Greener Journal of Agricultural Sciences Vol. 9(2), pp. 119137, 2019 ISSN: 22767770 Copyright ©2019, the copyright of this article is
retained by the author(s) DOI Link: http://doi.org/10.15580/GJAS.2019.2.010919010 http://gjournals.org/GJAS 

Impact
of Improved Soybean (Belessa95) Variety on Income among Smallholder Farmers
in Bambasi Woreda, Benishangul Gumuz Regional State
Ethiopian Institute of Agricultural Research
(EIAR), Planning, Monitoring and Evaluation Researcher
^{ }
ARTICLE INFO 
ABSTRACT 
Article
No.: 010919010 Type: Research DOI: 10.15580/GJAS.2019.2.010919010 
The importance of
agricultural technology in enhancing the welfares of farmers can be realized
when yield gain from the technologies results in meaningful income gain.
This article aimed to assess economic impact of improved soybean
(Bellessa95) variety on income among farm households in Bambasi District,
BGRS. In this study a multistage stratified sampling technique was employed
to select rural kebeles and households. Three rural kebeles were selected
randomly. Structured interview schedule was developed, pretested and used
for collecting the essential quantitative data for the study from 134
randomly selected households. Descriptive statistics and propensity score
machining (PSM) models were employed to analyze data. Results of descriptive
analysis showed that there were statistically significant differences
between adopter and nonadopter households with distance to market,
livestock ownership, and frequency of extension visit, farm income as well
as number of oxen owned. Consistent with the findings of previous studies,
regression results showed that adoption of improved soybean has a positive
and significant effect on farm income by which adopters are betteroff than
nonadopters. Based on results obtained it is recommended to continuous
training in improved soybean production. Promoting farmers to form or join
cooperatives. Strengthening demonstration centers and Farmers Training
Centers (FTC). Transaction costs should be reduced and scaling up and
diffusion of improved soybean varieties in the study area should be broadened. 
Submitted: 09/01/2019 Accepted: 12/01/2019 Published: 06/04/2019 

*Corresponding
Author Mr.
Musba Kedir Email:
Kedirmusba44@ gmail.com Phone:
+251913917911 

Keywords: 



In
Ethiopia, agriculture takes the lion’s share (72.7%) in terms of employment.
The sector is the livelihood of the 75.26 million (79.77%) of the population.
It is the source of food and cash for those who are engaged in the sector and
others. Most agricultural holders acquire the food they consume and the cash
they need to cover other expenses only from farming activities. Since farming
in Ethiopia is often precarious and usually at the mercy of nature, it is
invariably an arduous struggle for the holders to make ends meet (UNDP, 2014).
The
last fiveyear Growth and Transformation Plan (GTP) (2010/112014/15) and the
second GTP (2015/162020/21) gives special emphasis to the role of agriculture
as a major source of economic development. Following the Agricultural
DevelopmentLed Industrialization (ADLI) strategy and building on PASDEP
achievements, the GTP has the priority to intensify productivity of
smallholders and strongly supports the intensification of marketoriented
agriculture, either at national or international level, and promotes private
investments. The plan includes scaling up of best practices to bring average
farmers’ productivity closer to those of best farmers, expanding irrigation
coverage and shifting to production of high value crops to improve income of
farmers and pastoralists, with complementary investments in market and
infrastructure development (FAO, 2012).
Soybean is among the cash crops which have been
given priority in the national effort of meeting increased income and
nutritional security of households. To this effect number of improved soybean
varieties have been developed by the national research system and disseminated
among smallholder farmers. Benishangul Gumuz region is one of the beneficiary
regions of the country since 2006. The region has 156,000 ha of land that is
suitable for soybean production. However, the contribution of the already
disseminated varieties has not been well understood among the beneficiary
farmers for informed decision making and to justify further expansion of
soybean production zone in the region.
Kathelen (2010) reported on his study that a welldesigned impact
assessment study can provide insight in to the causal factors behind the
success and failure of various improved variety adoption activities. Impact
assessment thus provides information that allows research and extension
institutions to improve their services, and improve the welfare of the farmer.
Moreover, ADB (2006) indicated that project impact evaluation established
whether the intervention had a welfare effect on individuals, households, and
communities, and whether this can be attributed to the concerned intervention.
Generally, the focus of
impact assessment goes beyond the products of research (such as improved
variety) to determine the effects of adoption of its products. In other words,
adoption of the products of research is a prerequisite for attaining impact.
Impact assessment is done for several reasons including accountability,
improving future design, prioritizing and implementation of similar programs.
As FAO (2000) put it; the results of this process provide continuous feedbacks
to the project planning, prioritizing and implementation. It can be undertaken
before initiating the project (exante), during the project period (midterm) or
after the completion (expost) of the project or activity (Anandajayasekeram, et al., 1996).
Though
traditional agriculture prevails in developing countries in general and in
Ethiopia in particular, some progress has been made in using improved
agricultural technologies and inputs. High yielding soybean varieties are among
important technologies promoted by the country’s agricultural research and
extension system.
A
good technology has to guarantee sustainable productivity across niche agro
ecologies and over a period of time. The possible outcomes of a research
undertaking are commonly conceptualized in terms of yield increases or reducing
yield losses. However, such yield increases often require additional inputs,
which lower the effective value of yield gains. Farmers particularly resource
poor ones, will only adopt technologies if net yield gains are significantly
greater than zero. (Mills, 1997 cited by Mengistu, 2003).It is then very
important to know whether additional yields or returns obtained as a result of
using an improved technology are sufficient enough to qualify technology for
wide scale dissemination. In this respect, technical changes brought about
research have to be evaluated for the benefit that the farmers get from them.
Added to this, in
developing countries like Ethiopia, some new agricultural technologies have
been only partially successful in improving productive efficiency. One of the
probable reasons might be farmers’ deviation in the use of associated inputs
from the developed technological packages. Farmers may not be able to operate
at higher levels of inputs because of capital scarcity or lack of access to
purchased inputs. Consequently, analysis of the impact of new varieties should
begin with how the characteristics of the variety and characteristics specific
to small farmers interact. More precisely, increasing production and ultimately
the income levels of smallholder farmers through technological advancement
requires better understanding of the production system and producers’ behavior
in the use of farm resources.
The adoption of
improved technologies that enhance agricultural productivity and improve annual
income of smallholder farmers in Benishangul Gumuz Region is instrumental for
achieving better economic growth of the farmers. Dissemination of improved
soybean variety known as Bellesa95 was started in 2006 with few
early adopters in the region especially in Bambasi woreda in order to increase crop productivity and incomes for small
scale farm household. However, its impact on farm income were
not known and no effort had been made to evaluate the program and its
activities hence creating an information gap that needed to be filed. In spite
of the government‘s efforts to address the issue of low productivity, scaling
up and dissemination of improved varieties are in progress such as improved
soybean varieties. But the impact of the improved soybean on income is not
studied in the study area. Therefore; this study was intended to assess the
impact of improved soybean (Bellesa95)
on household
income.
A comprehensive understanding of impact of improved
varieties in a diverse environment is necessary to design appropriate
strategies to harness their potential benefits in target domains. Research and
extension workers engaged in development and transfer of production
technologies can utilize the results of this study in setting research and
extension agenda. The assessment is important to better inform policy makers
about the contribution of soybean varieties. It helps draw good lessons about
the performance of existing approaches to development and poverty reduction thus
improves targeting of research programs. Smallholder soybean farmers also
benefit to foster adoption and diffusion of new agricultural technologies.
Other similar studies elsewhere will also benefit by using the information
generated by this study.
The
study covers only Bambasi Woreda of
Benishangul Gumuz region. This is mainly because of the limited resource
available and time to undertake the study on a wider scale. The
study impact of improved soybean (Bellesa95) variety on the farm income
of smallholder farmers.
The study was based on cross sectional data and single variety and may not show
the dynamics of impact from soybean technology adoption.
1.5. Objectives of the study
The general
objective of the study is to analyze impact on income of improved soybean (bellessa95) variety among farm
households.
Specific objectives
include
1.
To assess socioeconomic factors that affect
use of improved soybean varieties.
2.
To analyze the impact of soybean adoption on
farm income of farmers in the study area.
Assosa zone is
located 664 km southwest of Addis Ababa. Based on CSA (2013), this Zone has a
total population of 310,822, of whom 158,932 are men and 151,890 women. When we
see number of urban inhabitants it is 39,957 or 12.86% of the population in the
woreda. A total of 72,879 households
were counted in this Zone.
Bambasi woreda is located 629
km southwest of Addis Ababa and about 45 km southeast of Assosa, the capital
city of Benishangul Gumuz Regional State (BGRS). The woreda is geographically located at 9°45′N latitude and 34°44′E
longitude (Figure 3). It is one of the 20 woredas in the BenishangulGumuz Region of Ethiopia. Part of the Assosa Zone, it is bordered by the MaoKomo special woreda on the southwest, Asossa in the northwest, Oda Buldigilu in the northeast,
and by the Oromia
Region in the southeast. The Woreda
has an area of about 784.4 square kilometers.
According to CSA (2015) total population for this woreda is 71,279 out of this 37543 (52.65%) are male and 33,736
(47.34%) are female. The altitude of the area ranges from 13001470 m.a.s.l.
The temperature of the area ranges from 2135^{o}C. The rainfall of the
area has unimodal pattern extending from May to October with and annual mean of
1350 to 1450 mm.
In the woreda the
common crops grown are maize, sorghum Teff,
chickpea and groundnut. The study area is also suitable for oil crops like
soybean and noug. In addition to this, perennial crops such as mango and
avocado are common in the woreda.
Figure 1. Map of Study Area
Source: Own
sketch (2017)
2.2.
Sampling Technique and sample size
A clear and precise
identification and definition of the population of the study is an important prerequisite
for research sample design. In this study, a three stage sampling procedure was
adopted to collect the required primary data. In the first stage, among the
seven districts in Assosa Zone, Bambasi woreda was selected purposively since
scalingup of improved soybean variety (Bellesa95) was implemented in the
district. In the second stage among 26 kebeles in the woreda three of them
(Dabus, Amba16 and mender46) were randomly selected. Finally, stratified
random sampling method was employed to identify sample households for inclusion
in the study. To this effect, list of adopter households was obtained from
district agricultural office and from development agents at each sample kebeles
and then households in the area were categorized into 2 strata, that is 483
adopter of improved soybean households, and 612 nonadopter households. The
total sample was determined using Cochran (1977) stratified sampling procedure
considering resource limitation of the study for the Woreda. The sample size
thus obtained was assigned to each kebele based on probability proportional to
size of the households. 134 sample respondents, 67 sample households from each
category were drawn randomly based on probability proportional to sample size
(PPS). These both groups were chosen based on their close similarity in their
socioeconomic characteristics.
1
Where, n_{0}
is the sample size, Z is the
selected critical value of desired confidence level (1.96), p is the estimated proportion of an
attribute that is present in the population. q =1p and e is
the desired level of precision which is 0.08.
Table 1:
Soybean producers selected from each identified kebeles

Name of the kebeles 
Adopter households 
Nonadopter households 
Total Sample households selected 



Total 
Sample 
Total 
Sample 
Total sample 

Dabus 
123 
17 
202 
22 
39 

Amba16 
210 
29 
240 
26 
55 

Mender46 
150 
21 
170 
19 
40 
Total 
483 
67 
612 
67 
134 
2.3.
Type
and Methods of Data Collection
Primary
and secondary data were collected through semistructured schedule and
checklists respectively. Trained enumerators were used to collect the data. Primary
data was collected from sample households and secondary data sources were
woreda administration and some published documents.
The
type of primary data collected include households’ demographic characteristics,
asset endowments, access to market, access to credit, membership in different
rural institutions, and income sources. Interviews were conducted with district
level agricultural experts on production and productivity of improved soybean
to generate supplementary (qualitative) data. In addition, Focus Group
Discussions (FGD) was conducted with groups of selected farmers from each
sample kebeles to support interpretation of results obtained from field survey
on changes in adoption of improved soybean technologies and their impact on
household income
Descriptive statistics
such as mean, standard deviation, percentages, frequency, charts, and graphs,
used to describe different categories of sample units with respect to the
desired socioeconomic characteristics. Moreover, inferential statistics such as
chisquare test (for categorical variables) and ttest (for continuous
variables) were used to compare and contrast different categories of sample
units with respect to the desired characters so as to draw some important
conclusions.
There are three
potential source of bias in impact studies. The first one is that participant
households may significantly differ from nonparticipants in a community as
well as household level due to observable characteristics (such as geographic
remoteness, or a household’s physical and human capital stock) that may have a
direct effect on outcome of interest. Secondly, the difference arises due to
unobservable community level characteristic. For instance, dynamic leaders at
community level may in part drive the existence of a project. At the household
level, some households may have certain advantages such as good entrepreneurial
spirit or relationship with other programs/projects that may significantly
influence behavior. Thirdly, externalities (spillover effect) exerted by
project on nonparticipants (Davis et al., 2010). Because of the above
problems, differences between participants and nonparticipants may, either
totally or partially, reflect initial differences between the two groups rather
than the effects of participating in the scaling out program.
The use of PSM to
minimize selectivity bias thus suggests that these differences are the result
of participating in scaling up/out program rather than the intrinsic
characteristics of the sampled households. However, like the simple mean
comparison, PSM may misinterpret the treatment effect, because it only controls
for observed variables, and hidden selfselectivity bias may remain. PSM is
chosen for this study because it does not require baseline data, the treatment
assignment is not random and considered as secondbest alternative to
experimental design in minimizing selection biases mentioned above (Baker,
2000).
Moreover, assuming
that technology adoption is a function of a wide range of observable
characteristics at household level, removing the assumption of ‘‘constant
technology effect’’ allows us to follow the PSM method (Mendola, 2007).
Rosenbaum
and Rubin (1983) were the first to develop the PSM statistical tool. The
technique has attracted attention of social program evaluators (Jalan and
Ravallion, 2003; Dehejia and Wahba, 1999). PSM is a nonparametric estimation
method that works by reweighting the comparison sample to provide an estimate
of the counterfactual of interest what the outcome of a beneficiary household
would have been had it not received program benefits. Since dissemination of
improved soybean variety is targeted small holder farmer’s comparison of mean
outcomes between beneficiaries and nonbeneficiaries would lead to bias
estimates. In order to solve this problem the study used the matching technique
called propensity score matching method, which is capable of extracting
comparable pair of treatment comparison households in a nonrandom program
setup and absence of base line data.
According
to Caliendo and Kopeinig (2008), the estimation of the impact of household’s
adoption of improved soybean on a given farm income (Y) is specified as:
2
Where
τi is treatment effect (effect due adoption of improved soybean),
Yi is farm income of household i, Di is whether household i has
got the treatment or not (i.e., whether a household is adopt or not).
However,
one should notice that) Yi (Di = 1) and Yi (Di =0)
cannot be observed for the same household at the same time. Depending on the
position of the household in the treatment (adoption of improved soybean),
either Yi (Di =1) or Yi (Di =0) is unobserved
outcome (called counterfactual outcome). Due to this fact, estimating
individual treatment effect τ_{i} is not possible and one
has to shift to estimating the average treatment effects of the population than
the individual one. Most commonly used average treatment effect estimation is
the ‘average treatment effect on the treated (τ _{ATT}),
and specified as:
_{ }(3)_{}
As the counterfactual mean for those being treated, E [Y (0)/ D = 1] is not observed, one has to choose a proper substitute for it in order to estimate ATT. One may think to use the mean outcome of the untreated individuals, E[Y (0)/D = 0] as a substitute to the counterfactual mean for those being treated, E[Y (0)/D =1]. However, this is not a good idea especially in nonexperimental studies, since it is likely that components which determine the treatment decision also determine the outcome variable of interest._{ }
In this particular case, variables that determine household’s
decision to adopt improved soybean could also affect household’s farm income.
Therefore, the outcomes of individuals from treatment and comparison group
would differ even in the absence of treatment leading to a selfselection bias.
By rearranging, and subtracting E[Y (0)/D =
0] from both sides of equation (2), one can get the following specification for
ATT.
(4)
Both
terms in the left hand side are observables and ATT can be identified, if and
only if E[Y (0)/D =1] − E[Y (0 /D = 0] = 0. i.e., when there is
no selfselection bias. This condition can be ensured only in social
experiments where treatments are assigned to units randomly (i.e., when there
is no selfselection bias). In nonexperimental studies one has to introduce
some identifying assumptions to solve the selection problem. The following are
two strong assumptions to solve the selection problem.
i.
Conditional Independence
Assumption (CIA)
The
Conditional Independence Assumption is given by:
(5)
Where: C
indicates independence, X is a set of observable characteristics,
Y (0) is nonadopter, and Y (1) is
adopter.
Independence indicates that given a set of observable covariates
(X) which are not affected by treatment (in our case, adoption of improved
soybean) and potential outcomes (farm income) are independent of treatment
assignment (independent of how adoption decision is made by the household).
This assumption implies that the selection is solely based on
observable characteristics (X) and variables that influence treatment
assignment (households’ adoption decision) and potential outcomes (farm income)
are simultaneously observed (Bryson et al. 2002; Caliendo and Kopeinig,
2008). Hence, after adjusting for observable differences, the mean of the
potential outcome is the same for D = 1 and D =0 and
Instead of conditioning on X, Rosenbaum and Rubin (1983),
suggest conditioning on a propensity score (propensity score matching). The
propensity score is defined as the probability of participation for household i given a set X which is
household’s characteristics, P(X ) = pr (D = 1/ X
). Propensity scores are derived from discrete choice models, and are then
used to construct the comparison groups. Matching the probability of
participation, given covariates solves the problem of selection bias using PSM
(Liebenehm et al., 2009). The distribution of observables X is
the same for both participants and nonparticipants given that the propensity
score is balancing score (Liebenehm et al., 2009).
Matching can be performed conditioning on P(X) alone rather than
on X, where P(X) = Prob (D=1X) is the probability of adoption of improved
soybean conditional on X. If the outcomes without the adoption are independent
of participation given X, then they are also independent of participation given
P(X). This reduces a multidimensional matching problem to a single dimensional
problem. Due to this, differences between the two groups are reduced to only
the attribute of treatment assignment, and unbiased impact estimate can be
produced (Rosenbaum and Rubin, 1983).
ii.
Common support region assumption
Imposing a common support condition ensures that any combination
of characteristics observed in the treatment group can also be observed among
the control group (Bryson et.al. 2002). The common support region
is the area which contains the minimum and maximum propensity scores of treatment
and control group households, respectively. It requires deleting of all
observations whose propensity scores is smaller than the minimum and larger
than the maximum of treatment and control, respectively (Caliendo and Kopeinig,
2008). This assumption rules out perfect predictability of D given X.
That is (6)
This assumption improves the quality of the matches as it excludes
the tails of the distribution of P(X), though this is done at the
cost that sample may be considerably reduced. Yet, nonparametric matching
methods can only be meaningfully applied over regions of overlapping support.
No matches can be formed to estimate the parameters when there is no overlap
between the treatment and comparison groups. This assumption ensures that
persons with the same X values have a positive probability of being both
participants and nonparticipants.
Given
the above two assumptions, the PSM estimator of ATT can write as:
(7)
Where p(X) is the propensity score computed on the covariates X.
Equation (10) shows that the PSM estimator is the mean difference in outcomes
over the common support, appropriately weighted by the propensity score
distribution of participants.
According to Caliendo and Kopeinig (2008), there are steps in
implementing PSM. These are estimation of the propensity scores, choosing a
matching algorithm, checking on common support condition testing the matching
quality and sensitivity analysis if it is necessary.
Propensity score
involves a series of empirical steps. First, logit model that predicts the
probability of each household adopting improved soybean (propensity score) as a
function of observed household and community characteristics was estimated,
using a sample of adopters and nonadopters. In estimating the propensity
score, the dependent variable used in the model was a binary variable a value
of 1 for adopters of improved soybean variety and 0 otherwise.
The estimates of
individual adoption using logit model are useful for two reasons. First, it
gives some insight regarding the observable variable that should be included in
the balancing function. Second, it provides a better understanding of adoption
of improved soybean variety by the
kebeles people.
According to Gujrat
(2004), the logistic distribution function for the determining factors for
adoption of improved soybean of the household can be specified as:
(8)
Where p(i) is a
probability of a household being
diversified for i^{th} household and Z(i) is a function of m
explanatory variables (Xi) and expressed as:
(9)
Where is the intercept and ^{ }is the slopes parameter
in the intercept in the model which is estimated using maximum likelihood
method.
The probability that a household belongs
to not adopt is:
(10)
Therefore,
(11)
And
(12)
Taking the natural logarithms of the odds
ratio of equation (15) will result in what is known as the logit model as
included below:
(13)
If the disturbance term U_{i }is
taken in to account the logit model becomes:
(14)
After estimation of the propensity score,
seeking an appropriate matching estimator is the major task of program
evaluator. There are different matching estimators in theory. However the most commonly
applied matching estimators are Nearest Neighbor matching (NN) Caliper
matching, Kernel and local linear matching (Smith and Todd, 2005). The
performance of different matching algorithms varies casebycase and depends
largely on the data sample (Caliendo and Kopeing, 2008).
NearestNeighbor
Matching:
In NNM, an individual from a comparison group is chosen as a matching partner
for a treated individual that is closest in terms of propensity score Caliendo
and Koping, (2008). For each control unit this method assigns a weight equal to
one for the nearest comparison unit in terms of the balancing score and zero to
all the other comparison observations. NN matching can be with replacement, and
without replacement. In the case of matching with replacement a single
comparison unit can be used as a match for more than one treatment unit. On the
other hand, NN matching without replacement, a comparison individual can be
used only once and increases bias but it could improve the precision of the
estimates (Abadie and Imbens, 2006).
Kernel Matching: This is another matching method
whereby all treated units are matched units with a weighted average of all
controls with weights which are inversely proportional to the distance between
the propensity score of treated and controls (Becker and Ichino, 2002). The
weighting function is a (Gaussian) kernel density. All the observations in the
comparison group inside the common support region are used, the farther the
comparison unit from the control unit the lower the weight. The kernel
estimator matches treatment units to a kernel a weighted average of comparison
units. This can be thought as a nonparametric regression of the outcome on a
constant term. For the treatment units, weights are given by:
(15)
Where G (i,j) is a kernel function and h_{n }is
a bandwidth parameter.
Radius matching method: is each treated unit is matched only with the control units whose
propensity score falls in a predefined neighborhood of the propensity score of
the treated unit. If the dimension of the neighborhood (i.e. the radius) is set
to be very small it is possible that some treated units are not matched because
the neighborhood does not contain control units. On the other hand, the smaller
the size of the neighborhood the better is the quality of the matches.
Caliper Matching: With radius matching each treated unit is matched only with the
untreated units whose propensity score falls within a prespecified range of
neighborhood of the propensity score of the treated unit. If the range of the
neighborhood, i.e. the radius, is to be very small it is possible that some
treated units are not matched because their neighborhood does not contain any
untreated units. On the while, the smaller the size of the neighborhood the
better is the quality of the matches. However, by using more untreated
observations, one can increase the precision of the estimates, but at the cost
of increased bias.
Secondly, radius matching (RM) involves all
neighbors within a maximum propensity score distance (caliper), a priori
defined, and thus corresponds to the common support assumption. Additionally,
poor matches through toodistant neighbors are avoided (Dehejia and Wahbha,
2002; Smith and Todd, 2005).
The “balancing properties” of the data was checked
by testing that treatment and comparison observations had the same distribution
(mean) of propensity scores and of control variables within grouping (roughly
quintiles) of the propensity score.
The study was employed two types of balancing test.
First a simple ttest was used to examine whether the mean of each covariate
differs between the treatment and the control group after matching. Secondly,
it was run logit using sample after matching and compare the pseudo R^{2 }with
that obtained from the logit estimation using the sample before matching. As
proposed by Smith and Todd (2005), if matching is successful, the after
matching logit should have no explanatory power so that the pseudoR^{2 }should
be fairly low.
The methods to detect lack of overlap are to plot
distribution of covariates by treatment groups and inspection of the propensity
score distribution in both treatment and control groups. According to Imbens
(2004), one obvious approach to check this assumption, is through visual
inspection of the propensity score distributions, can typically give the
researcher a good, initial reading of the extent to which there is overlap in
the propensity score of the treatment and comparison units. In this study, lack
of overlap was detected by plotting the distribution of propensity score and
visually assess whether the overlap assumption holds.
There are different approaches in applying the
method of covariate balancing (i.e., the equality of the means on the scores
and all the covariates) between treated and nontreated individuals. Among
different procedures the most commonly applied ones are described below.
Standard bias: One suitable indicator to assess the distance in marginal distribution of
the X variable is the standard bias (SB) suggested by Rosenbaum and Rubin
(1985). It is used to quantify the bias between treated and control groups. For
each variable and propensity score, the standardized bias is computed before
and after matching as:
(16)
Where and are the sample means for the treatment and control
groups V1(X) and V0(X) are the corresponding variance (Caliendo and Kopeing,
2008). The bias reduction (BR) can be computed as:
(17)
One
possible problem with the SB approach is that one does not have a clear
indication for success of the matching procedure.
Joint significance and pseudo R^{ 2}:
Additionally, Sianesi
(2004) suggests reestimating the propensity score on the matched sample, i.e.
only on participants and matched nonparticipants, and comparing the pseudo R^{ 2} before and after matching. The pseudo R^{ 2} indicates
how well the repressors X explain the participation probability. After matching
there should be no systematic differences in the distribution of covariates
between both groups and therefore the pseudo R^{ 2} should be fairly low. Furthermore, one can also
perform a likelihood ratio test on joint significance of all covariates in the
probit or logit model. The test should not be rejected before, and should be
rejected after, matching. In our cases the combination of the above procedures
were applied.
Bootstrapping: Standard errors in psmatch2 are invalid,
since they do not take into account the estimation uncertainty involved in
probit/logit regressions (pscore). One way to deal with this problem is to use
bootstrapping as suggested by Lechner (2002). This method is a popular way to
estimate standard errors in case analytical estimates are biased or
unavailable. Recently it has been widely applied in most of economic literature
in impact estimation procedures. Each bootstrap draw includes the reestimation
of the results, including the first steps of the estimation (propensity score
and common support). Bootstrap standard errors attempted to incorporate all
sources of error that could influence the estimates.
Sensitivity analysis: Matching estimators are not robust against ‘hidden biases’. Different
researchers become increasingly aware that it is important to test the robustness
of results to departure from the identifying assumption. Estimation of
treatment effects with matching estimators is based on the unconfoundedness or
selection on observables assumption. However, if there are unobserved variables
which affect assignment into treatment and the outcome variable simultaneously,
a ‘hidden bias’ might arise. The size of the bias depends on the strength of
the correlation between the unobservable factors, on the one hand, and
treatment and outcomes, on the other (Rosenbaum, 2002).
The basic question to be answered here is whether
inference about treatment effects may be altered by unobserved factors. In
other words, one wants to determine how strongly an unmeasured variable must
influence the selection process in order to undermine the implications of
matching analysis. This is the main motive behind sensitivity analysis.
The bounding approach proposed by Rosenbaum (2002)
in order to check the sensitivity of the estimated ATT with respect to
deviation from the Constant Independence Assumption. The bounding approach does
not test the confoundedness assumption itself. However, it provides evidence on
the degree to which the significance of results hinge on this assumption. If
the results turn out to be sensitive, the evaluator might have to think about
the validity of his identifying assumption and consider other estimation
strategies.
Adopter (treatment) and non adopter
categories was identified based on the adoption of improved soybean variety. In
this study, two years data (2014 and 2015) on area allocated to improved
soybean variety was used to categorize the two groups. Adopters (participants)
are those that allocated land to improved soybean varieties for two or more
years while nonadopters (nonparticipants) are those who did not allocate land
for this variety at all. It is equal to one if the farm household has adopted
the variety and zero otherwise.
Farm income (FARMICOM): It is a continuous variable, which refers to the annual total income
obtained from farm output including soybean production.
Age of the farm
household head (AGEHH): it is defined as
the period from the respondent’s birth to the time of the interview measured in
years. It is a continuous variable. Those farmers having a higher age will have
much better probability to adopt the improved soybean variety and increase its
income due to experience obtained throughout his life time. Hossain
and Crouch (1992) revealed that the probability of adoption of new farm
practices increases with farmer’s age in Bangladesh.
Sex of household head
(SEXHH): This is dummy variable, which takes value of 1 if
the head of household is male and 0 otherwise.
Gender differentials in the farm household play a significant role in
economic performance of a given household. One may expect that male headed
households have more probability to adopt improved soybean variety than female
headed households. This may be because male headed households have a chance to
move and participate in different meetings and have opportunity to get more
information. While female headed households are more preoccupied with childcare
and home management than interaction with the external environment. According
to Yenealem (2013) men are more likely to adopt improved variety than women. Therefore, it
is hypothesized that sex of household head and adoption of improved soybean
variety are positively related.
Level of education of
the household head (EDUCTGRY): This refers
to the number of schooling years attained by the respondents up to the time of
the survey. According to Asfaw (2013) those farmers who have better level of
schooling have high chance of being project participant. It is hypothesized
that household heads that are literate have a better knowledge about improved
varieties. This variable is a categorical variable and is expected to have a
positive relationship with annual income. According to Alene
(2000) level of education of the head of
the household has a positive and significant influence on the adoption and use
of improved maize variety with each additional year of schooling increasing the
probability of adoption by 3.9%.
Family size
(FAMSIZE): A continuous variable measured in numbers refers to the number of
family members of a given household. The family members are important in the
operation of farm activities, such as weeding, harvesting. Adoption of new
technology often implies a need for additional labor. Family size contributes
to the variation in the decision to adopt improved soybean variety. A research
result reported by Solomon (2012) show that family size influences adoption
positively. Therefore, it is hypothesized that family size is influence the
adoption of improved soybean variety positively.
Farming experience
(FARMEXP): Farming experience is measured in
years. The study conducted by Solomon (2012) indicated that farmers experience
was found to be significantly influenced the probability of adoption of barley
variety. A positive influence of farming experience on the probability
of adoption of improved soybean varieties is hypothesized.
Membership to cooperatives
(COOPMEMR): membership to cooperatives represents
whether a household is member to cooperatives or not. Cooperatives worldwide
are committed to the concept of mutual selfhelp. This makes them natural tools
for social and economic development, and provides significant additional
benefit to communities and social systems. Formal as well as informal
associations, such as indigenous cooperation groups, enforcing widely agreed
standards of behavior, and uniting people with bonds of community solidarity and
mutual assistance. As such, they embody important forms of social capital
representing forums where in local communities can unit and act collectively Mekuria (2013). Membership to
cooperatives also will increase households’ access to services that might be
granted by being member. This variable was expected to be positively related to
adoption of improved soybean variety.
Market distance
(MARKTDIST): It refers to distance from the
residence of the farm household in kilometers to the market. Farmers near to
markets have more information on price of agricultural commodities such as
soybean compared to those away from the town. And can sell their products with
better margin and transportation cost will be lower. The result of Berhanu et
al., (2014) revealed that, distance to
market affects agricultural
technology adoption decision of farm households negatively. Therefore, nearness to the market is expected to improve adoption of
improved soybean variety and increase farm income of household.
Distance to the nearest
road (DISTROAD): This is a
continuous variable that is an average kilometers to main road. Proximity to
main road has indirect effect on adoption of improved soybean variety. It was
hypothesized that nearness to main road affects adoption of improved soybean
variety positively.
Dependency ratio
(DEPRATIO): It is a continuous variable and
defined as the ratio of active to total family members. This ratio tells us the
proportion of household members who are able to feed the dependent active
members of the family. The more the dependency ratio in a household more of the
production would be utilized for consumption rather than exchanged with
improved soybean variety. Hence, it is hypothesized that this variable has
negative influence on adoption of improved barley varieties. Tadesse and Belay
(2004) reported the number of economically active family members is found to
influence adoption of soil conservation mechanisms negatively and
significantly.
Frequency of visit by
extension agent (FRQNCVST): The number of
times that the extension agents visit the farmer is determinant factor to
technology adoption. Those farmers who are more visited by extension agents are
expected to positively influence adopt improved soybean variety than rarely
visited. Bingxin et al., (2011) found that access
to fertilizer and improved seed adoption was related to access to extension
services.
Access to credit
service (CREDIT): It is hypothesized that
access to credit and adoption of improved soybean variety have positive relationship. The variable is entered in the model as a
dummy variable (it takes a value 1 if the household has access to credit
service and 0 otherwise). Geremew (2012) on his thesis analysis of smallholder
farmer’s participation in production and marketing of export potential crops: the case of sesame in Diga district, East
Wollega zone of oromia regional
state, those farmers who have access to
credit service are more likely to produce significant amount of sesame than
their counter parts.
Participation in demonstration
(PARTDMRN): Demonstration is one of the means by
which farmers acquire new knowledge and skills and it is measured by the number
of times the farmer has participated in demonstration after improved soybean
variety was introduced. Hence, participation in demonstration is expected to
positively influence farmers’ adoption behavior. Tesfaye and Alemu (2001)
reported that participation in onfarm demonstration contributed positively to
farmers’ adoption decision.
Number of oxen owned
(NBROXEN): It is a continuous variable to number of oxen owned
by the households who own oxen have better chance to produce more. This is
because oxen possession allows undertaking farm activities on time and when
required. Therefore, it is expected that possession of oxen size increases the
probability of adoption of improved soy bean and thereby increase farm income
According to Solomon (2012) on his study the impact of livestock development
program on farm household cash income: the case of Umbullo Wacho integrated
watershed, Dore Bafano Woreda in
Southern Ethiopia, households who do not own oxen are more like to participate
in livestock development program.
Participation in
training (PARTRAN): Training is
one of the means by which farmers acquire new knowledge and skills and it is
measured by the number of times the farmer has participated in training after
improved soybean variety was introduced. Hence, participation in training was
expected to positively influence farmers’ adoption behavior. Minyahel (2007)
reported that adoption of improved bread wheat was positively affected by
participation in training.
Residence status
(RSDCSTATS): Residence status is dummy variable if
the farmer is indigenous to the region 1, otherwise 0. Being permanent
residence of the woreda was expected to have positive relationship with
adoption of improved soybean variety.
In the area, the average improved soybean farm size of sample grower
households was 2.91ha, and the maximum farm size was 12 ha and the minimum was
0.8 ha. The average quintal obtained was 6.47 quintal and the maximum was 85
quintal while the minimum was 0.75 quintal. The average gross income from
improved soybean production of the sample adopter households from one season
harvest during 2014/2015 production year was 3897.46 birr.
From the sample of the population 50% are nonadopters and 50% of them are
adopters. Improvement in production and productivity of a given crop depends,
among other things, on presence and use of better and improved varieties.
The landholding of the sample households ranges from 0.8 ha to 12.5 ha
with an average figure of 1.51 hectares. The average livestock (including
cattle, sheep, goats, pack animals, and poultry) was 3.45 TLU with the minimum
and the maximum holdings of 0 TLU and 38.22 TLU respectively. The average labor
force available was 6 man equivalents. Households generated cash income by selling crops (such as soybean,
sorghum, maize, chickpea and noug) about 54% could do also from offfarm
activities. About 72% had access to institutional credit to purchase farm
inputs. The average distance from the nearest market place was 3.46 km with the
minimum and maximum figures equal to 0.2 km and 8 km, respectively.
A combination of different descriptive, the means and standard deviation and
inferential, the ttest and X^{2}test,
statistics for explanatory variables of sample households were performed on the
household level data to inform the subsequent empirical data analysis.
The descriptive and inferential results presented on Table 3 show that
there was statistically significant difference between adopters and
nonadopters in terms of distance to market, TLU, frequency of extension visit,
farm income and oxen ownership in favor of the adopters.
Table 2. Descriptive statistics of continuous independent
variables
Variables 
Mean across
adoption categories 
ttest 
P Value 

Adopter 
Nonadopter 

Distance to
market 
2.91 
4.00 
3.66*** 
0.0001 
Age of household
head 
46.71 
47.597 
0.37 
0.7087 
Livestock
holding(TLU) 
4.22 
2.52 
2.50 ** 
0.0134 
Family size 
6.447 
5.83 
.26 
0.2096 
Frequency of
Visit 
21.94 
15.32 
1.74 * 
0.0825 
Distance to main
road 
1.05 
1.12 
0.36 
0.7131 
Farm income 
8.99 
8.28 
3.97*** 
0.0001 
Farm experience 
27.84 
28.74 
0.43 
0.6689 
Oxen ownership 
1.16 
0.5970 
2.46 ** 
0.0148 
Source: own survey,
***, **, and * indicates that at 1%, 5% & 10 significance level
respectively
The descriptive and
inferential statistics results presented in Table 4 show that 94.05% of
adopters were male headed households. Regarding to participation in training
97.61% of adopters were participants of training and 84.52% of them were
members of cooperatives. On the other hand, 94.03% of adopters and 76.22% of
nonadopters participated in demonstration in 2014/2015 cropping season.
Table 3. Descriptive
statistics of Dummy/ discrete Independent Variables
Variables 
Percentage of adoption category 

Adopter 
Nonadopter 
χ2 value 
pvalue 

Sex of household head 


4.97** 
0.026 
Male 
94.05 
79.10 


Female 
4.95 
20.90 


Participation in training 


35.47*** 
0.000 
Yes 
97.61 
53.73 


No 
2.39 
46.27 


Cooperative membership 


10.45*** 
0.001 
Yes 
84.52 
58.21 


No 
15.48 
41.79 


Residence status 


2.51 
0.472 
Indigenous 
31.34 
35.8 


settler 
68.66 
64.2 


Participation in demonstration 


33.81*** 
0.000 
Yes 
94.03 
76.22 


No 
5.97 
23.78 


Source: own survey
(2017), ** and *** indicates 5% and 1% of significance probability level
This section presents the results of the logistic regression model which
is used to estimate propensity scores for matching adopted households with nonadopters.
As indicated earlier, the dependent variable in this model is binary variable
indicating whether the household has adopted improved soybean or not and the
outcome variable is the farm income. In the estimation, data from the two
groups; namely, adopters and non adopter households were pooled such that the
dependent variable takes a value 1 if the household is adopter of improved
soybean and 0 otherwise.
Table 7 shows the estimation results of the logit
model. The estimated model appears to perform well for our intended matching
exercise. The pseudoR^{2 }value is 0.4571, which is fairly low. A low pseudoR^{2} value means that household adopted improved soybean do not
have many distinct characteristics overall and as such finding a good match
between adopters of improved soybean and nonadopters of improved soybean
households becomes easier, and the pseudoR^{2 } indicates how well independent variables
explain the probability of adoption.
Table 4. Propensity
score matching estimation (logit model)
Variables 
Coefficient 
Std. Err. 
Zvalue 
P value 
Sex of household head 
0.9165 
0.4418 
2.07 
0.0380** 
Total family size 
0.0189 
0.0589 
0.32 
0.7490 
Education category 
0.1941 
0.2341 
0.83 
0.4070 
Religion of HH 
1.1022 
0.3741 
2.95 
0.0030*** 
Residence status of HH 
0.0724 
0.2648 
0.27 
0.7840 
Distance to market 
0.1015 
0.0290 
3.50 
0.0000*** 
Distance to main road 
0.1449 
0.1377 
1.05 
0.2930 
Cooperative membership 
0.8185 
0.3785 
2.16 
0.0310** 
Dependency ratio 
0.1452 
0.6119 
0.24 
0.8120 
Credit service 
0.0005 
0.0034 
0.13 
0.8930 
Frequency of visit 
0.0044 
0.0072 
0.61 
0.5410 
_cons 
3.3186 
1.0639 
3.12 
0.0020*** 
Number of obs 
134 



LR chi2 (11) 
84.90 



Prob > chi2 
0.0000*** 



Pseudo R2 
0.4571 



Log likelihood 
50.429 



Note: ***, **, *, show significance at 1%, 5% and 10% level, respectively
Figure 4 portrays the distribution of the household with
respect to the estimated propensity scores. In case of treatment households, most
of them are found in partly the middle and partly in the right side of the
distribution. On the other hand, most of the control households are partly
found in the center and partly in the left side of the distribution.