the labor force age category
2.96
2.28
6.06***
Headschooling
Educational level of the head of the household
.95
.81
-1.35*
Farm characteristics
lnlandsize
Current Household land in tsemad ( 1ha=4 tsemad)
4.89
3.80
-4.4273***
Lnlagged TLU
Lag Livestock resource holding of the household in tropical livestock unit
1.2317
0.7286
-6.8554***
lnCurrentTLU
Current Livestock resource holding of the household in tropical livestock unit
1.1544
0.7586
-6.3434***
AccessIrigation
=1 if the household has Access to irrigation and 0 otherwise
.24
.20
1.3497*
Social Capital
Edirmembership
= 1If the hh is a member of edir 2 otherwise
.29
.16
-4.1183***
Distance to market
DistanceMekelle
Tabia distance to Mekelle market in Kms
72
68.18
-1.7452*
DistanceWoreda
Tabia distance to Woreda market in kms
13.11
11.35
-2.6531***
extenagents
Average number of extension agents working in the Tabia
3.2385
3.2879
1.7892*
disextenhr
Distance to extension agent/ extension center (walking hours)
.6853
.7320
1.3383*
Landsize*tlu
Interaction between land size and TLU
15.36
16.42
3.7301***
20
Table 2 Farm and non-farm income by participation status
Income source
Non-part (Obs=344)
Part(Obs=390)
t-values
Crop income
3707
4611
-2.1656**
Livestock income
525
776
-3.2390***
Total farm-income
4232
5387
27.3
Farm income/total income in %
68.14
69.10
Transfer income
354
259
1.4853*
Migration income
186
134
0.9953
Wage income
1071
1543
-3.9947***
Business income
368
473
-0.7392
Total non-farm income
1979
2409
21.7
Non-farm income/total income in %
31.86
30.90
Total income
6211
7796
25.5
21
Table 3 Mean income of the different income categories by income and diversification quintiles (n=734)
Income diversification index quintiles*
Income quintiles
Mean values
1
2
3
4
5
1
2
3
4
5
total income ( Eth. Birr)
9473
6394
5954
6757
6685
1744
3760
5649
8145
16032
crop income
8379
4412
3128
2981
2022
834
1885
2857
4463
10943
livestock income
222
448
595
808
1223
134
436
627
930
1168
total farm income
8601
4860
3723
3789
3245
968
2321
3484
5393
12,111
% share of farm income
90.8
76.00
62.5
56.00
48.5
55.5
61.7
61.7
66.2
75.5
Wage income
442
902
1550
1911
1806
549
1005
1431
1751
1876
business income
331
334
365
331
762
46
170
281
483
1145
transfer income
97
200
213
410
601
135
226
268
362
530
migration income
4
98
102
315
272
46
38
184
155
369
Total non/off-farm income
874
1534
2230
2967
3441
776
1439
2164
2751
3920
% share ofnon/off- farm income
9.2
23.99
37.5
43.9
51.5
44.5
38.3
38.3
33.8
24.5
Extension participants (%)
48.3
51
55.1
55.8
55.1
44.2
46.9
53.1
39.5
60.1
Non-participants (%)
51.7
49
44.9
44.2
44.1
55.8
53.1
46.9
60.5
39.9
*Note: due to rounding the % do not add-up to 100%.
22
Table 4 Probit analysis of factors affecting the probability extension participation
β*
SE
γ*
SE
Householdheadage
.0491*
.0261
.0195
.0103
Agesqr
-.0004*
.0002
-.0001
HHHeadgender
-.0885
.1396
-.0351
.0001
Headschooling
.1378***
.0419
.0548
.0552
Adultlaborforce
.1521***
.0431
.0605
.0166
No.dependent
.0884**
.0389
.0351
.0171
AccessIrigation
.0784
.1320
.0311
.0154
lnlandsize
.2821**
.1303
.1122
.0522
lntlucurrent
.0542
.0771
.0215
.0518
DistanceMekelle
.0030
.0028
.0012
.0306
DistanceWoreda
.0254***
.0087
.0101
.0011
Landsize*tlu
-.0060**
.0025
-.0024
.0034
Edirmembership
.3880***
.1361
.1512
.0010
lntlubefore1
.1917***
.0728
.0762
.0514
extenagents
-.7252***
.2183
-.2883
.0289
disextenhr
.1012
.1244
.0402
.0868
_cons
-.6307
.9245
.0495
Correctly predicted
69%
LR chi2(16) 142.10
chi2( 16) 119.8
* The β values are the coefficient estimates and the γ values are marginal effects at mean.
23
Table 5 Extension participation, household income and income diversification 2SLS and 3SLS estimate
lnTincome
SID
2SLS
3SLS
2SLS
3SLS
β
SE
β
SE
β
SE
β
SE
Extenpart
.4341*
.2591
.1468*
.0803
.3779***
.1178
.1267***
.0298
Householdheadage
-.0067***
.0017
-.0067***
.0016
-.0029***
.0008
-.0029***
.0006
HHHeadgender
.1721***
.0609
.1721***
.0565
.0516*
.0276
.0514**
.0210
Headschooling
.0018
.0196
.0007
.0185
-.0204**
.0089
-.0212***
.0068
Adultlaborforce
.0396
.0251
.0402*
.0230
-.0089
.0114
-.0082
.0085
No.dependent
-.0039
.0204
-.0032
.0186
-.0087
.0093
-.0080
.0069
AccessIrigation
.2142***
.0547
.2106***
.0509
-.0590**
.0248
-.0621***
.0189
lnlandsize
.2434***
.0564
.2415***
.0527
-.0822***
.0256
-.0838***
.0196
lntlucurrent
.0985***
.0332
.0981***
.0309
.0042
.0151
.0039
.0115
DistanceMekelle
-.0051***
.0012
-.0051***
.0011
-.0014***
.0005
-.0014***
.0004
DistanceWoreda
.0007
.0030
.0007
.0028
-.0058***
.0014
-.0057***
.0010
Edirmembership
.1083
.0708
.1104*
.0648
-.0316
.0322
-.0293
.0241
Landsize*tlu
.0027***
.0008
.0027***
.0008
.0002
.0004
.0003
.0003
_cons
8.3740***
.1413
8.5920***
.2094
.6902
.0642
.8777***
.0778
Obs
734
734
734
734
LR/Wald chi2 (13)
326.87
378.39
57.14
99.17
Prob>chi2
0.0000
0.0000
0.0000
0.0000
R-squared
0.2396
0.3429
----
0.1203
Note: *** significant at 1%, ** significant at 5% and * significant at 10%.
24
Appendix A. Robustness test of the different models
Model
Hypothesis
Type of Test
p-value
Decision
2SLS
Null hypothesis that the instruments used are valid
Sargan statistic (overidentification test of all instruments
0.56914
Accept
3SLS
Null hypothesis that all the instrumental variables
are uncorrelated with the residuals
Tests of over identifying restrictions
nR
2
< χ
q
2
; q number
of IV (1 % level)
Accept
2SLS
lnTincome
Ho: Disturbance is homoskedastic
Pagan-Hall general test statistic
0.5080
Accept
2SLS
SID
Ho: Disturbance is homoskedastic
Pagan-Hall general test statistic
0.6566
Accept
2SLS
lnTincome
H0: Regressor is exogenous
Wu-Hausman F test:
0.00068
Reject
2SLS
SID
H0: Regressor is exogenous
Wu-Hausman F test:
0.00054
Reject
