EFFECT OF THE ADOPTION INTENSITY OF IMPROVED BREAD

WHEAT PACKAGE AND PRODUCTIVITY CHANGES: APPLICATIONS

OF CRAGG’S DOUBLE HURDLE AND REGRESSION ADJUSTMENT
MODELS

Yirgalem Eshete Fithaneges1*, Bamlaku Alamirew Alemu2,

1
Addis Ababa University, College of Development Studies, Center for Rural Development.
Email: [email protected]. ORCID: https://orcid.org/0000-0002-8143-5692. (+251) 913
655606

2
Associate Professor at Yom Institute of Economic Development, (+251) 911666820, P.O. Box
62539, Addis Ababa, Ethiopia. [email protected].

1
Abstract

Smallholder farmers’ can enhance their productivity by adopting the entire components of the
improved bread wheat package (IBWP). However, farmers often make partial adoption of the
package, and there is limited research on the factors influencing adoption intensity and its
impact on productivity. To address this gap, we conducted interviews and discussions with 450
smallholder farmers and experts selected using a multistage stratified sampling approach. We
employed Cragg's Double-Hurdle and Regression Adjustment Treatment Effect models to
identify determinants of adoption intensity and its effect on productivity. The coefficients from
the probit and truncated models turned out quite substantial indicating the significance of the
variables. Lack of access to improved fresh seeds and insufficient labor were identified as major
reasons for non-adoption of the entire package. Furthermore, we found that farmers who
adopted the entire IBWP had a lower productivity of 8.24 quintals per hectare compared to the
average of 40.21 quintals per hectare, assuming no reliance on unimproved bread wheat seeds.
This suggests that full bread wheat package adoption significantly boosts productivity.
Increasing access to improved bread wheat seed through seed aid and credits, encouraging full
package adoption, and introducing affordable, laborsaving, and improved machinery are
strongly recommended.
Key Words: Complete adoption, bread wheat package, seed aid, productivity, treatment Effect

2
1. INTRODUCTION

In the developing countries, agriculture is the most important sector for spurring growth,
eradicating poverty, and ensuring food security. Improving crop productivity is not only essential
but a necessary condition to break the vicious poverty cycle under which the majority of the
population in poor countries have persisted for millennia (Dereje, 2021). The role of wheat
production is critical in ensuring global food security because approximately 1.2 billion people
are wheat dependent, 2.5 billion wheat consuming poor men, women, and children live on less
than 2 USD per day, and it supports the livelihood of approximately 30 million poor wheat
producers and their families. Wheat is grown on more than 240 million hectares exceeding other
commercial crops (Bekele et al., 2013, CIMMYT, 2012). In Ethiopia, wheat is cultivated on
1,696,907.05 hectares of land, which is 13.38% of the total grain area. Cereal production
contributes 87.48% (about 26,778,976.40 tons) of the grain production. Wheat contributes
15.17% (4,642,965.71 tons) to the total cereals production. The national average wheat
productivity was 2.76 tons per hectare. The average wheat productivity in Ethiopia in general
and in ANRS (Amhara National Regional State) in particular is very low as compared to the
global average (CSA, 2018). Improving wheat productivity, therefore, is essential in the fight
against food insecurity and poverty in Ethiopia (Bekele et al., 2013, Dawit, 2012). With the
intention of bridging the potential and the observed productivity gaps in wheat, wider diffusion
and scaling-up of the Improved Bread Wheat Package (IBWP) has become the major concern of
government intervention in the country (Tariku, 2012). Yet, more than 90% of the seed used by
the smallholder farmers is supplied by the traditional, local, and informal seed supply system
(Abebe and Lijalem, 2011, Bishaw et al., 2012). Besides, the adoption of the whole practices in
the IBWP significantly enhances the productivity and food security of smallholder farmers (Ali
et al., 2015). Yet, farmers often make partial adoption of the IBWP at the expense of certain
components of the package. Thus, there is a need to investigate the determinants that affect the
adoption and Adoption Intensity (AI) of the IBWP and this study aimed at addressing this need.

Moreover, the nexus between productivity changes and the adoption of the different levels of the
IBWP is not sufficiently addressed by empirical researches. Thus, this study also aimed at
investigating the determinants of smallholder farmers AI of IBWP and its nexus with

3
productivity. Besides, most previous empirical studies were flawed and had been dissimilar in
the choice of analytical models for adoption. Binomial or dichotomous regression models are
commonly employed to elucidate the probability of adoption or non–adoption. However,
dichotomous models fail to capture how much of the technology is adopted or rejected. Such
understanding of the probability of technology adoption does not provide sufficient information
about farmer's behavior as the farmer may allot some percent or 100 percent of their farm for the
new technology. Thus, a strict binary model is insufficient for examining the adoption intensity
of agricultural technologies (Gujarati and Dawn, 2009, Tobin, 1958). The Tobit model was
suggested as a remedy for the failures of the dichotomous model and frequently applied in the
analysis of the probability of adoption and adoption intensity of technologies with the
assumption that these two decisions are affected by the same set of factors. Yet, the decision to
adopt or reject and how much to adopt (adoption intensity) may not be necessarily affected by
similar variables. In this study, the adoption decision and how much to adopt are considered as
joint decisions and should be addressed together. Furthermore, the productivity change across the
different levels of adoption intensity has remained beyond the horizon of empirical investigation.
As a result, farmers do not know how much of the IBWP should adopted to hone their
productivity. The current study has employed the Regression Adjustment Treatment Effect
model to analyze the productivity change across the varying adoption intensity of IBWP.
Therefore, this study has a methodological contribution by applying Cragg's Double–Hurdle
model along with the Regression Adjustment model, which have been rarely used together.

1.2. Objectives of the study

This study had the following objectives:

1) To analyze the major determinants of smallholder farmers adoption intensity of improved
bread wheat package
2) To investigate the productivity changes in bread wheat across the varying levels of
improved bread wheat package adoption intensity.

4
2. RESEARCH METHODS

2.1. Description of the study area

The East Gojam zone in Ethiopia comprises 18 districts and is situated approximately 300 km
and 260 km away from the country's capital city of Addis Ababa and the regional capital city of
Bahir Dar, respectively. Its coordinates are 10° 19' 60.00" North and 38° 00' 0.00" East. The
zone has a total population of 2,358,051, with 1,154,740 males and 1,203,311 females. The
primary occupation of farmers in the area is mixed farming, and the zone is dominated by mixed
agricultural systems (Bureau of Finance and Economic Development, 2013). The map of the
study areas is presented in Figure 1 below.

Figure 1: Map of the study area

2.2. Sampling technique and sample size determination

The sampling strategy used in this study involved a combination of multi-stage purposive and
random sampling methods, as well as Probability Proportional to Size (PPS) sampling. East
Gojam Zone of the Amhara National Regional State was purposefully selected as the first stage,
and the specific study areas of Baso Liban and Debre Elias districts were selected in the second

5
stage. These districts were chosen due to their potential for wheat production, availability of
representative sample respondents, and the researcher's familiarity with the study areas. From the
37 kebeles in the two districts, eight rural kebeles were selected as a representative sample. PPS
sampling was then used to select three kebeles from Debre Elias and five kebeles from Baso
Liban as specific study areas. As indicated in Figure 2.1 above, Yelemelmo, Lemechem,
Dendegeb, Kork, and Dogemo kebeles were considered from Baso Liban district while Guay,
Yekegat, and Debre Elias zuria kebeles were considered from Debre Elias district. The total
sample size of the study was determined to be 450 farmers using a sample size determination
formula (Yamane, 1967).

2.3. Data Collection Methods

To gather primary data, this study employed structured interview schedules, Focus Group
Discussions (FGDs), Key Informant Interviews (KIIs), and personal observations. These
methods were used to collect both qualitative and quantitative data from a sample of smallholder
farmers involved in wheat production, development agents, crop experts, and key informants.
The data gathered from these primary sources were then supported and triangulated with
available secondary data sources including regional and district annual reports, research findings,
journals, publications, thesis, dissertations, books, proceedings, and other relevant sources.

2.4. Data Analysis Methods

2.4.1. Specification of the Double - Hurdle Model

This study has two important outcome variables. The first dependent variable is the adoption of
Improved Bread Wheat Seed (IBWS) that assumes a simple yes or no (adopter or non-adopter)
answer, which is a classic dichotomous variable. The second dependent variable is the Adoption
Intensity of the IBWP, measured using adoption index scores. The adoption index for each
respondent is calculated using equation 1 below following the formula by Tiruneh et al. (2015)
and slightly modified to fit the realities and technologies considered under the current study.
n
AI i=∑ ¿ ¿
i =0

6
Where, AI1i-is the adoption index of the i th farmer measured based on the level of adoption of
each recommended component divided by the total number of recommended practices
considered in this study (TNP), ATF/RTF-Refers to the ratio of Applied Tilling Frequency to
Recommended Tilling Frequency, ASP/RSP-Ratio of Applied Row Spacing to Recommended
Row Spacing, AHC/RHC-Ration of applied Hand Weeding frequency to Recommended Hand
Weeding Frequency, IWSU-refers to Improved Wheat Seed Use (1 if s/he has used, 0 otherwise),
ASR/RSR- ratio of Applied Seed Rate to Recommended Seed Rate, ANPSRATE/RNPSRATE-
Ratio of Applied NPS Rate to Recommended NPS Rate, AUREARATE/RUREARATE- Ratio of
Applied Urea Rate to Recommended Urea Rate.

The model of this study is based on the micro-level data, where each of the sample respondents
(Decision Making Units) must choose whether to adopt or reject the IBWP and then its intensity
if adoption. Several studies employ the conventional binary dependent variable models like Logit
and Probit models in studying the probability of adopting or rejecting a technology (Finger and
El Benni, 2013, Mariano et al., 2012). However, the objective here goes one-step beyond mere
understanding of probabilities of adopting or rejecting to understanding how much of the
components in the wheat production package are adopted. As a result, this study reverts to apply
the Double Hurdle model, which is developed by (Cragg, 1971). The DH model which is a
parametric generation of the (Tobin, 1958) was applied in this study due to the assumption that
adoption and adoption intensity of IBWP are influenced by different sets of variables. Thus, the
two decisions in this study were the decision to adopt Improved Bread Wheat Seed (IBWS) and
the decision how much of the practices in the IBWP to adopt (Adoption Intensity). The first
decision variable (Ƴ) takes the value 1 for farmers who have adopted improved bread wheat seed
and 0 otherwise. However, the expected utility of adopting a technology is a latent variable.
Thus, the first decision (adoption hurdle) equation can be written as:
¿
Υ i =β 1 X 1 i + β 2 X 2i +. . . .+ β11 X 11i +ε 1 i ............... .[2]

( )
¿
1if Υ i >0
Υ i=
0 Otherwise

¿
Where, Υ i is the latent variable for the decision to adopt Improved Bread Wheat Seed (IBWS)
by the ith farmer, β 1 to β n are the coefficients for the explanatory variables X 1 i to X ¿ and ε 1i is the
error term.
7
The Double Hurdle model estimates the probability of adoption (Ƴi = 1) and the level of
adoption intensity (Wi) separately, and then combines them to estimate the overall impact of the
explanatory variables on adoption and intensity of adoption of the IBWP. Therefore, the AI (the
intensity hurdle) of IBWP is specified in a Tobit like function as specified below:
¿
W i =Z X 1 i + β 2 X 2 i+ .. . .+ β 11 X 11i +ε 1 i .........................(3)

¿
W i =Z i β i + μi … … … … … … … … … … … … … … … … … … … … … … … .(4)

( )
¿ ¿
¿ 1W i >0∧Υ i >0
W i =W i =Z i βi + μi if
0 Otherwise

Where Wi here is the observed responses (Adoption index that refers to the ratio of the total
number of actually applied recommendations to the total number of recommendations included
in the IBWP), Z is a vector of the household characteristics while β is a vector of parameters.
Farmers decision whether to adopt or reject and their AI can be modeled jointly if they are
simultaneously made by farmers while it can also be modeled independently if they are made
separately or sequentially if one decision is made first and affects the other one (this is the
dominance model) (Finger and El Benni, 2013). If the independence model applies, the error
terms are distributed as follows: Ɛi ~ N (0, 1) and Ui ~ N (0, δ2). If both decisions are made jointly
(the Dependent DH), the error term can be defined as (Ɛi Ui) ~ BV N (0, Y) where

[ ]
δ
1 ρ
Υ= ....................................................................[5]
ρδ δ 2
If the sample household’s adoption decision and its intensity tend to have relationships, then the
model is said to be a dependent model and such relationship can be expressed as follows:
Cov (ϵ i ui)
ρ= ................................................... [6 ]
√ Var (ϵi )Var (ui)
If ρ = 0 and there is dominance (the zeros are only associated with non-participation, not
standard corner solutions) then the model decomposes into a Probit for participation and standard
OLS for Y. Following (Tobin, 1958) we assume that the error terms and εi and ui are
independently and normally distributed and thus we have the following expression:
ϵ
⌊ i ⌋ (N ) ⌊ , ⌊
ui ()
0 1 0
0 0 δ2
⌋ ⌋ ................................................................ [7 ]

8
Finally, the observed variable in a DH model is Wi = Y iWi* and the log-likelihood function for
the DH model is:

LogL=∑ ln ⌊ 1−Φ ( χ i α )Φ
0
( zδ ) ⌋ +¿∑ ln A=⌊ Φ (z α ) 1δ ( Υ −δ χ ) ⌋ ............[8]
i

1
i
i i

Therefore, we employ the Double – Hurdle model to estimate smallholder farmers’ adoption
decision and the AI. A simple specification test that evaluates for Cragg’s DH-model against the
Tobit model can be used using the same set of explanatory variables, through a comparison of
the log-likelihood function values of the Tobit, Probit and Truncated models estimated.
Assuming that the same set of independent variables appears in all the three equations, the
following value λ will be distributed as a chi-square random variable with degrees of freedom
equal to the number of explanatory variables under the null hypothesis that the Tobit model is the
correct (Goodwin et al., 1993, Heckman, 1979).

λ=−2 { Ӻ Tobit −Ӻ Probit −Ӻ Truncated } … … … … … … … … … … … … … … … … . ( 9 )

Where, Ӻ i ’s represent the respective estimated log-likelihood function values.

2.4.2. The Average Treatment Effect Estimation

In an experimental impact evaluation approach, the causal effect of technology adoption on

smallholder farmers’ productivity has been investigated by comparing the mean productivity
between adopter and non – adopter groups. This approach, however, assumes a pure random
assignment of farmers’ to treatment and control groups. It also fully attributes any difference in
outcome (productivity under this case) between the groups to the treatment (Improved Bread
wheat Seed adoption here) at the complete exclusion of pre-existing differences between adopter
and non – adopter groups (Reed et al., 2021, Tesfaye et al., 2016). The other important
assumptions in the experimental approach is absence of spillover effects, stable treatment effects,
compliance, validity of assumptions, and outcome measures. The no spillover effect assumption
refers to the fact that the treatment has no effect on farmers in the control group or farmers’
outside the study. Isolating the impact of the treatment (adoption of IBWS) on the treated group
(adopters of IBWS) is plausibly possible if the assumption of no spillover effect does hold water
(Morgan, 2018). The stable treatment effects assumption implies that the effect of the treatment

9
don’t vary on temporal differences that allows for generalization of the treatment effect beyond a
specific study period (Morgan, 2018, Reed et al., 2021).

The assumptions of the experimental impact evaluation explained above does not hold water
under the current study. First, technology (IBWS) is not randomly assigned, but shaped by
farmers’ own decision and self-selection. Second, the decision to adopt or reject the technology
is determined by a mix of agronomic, socio-economic, institutional, and psychological variables
and these factors may directly affect farmers’ level of bread wheat productivity. If adoption of
IBWS has an effect on farmers productivity, one cannot safely distinguish whether the effect on
productivity is caused by the adoption of the IBWS or other covariates considered in the study
such as farm experience, gender, amount of credit, frequency of extension contact, and other
agronomic factors. Identifying the exact factor contributing to the productivity of the smallholder
farmers’ demand the application of ATE models. Following Rosenbaum and Rubin (1983), the
‘treatment effect’ can be defined as the difference between bread wheat productivity farmer i
receives in the two states of the world:

(i) The farmer that adopts IBWS, A = 1, and

(ii) The farmer that does not adopt IBWS, A = 0:

ATE(λ ¿¿ i)=Υ i ( 1 )−Υ i ( 0 ) ...............................................................(10)¿

The Average Treatment Effect (ATE or λi) is defined as the expectation of the treatment effect
across all farmers. A problem arises when using non-experimental data because only one of these
states is actually observed; that is, either Yi (1) or Yi (0) is observed for each farmer i, but not
both. The unobserved productivity is called the counterfactual productivity. Accordingly, it is co
nvenient to express the ATE (λi) as:

1 0
Ε( λi )=P∗⌊ Ε(Υ )∨ A=1−Ε (Υ ∨ A=1)⌋ +¿

Where P is the probability of observing a farmer adopting the IBWS. In Equation (11), the ATE
for the whole sample is the weighted average of the technology (treatment) effect for IBWS

10
users and non-users. In estimating the ATE, both counterfactual productivity, E (Y0|A = 1) and E
(Y1|A = 0), should be constructed. Intending to avoid complications, several studies emphasize on
one or the other of the respective counterfactuals. The average IBWS adoption effect on the
adopter category or the Average Treatment Effect on the Treated (ATT) is the most important
evaluation parameter and is given by:

λ i=Ε ( λ∨A=1 ) =Ε [ Υ (1)∨A=1 ] −Ε [ Υ (0)∨ A=1 ] . . . .. . . .. . . .. . . .. . . .. . . .. .(12)

Now, given Equation (12) above, one can directly see the problem of selection bias as the second
term on the right hand side is unobservable. If E[Y (0) | A = 0] = E[Y (0) | A = 1], non-adopters
of IBWS can be used as an adequate comparison group. Yet, this assumption is rarely satisfied as
this study rely on non-experimental data. According to Rosenbaum and Rubin (1983), matching
is the best way-out for such a problem. The rationale behind the matching exercise is that one
infers the behavior of a given technology adopter by matching them with an observationally
equivalent technology non - adopters. This would actually assist the researcher to have a similar
situation to what really would have been observed in a random experiment. To apply this
approach, however, the following two assumptions are expected to hold true.

The condition independence assumption: Given a series of observable covariate Xs, which are
not affected by technology adoption, the potential well-being is independent of technology
assignment:
Y0, Y1Џ T|X, ⱯX
Where Џ denotes independence. If the assumption is fulfilled, holding the observable covariates
constant, the non-adopter’s well-being has the same distribution that adopters would have
experienced had they not adopted the technology. Accordingly, technology can be deemed as
randomly assigned. It follows that the ATT can be expressed as follows:

λ ATT =E ( λ| A=1 ) =E [ Ƴ (1 )| A=1 , X ] −E ¿

However, it is quite difficult if there are multiple observable co-variates that affect smallholder
farmers’ IBWS adoption and productivity. For example, if X contains s covariates that are all
dichotomous, the adopter must be matched with a non-adopter having identical values for all 2 S
variables. This is obviously a problem when S is relatively large and/or the sample size is relativ

11
ely small. To deal with this dimensionality problem, Rosenbaum and Rubin (1983) suggest using
a balancing scores method. The method shows that if potential adoption is independent of
technology adoption conditional on covariates X, it is also independent of technology adoption
conditional on a balancing score. The propensity score, defined as the possibility of farmer
adopting technology conditional on covariates X, is one possible balancing score:
Pi ( X )=Prob ( A=1| X ) … … … … … … … … … … … … … … … … … … … … …(14)
This conditional probability is the propensity score, which allows one to identify similar farmers.
The second assumption relates to the common support or overlap condition. It rules out perfect
predictability of A given X, so that
0< P ¿

This condition guarantees that farmers with the same X values have a positive probability of
being both an adopter and a non-adopter. Accordingly, ATE is only defined in the region of this
common support. Farmers falling outside this region are not included in the estimation of ATE.

The common support assumption: improves the matching quality by excluding farmers at the
tails of the propensity score distribution. It ensures that characteristics observed in the
technology adoption group can also be observed among the non-adopters (Bryson et al., 2002). A
downside of the common support assumption is that it reduces the sample size. If the proportion
of lost farmers is too large, this might raise concerns that the remaining farmers are insufficiently
representative of the population, thus casting doubt on the associated ATE estimates. Assuming
that both the conditional independence and common support conditions hold, the propensity
score-matching estimator for ATT can be expressed as follows:

=E p ( x|A =1) { E [ Ƴ ( 1 )| A=1 , P [ X ] } −E [ Ƴ ( 0 )| A=0 , P ( X ) ] } … … … … ..(16)

PSM
ATT λ
Equation (16) reveals that the propensity score matching estimator is simply the mean difference
in the economic efficiency of the two groups (IBWS users and non - users) of the study area.

2.4.3. The Regression Adjustment Treatment Effect

Studies in randomized experiments, project and policy interventions, agronomic and clinical
trials often use the Average Treatment Effect (ATE) as a metric to compare outcomes between

12
treatment and control groups. The ATE measures the mean outcome difference between decision
– making units assigned in the treatment and control arms. Though the origin for the ATE is in a
pure experimental science, it has been frequently applied in observational studies. There are
different variants of ATE estimators that includes Propensity Score Matching (PSM), Nearest
Neighbor Matching (NNM), Augmented Inverse – Probability Weighting (AIPW), Inverse –
Probability – Weighted Regression Adjustment (IPWRA), Inverse – Probability Weighted
(IPW), and Regression Adjustment (RA) (Li et al., 2016, Becker and Ichino, 2002, Grilli and
Rampichini, 2021). The RA estimator is the best option ATE estimator in times when the
dependent outcome is continuous (like productivity, wage, birth weight) while the treatment
variable is binary (like adoption status expressed as adopter versus non - adopter, survival status
expressed as dead versus alive) with a mix of various outcome independent variables. The
interest in the current study is to estimate the effect of farmers’ adoption status of IBWS on their
productivity controlling for the various outcome explanatory variables. Thus, the RA variant of
the ATE can best fulfill the interest of this study.

3. RESULT AND DISCUSSION

3.1. Characteristics of Adopters and Non-adopters

Table 3.1 below summarizes the mean, standard deviation, t-values, and p-values of the
continuous variables. Adopters and non-adopters had shown a significantly high difference in the
calculated mean values of their adoption index as suggested by a t-value of –30.62 at p = 0.000.
Adopters and non – adopters of this study were found to be quite different in terms of the various
socio – economic, institutional, and agronomic variables. Adopters were found to be older with
more years of experience in bread wheat production. They also own a greater number of
livestock and have greater plot and family size than non – adopters. Yet, no significant difference
was manifested in the grade level of adopters and non – adopters as can be proved from a
calculated t – value of 1.3901 (p = .9171). Moreover, these two categories had no any
outstanding difference in terms of the amount of credit. Adopters, on average have received only
273.19 Ethiopian birr per year, which do not bring any notable difference with non-adopters. In
addition, adopters had better frequency of contact with the extension workers than non-adopters.
Additionally, a statistically discernible difference was observed between these two groups in
terms of making use of different agronomic practices. For instance, adopters, on average had

13
practiced a hand weeding frequency closer to the recommendation (a twice hand weeding
frequency) than non-adopters. Similarly, adopters have been applying bread wheat seed rates,
fertilizer, and hand weeding frequencies closer to what is recommended than non-adopters. A
quite agreeable finding was also reported by other studies (Abeyo et al., 2020). A χ2- test has
been carried out to check if there exists a significant difference between adopters and non–
adopters for all categorical variables considered for the study. Accordingly, an insignificant
difference has been witnessed between adopters and non-adopters in terms of gender (χ 2 = .781,
p= .377). Similarly, adopters and non-adopters were not significantly different in terms of their
membership in seed multiplication and marketing cooperatives and sowing methods (Appendix
Table A1).

As can be inferred from Table 3.1, the adoption index showed a significant difference between
adopters and non-adopters with a t-value of -30.62 and p-value of 0.000. This suggests that
adopters have a higher level of adoption of the recommended practices than non-adopters do.
The study also found that adopters were older and had more years of experience in bread wheat
production compared to non-adopters. Adopters also had a greater number of livestock and larger
plot and family sizes. However, there was no significant difference in the grade level of adopters
and non-adopters, and the amount of credit received. Adopters had better frequency of contact
with extension workers than non-adopters, and they made use of different agronomic practices
closer to what is recommended, such as hand weeding frequency, bread wheat seed rates,
fertilizer, and hand weeding frequencies. This finding is consistent with other studies. The study
also conducted a chi-square test to check for significant differences between adopters and non-
adopters for all categorical variables considered in the study. There was no significant difference
between adopters and non-adopters in terms of gender, membership in seed multiplication and
marketing cooperatives, and sowing methods. The results are presented in Appendix Table A1.

14
 Table 3. 1: Descriptive statistics for Improved Bread Wheat Seed adopters and non-adopters
Variables Adopter Non-Adopter Combined Mean Mean Dif. t- value p value
Mean Std. Err Mean Std. Err
Adoption Index .698 .009 0.1992 .01304 .5491 -.498 -30.617 .0000
Productivity 41.525 .22845 31.791 .2287 38.6267 -9.734 -25.533 .0000
Age of the respondent 47.06 .454 3.723 .1399 42.1822 -16.38 -21.934 .0000
IBWP use Experience 13.95 .4538 7.381 .6604 10.907 -10.229 -14.554 .0000
Grade level 1.6519 .1567 2.0522 .2416 1.7711 .4003 1.3901 .9171
Family size 7.294 .151 2.918 .0821 5.9911 -4.376 -18.397 .0000
Dependency Ratio .4147 .0109 .8317 .0066 .5389 .4171 23.9234 .0000
Tropical Livestock Unit 9.164 .2344 1.4641 .1292 6.8718 -7.699 -20.8201 .0000
Land covered with IBWS .9680 .0409 .1006 .0069 .7097 -.8674 -13.761 .0000
Hand Weeding Frequency 1.851 .02614 .9627 .0630 1.5867 -.8886 -15.4884 .0000
Row spacing 21.4241 .4188 6.1567 .1828 16.8778 -15.2673 -23.3289 .0000
Extension contact frequency 1.7911 .06665 .9478 .0611 1.54 -.8434 -7.676 .0000
Credit Amount 4085.133 291.2287 3811.948 406.5004 4003.784 -273.1852 -0.5463 .2926
Seed rate 171.3608 4.1548 248.694 10.4646 194.3889 77.33327 8.2883 .0000
NPS Per Hectare 157.5403 4.8505 70.4179 4.9278 131.5972 -87.1224 -12.6000 .0000
Urea Per Hectare 135.5918 4.2998 50.7127 3.1411 110.3167 -84.87909 -15.9400 .0000
Source: Own Computation (2022)

15
 3.2. Adoption of Agronomic Recommendations in Bread Wheat Package

Adoption of improved agricultural technologies has often been viewed as a matter of using one
or few components of recommended production packages. However, empirical evidence has
shown that the use of improved seeds alone, without a parallel application of recommended
fertilizers, agro-chemicals, and fundamental agronomic practices, does not fully exploit the yield
potential of improved crop varieties, especially improved wheat varieties (Idrisa et al., 2012,
Mideksa et al., 2017). Additionally, farmers may adopt agronomic and other recommendations
without actually using improved seeds. To address this issue, the International Maize and Wheat
Improvement Center (CIMMYT) developed the Integrated Bread Wheat Productivity (IBWP)
package, which includes a set of complementary inputs and agronomic practices aimed at
maximizing the yield potential of improved wheat varieties. This package has been widely
disseminated to smallholder farmers in different countries.

The current study found that adoption of the IBWP was promisingly high, with only 29.8% of
the total respondents failing to adopt it, while 70.2% adopted it. However, adopters were quite
heterogeneous in their level of adoption of the different components included in the IBWP. It
was evident that wheat growers in the study area did not fully adopt all the recommendations
included in the IBWP. This implies that smallholder farmers tend to adopt one or few
components of the recommended package without a parallel adoption of all complementary
inputs and agronomic practices. The findings of this study suggest that there is a need for
targeted extension services that emphasize the importance of adopting all the recommended
components of the IBWP. Furthermore, it is important to tailor these recommendations to the
specific needs and resource constraints of smallholder farmers in different regions, to ensure that
they are able to fully exploit the yield potential of improved wheat varieties.

Poor application of agronomic practices was frequently mentioned as an important factor behind
the low wheat yield in Ethiopia (Tiruneh et al., 2015, Mesay et al., 2013). To address this issue,
a package of improved bread wheat production recommendations intended to maximize the
bread wheat yield potential. Land preparation was one of the essential components included in
the IBWP recommendations, with a recommended tilling frequency of 3 to 5 times. However,
the use of these recommendations varied significantly between adopters and non-adopters of the
study, particularly in terms of land tilling frequency and application of other agronomic

16
practices, as shown in Table 3.1 above. The mean tilling frequency of non-adopters was
significantly lower than that of the adopters.

Row planting is another crucial recommendation in the wheat production package. However, this
study found that adopters and non-adopters differed significantly in their choice of sowing
methods, with adopters relying heavily on row planting, while non-adopters mostly used
broadcasting. Participants in the focus group discussions (FGD) revealed that a lack of sufficient
labor and inability to employ hired daily laborers were important factors pushing farmers to stick
to the traditional broadcasting method. The discussants also mentioned that this was a major
challenge they faced in implementing other recommended agronomic practices, including
multiple tilling and weeding practices. These findings suggest that there is a need for targeted
extension services to address the labor constraints faced by smallholder farmers in the study area.
This could include training on alternative laborsaving techniques such as the use of appropriate
machinery, intercropping, and crop rotations. Moreover, there is a need to tailor the
recommended agronomic practices to the specific needs and resource constraints of smallholder
farmers in different regions to ensure better adoption and implementation. Overall, this study
highlights the importance of understanding the underlying reasons for non-adoption of
recommended agronomic practices and tailoring extension services to address these challenges
effectively.

Agricultural technologies are often interdependent, and the application of one technology may
not result in the desired yield increase unless other complementary inputs are also adopted. This
is particularly true for improved seeds, which may only increase yields if accompanied by the
optimal amount of fertilizer and proper seeding rates applied (Mideksa et al., 2017, Gashaw et
al., 2018, Alemu and Abera, 2013). In the study areas, the recommended types of chemical
fertilizers were NPS and Urea, with the recommended application rates of 200 and 100
kilograms per hectare, respectively (ATA, 2015). However, as shown in Table 3.1, there was a
clear difference in fertilizer application rates between adopters and non-adopters. On average,
respondents applied 132 and 110 kilograms of NPS and Urea per hectare, respectively. Non-
adopters applied, on average, 87.12 kilograms less NPS and 84.88 kilograms less Urea than
adopters. When asked why they did not adhere to the recommended fertilizer rates, farmers
explained that they had followed the recommended rates for more than ten seasons but had

17
experienced diminishing yield gains over time, which prompted them to increase their fertilizer
application rates.

A study conducted in Kaffa zone of Southern Ethiopia found that wheat-growing farmers used
varying fertilizer rates, but generally below the recommended levels. The difference in fertilizer
rates could be attributed to various factors, including capital constraints, insufficient and
untimely distribution of fertilizers, and other socio-economic factors affecting the demand and
supply of chemical fertilizers. Therefore, the adoption of the IBWP recommendations should be
complemented by the parallel adoption of other complementary inputs and agronomic practices,
including the optimal application of chemical fertilizers, proper seeding rates, and other
recommended agronomic practices.

Effective management practices are essential for achieving maximum wheat yield. Studies have
identified various factors, including spacing between rows and plants, crop rotation, tillage
practices, and weed control methods, as critical in influencing wheat yield (Tariku, 2012, Kebede
et al., 2017). According to ATA (2015) recommendation, a spacing of 30 centimeters between
rows is recommended. However, a recent study revealed that the average spacing between rows
among farmers was only about 17 centimeters, which is narrower than the recommended
spacing. This difference in spacing between adopters and non-adopters was found to be highly
significant (t = -18.863, p = 0.0000). During key informant interviews and focus group
discussions, farmers identified lack of appropriate sowing machines, small land sizes, and labor
shortages as the primary challenges to maintaining proper spacing between rows. To control
weeds, which is also an important component of the wheat production package; farmers reported
using a combination of herbicides and hand weeding. Blending these two methods is highly
recommended since the disadvantages of one method can be compensated for by the advantages
of the other. For instance, weeds that are tolerant to chemicals can easily be avoided through
hand weeding (ATA, 2015). Farmers’ considered in this study reportedly used both chemicals
and hand weeding to control weeds.

3.3. Econometric Results

Results of the Cragg’s double hurdle model, which incorporates the probit model in the first tier
and truncated normal regression in the second tier, are summarized and presented in Table 3.2.
18
The Variance Inflation Factor (VIF) was calculated to check if there were problems of
multicolinearity among the quantitative variables considered in the study. Serious
multicolinearity problem has been observed for two of the variables considered in the study
(dwelling distance from the market and average annual income). To solve the problem, these
variables were dropped, and the test was re-conducted again. Finally, we have arrived at a
situation where all the VIF were less than 6.5 and the mean VIF turns out to be 3.8, indicating
multicolinearity is no longer a problem (Appendix Table A2). Besides, the Wald chi2 (14)
19542.69 with a p-value (Prob > chi2) = 0.0000 tells us that the Cragg’s DH model is statistically
significant as compared to the model with no predictors.

Among the surveyed farmers, the majority of them (70.2%) cultivated bread wheat using IBWS
against 29.8% who have used local seeds or seeds recycled more than 5 times. The highest
adoption status here could be attributed to improvement in farmer’s awareness due to extensive
efforts made by development workers and other stakeholders. Though a significantly high
segment of the farmers was adopters of IBWS, this study has witnessed a markedly high
difference in the adoption index of farmers within the adopter category itself. This implies that
adopters make a partial than complete adoption of all the components included in the IBWP. The
relatively better adoption of IBWS in the study area could be attributed to the strong farmer-
research-extension linkage that stayed for the past several years. Farmers of the study area have a
long-held tradition of working together with different governmental and Non-Governmental
Organizations (NGOs) including higher learning institutions, research centers, seed enterprises,
and the private sectors. The information and experience from these different sources along with
the benefits that farmers have enjoyed by working together with these stakeholders could
incentivize farmers’ adoption of the IBWS. The border effect of agricultural research centers and
academic institutions is reported in other studies too. For instance, and (Ragasa et al., 2013)
found that long-held customs of working together and proximity between stakeholders involved
in crop production ensure sustainable adoption of innovations and reduce technology dis–
adoption. The role of research and the seed sector is also well documented by (Bishaw and
Atilaw, 2016). Thus, the higher proportion of IBWS users in the study area is not surprising. This
finding also signals the need for efficient and sustainable linkage among the different actors
involved in technology development, diffusion, and adoption. The next section presents and
discuss the results of Cragg’s Double Hurdle model to see the relative influence of the socio-

19
 demographic, economic, institutional, and plot-level variables on adoption and adoption intensity
of IBWP.

3.3.1. Adoption and Adoption Intensity of Improved Bread Wheat Package

Research efforts that deal with mere identification of determinants of adoption of IBWP may not be
sufficient to come–up with the best policy actions. It is therefore highly imperative to understand the
relative influence of each determinant to allow policymakers and implementers in prioritizing their
interventions. Thus, this section deals with the influence of each variable on the probability of
farmers’ adoption decision and their adoption intensity (Table 2.2). Gender of the household head
was expected to have a significant effect on the decision to adopt or not and on how much of the
recommendations to adopt. Results of the Double Hurdle (DH) model turned out that gender of the
respondent has no significant influence on the adoption decision is (Tier 1). Contrary to its coefficient
in the first tier, its coefficient in the second tier turned out to be highly significant. This implies the
fact that while gender of the farmer had no a significant effect on the probability of adoption, its
effect on how much to adopt (AI) was significant (p<.1). This means that being male improves the
probability of AI by 0.014 units than being female. In other ways, this can be translated, as males
tend to adopt more components of the recommendations in the IBWP than their female counterparts
do.

There can be several explanations for the weak adoption intensity of female-headed farmers. During
the key informants’ interview and FGDs, farmers vehemently oppose the idea that there is gender
parity in the extension service provision. Therefore, women’s weak AI could be explained by lack of
inclusive agricultural extension service equally tailored to men and women. This result is consistent
with several other studies (Ogada et al., 2014, Gebre et al., 2019). The same authors revealed that the
probability of adopting inorganic fertilizer and improved maize variety is higher for male-headed
households and this possibly indicates that female – headed households are more resource constrained
to adopt and try new technologies. A similar argument and explanation could be adopted in this study
too. An opposite finding was reported by (Kebede et al., 2017) who claim that the chances of
adoption of agricultural technologies is higher for women than men, which suggests success in
targeting vulnerable female households in Ethiopia.

20
The age of the farmer had the same positive and significant effect both on the decision to adopt and
how much of the components in the IBWP are adopted. Keeping other factors constant, a one-year
increase in the age of the farmer increases the probability of adopting IBWS by 1.214. Similarly, the
coefficients in the second tier inform us that older farmers tend to adopt more components of the
recommendations in the IBWP. A one-year increase in farmer's age increases AI by .002. Yet,
Nazziwa-Nviiri et al. (2017) reported quite a different finding. They argued that aged farmers tend to
have a shrunken planning horizon that reduces their incentives to invest in the future while younger
farmers tend to be more flexible and interested in trying out new things. However, a study carried out
by Tariku (2012) has supported the findings of this study. This author asserts that longer age bestow
farmers a chance to have more resources than younger farmers as they accumulate enough resources
over the years that could enable them to be less risk-averse and cope with the risks associated with the
use of agricultural technology. Similarly, (Kebede et al., 2017) found that older farmers accumulate
more knowledge and are better able to exploit social networks, rendering innovations less risky,
capitalize knowledge and information, and reducing information inefficiencies. These explanations
could justify the current finding. In addition, a one-year increase farmers experience was associated
with a probability of adopting improved bread wheat seed with 2.457 units at p=0.001. Likewise, a
one-year increase in making use of IBWP increases farmers AI by 0.014 units. This may be because
increased experience in agriculture may bestow farmers with knowledge and information that
increase their rationality in making use of farm innovations. The finding in this regard is in agreement
with other findings (Altaye and Mohammed, 2013).

The coefficient for education was insignificant in both hurdles. Family size of the household was
found to have a significantly high effect on the adoption and AI of IBWP. A one-person increase in
the family size increases the probability of the adoption of IBWS by .504 units. Similarly, an
additional one-person increase in the family member has increased the AI by 0.3%. This is plausible
because the decision to adopt IBWP depends on labor availability, which is largely provided by
household members. Moreover, most of the recommendations included in the IBWP such as row
planting, hand weeding, repeated tillage, split fertilizer application, and so forth are quite labor-
intensive. Thus, farmers with more family members tend to have better adoption propensity than
those with fewer family sizes. The other probable reason could be the shortage of land sufficient to
satisfy the crop consumption demand of large family size. Therefore, it is logical that households with
larger family sizes undertake crop intensification to meet their crop consumption requirements

21
through the adoption of high external inputs such as seed, fertilizer, and other agrochemicals. Other
studies also concluded that large family size is normally associated with a higher labor endowment
that would enable a household to accomplish various agricultural tasks on a timely basis thus raise
productivity. Unlike the findings in this study, (Ghimire and Huang, 2015) found no correlation
between adoption and household size. Such a relationship between adoption and family size could be
attributed to the fact that larger family size does affect household disposable income and resource
allocation decisions. More family size could lead to subsistence needs leaving no or very few
resources available for purchasing farm inputs. The paradoxical effect of family size on adoption
could have different explanations. The first probable explanation is a difference in the type of crops
considered for the study. Some crops such as “tef” and wheat tend to be quite labor-intensive as
compared to maize. The use of hired labor and membership in labor sharing traditional cooperatives
for different farm activities could be the other explanation.

The dependency ratio ended up in thwarting both adoption and AI. A unit change in the dependency
ratio of the respondents has reduced the probability of IBWS adoption and the AI by 34.11 and 0.077
units, respectively. This means that higher dependency ratio has forced farmers to go for cheaper and
unimproved bread wheat production practices. There could be two plausible explanations for the
negative effect of the dependency ratio on adoption and AI. First, the dependency of several family
members on one or few members of the family may increase the drudgery and amount of active man-
hour spent on non–agricultural activities. This reduces the amount of family labor required to
implement the different labor-intensive recommendations included in the wheat production package.
Second, farmers with a higher dependency ratio tend to have urgent and high consumption demand
that they are compelled to reject technology or make a partial adoption of the technology that could
negatively reduce farmers AI. The effect of the dependency ratio witnessed under the current study
was supported by earlier studies too (Dawit, 2012, Teshome and Abate, 2013).

Number of animals owned by farmers have shown a significantly high positive impact on both
adoption status and AI. A unit change in the TLU has raised farmer's the probability of adoption and
AI scores by 0.381 and 0.044 units, respectively. The positive effect of TLU on adoption and AI
might be explained by the fact that farmers with better livestock sizes can sell animals and use it for
the purchase of farm inputs against those who do not own animals or own a few numbers of animals.
In the study area, animals are the most common liquid assets available at farmer’s disposal that serve

22
 as insurance against production failure and other production misfortunes. Similar studies made by
(Finger and El Benni, 2013, Nazziwa-Nviiri et al., 2017) concluded that livestock assets are important
risk-coping mechanisms that encourage the adoption of farm technology. Adoption of
recommendations inculcated in wheat production package demands availability of sufficient labor.
Another possible explanation for the significant effect of animals on AI would therefore be their role
in serving as a source of agricultural draught power. Therefore, it is not surprising that TLU has a
significantly high positive effect on the adoption intensity of IBWP. Though not inferred from the
data in this study, the complementarity of livestock and crop production justifies the direct nexus
between TLU and the adoption of crop technologies as crop residues can be used as animal feeds
while livestock manure could be used for soil conditioning and management (Gashaw et al., 2018).
Table 3. 2: Coefficients of Cragg’s Double Hurdle Model
Variables
Tier 1 (Probit Model) Coefficient Std. Err 95% Confidence Interval
Gender (Male) -2.557NS 1622.506 -3182.610 3177.496
Age of the Respondent 1.214*** 854.816 -1676.623 1674.195
IBWP use Experience 2.457*** 1362.560 -2673.025 2668.111
Grade level of the Respondent 0.504NS 190.978 -373.805 374.814
Family Size 0.504*** 1107.970 -2172.086 2171.077
Dependency Ratio -34.110*** 46419.370 -9.10e+04 90946.190
Tropical Livestock Unit 0.381* 291.819 -571.574 572.336
Land Size 155.490** 9954.660 -1.94e+04 19666.260
Frequency of Extension Contact 2.224*** 646.615 -1265.118 1269.566
Credit Amount -0.0001NS 0.186 -0.365 0.364
Cooperative Membership (No) -0.586*** 725.877 -1423.278 1422.107
_Cons 89.800 66552.010 -1.30e+05 1.31e+05
Tier 2 (Truncated Model)
Gender (Male) 0.014* 0.011 -0.007 0.036
Age of the Respondent 0.002*** 0.002 -0.002 0.006
IBWP Use Experience 0.001** 0.002 -0.005 0.002
Grade level of the Respondent -0.000NS 0.001 -0.002 0.002
Family Size 0.003*** 0.002 -0.001 0.008
Dependency Ratio - 0.077*** 0.074 -0.068 0.222
Tropical Livestock Unit 0.044** 0.003 -0.006 0.005
Land Size 0.095*** 0.022 0.052 0.137
Hand Weeding Frequency 0.002NS 0.009 -0.016 0.019
Sowing Method (Row Planting) 0.046*** 0.013 0.020 0.073
Row Spacing 0.003*** 0.001 0.001 0.004
Credit Amount -0.0001NS 0.000 -0.000 -0.000
Cooperative Membership (No) -0.008* 0.009 -0.026 0.010
_Cons -0.293 0.119 -0.526 -0.059
Sigma _Cons 0.062 0.002 0.057 0.066
23
Cragg hurdle regression Log likelihood = 559.773
Number of obs = 450 Pseudo R2 = .84125
LR chi2 (14) =1827.08 Wald chi2(14)= 19542.69
Prob > chi2 = 0.0000 Note: The coefficient for factor levels is the discrete change from the base level.
*** p<.01, ** p<.05, * p<.1 NS Not Significant

In addition, the coefficient in the first and the second hurdles for land were positive and highly
significant indicating higher land size enhances the adoption and AI of IBWP practices. A
hectare increase in plot size is associated with the probability of increasing adoption and AI of
IBWP by 155.49 and 0.095 units, respectively. This may be because greater land size encourages
farmers to try technologies on at least some portion of their land. Other studies also found a
positive association between plot size and the probability of adoption of various farm inputs.
Though it was not inferred from our data, large farm size and land tenure security may bestow
farmers the chance to lease out part of their landholding for some cash that could be used to
purchase farm innovations (Kebede et al., 2017, Mehmood et al., 2020). Moreover, this study
had expected a significant positive impact of credit on adoption and AI. Yet, the coefficient in
the first tier is insignificant indicating credit did not help farmers to adopt IBWS. This may be
because the amount of credit obtained by farmers is not sufficient to adopt all complementary
recommendations included in the IBWP. Poor access to credit, lack of sufficient capital, and
farmers’ poor purchasing power were also reported as major setback in farm technology
adoption (Kebede et al., 2017, Wondimu et al., 2015, Taremwa et al., 2022).

As expected frequency of monthly extension contact appears to have a noticeable effect on the
probability of adoption of IBWS. The significant contribution of the frequency of extension
personnel contact could happen due to special attention of the extension service to bread wheat
production than other crop commodities and technologies that were not introduced so far. Other
studies also indicated that lower farmer to extension agent ratio and the last mile penetration of
farmers training centers in different kebeles of the country have improved the adoption behavior
of farmers in Ethiopia (Tiruneh et al., 2015). Being a member of the seed multiplication and
marketing cooperative contributes to higher adoption intensity. Membership in seed
multiplication and marketing cooperatives increases the probability of adoption and increase
farmers AI. This is because wheat seed multipliers sell the produced wheat to seed enterprises at
a price higher than the prevailing market price. This could inspire farmers to dedicate a

24
substantial portion of their land to be covered by IBWP and hence higher AI. Besides, members
are often supplied wheat seeds in the right quantity and time by seed enterprises that it is easy for
cooperative member farmers to adopt the technology. The special support and advice provided
by extension workers and seed enterprises to cooperative members encourage adoption of
technologies (Dawit, 2012, Kebede et al., 2017, Ogada et al., 2014, Sisay et al., 2017). Besides,
the availability of local inputs supplier cooperatives increases the adoption of improved wheat
varieties by 36.77 kg (Manda et al., 2016, Sisay et al., 2017). These authors revealed that seed
companies' reluctance to expand their retail networks hinders improved crops adoption. The
exceptionally high adoption status and AI of IBWP witnessed in this study area can also be
explained by the availability of seed multiplication and marketing cooperatives closer to farmer’s
residence.

3.4. Productivity Changes along the Adoption Intensity Continuum

Based on the information provided, the study found a significant difference in productivity
between adopters and non-adopters of IBWS. Adopters had an average productivity of 41.53 ±
4.061 quintal/hectare, while non-adopters had an average productivity of 31.79 ± 2.647
quintal/hectare, indicating that IBWS adoption pushed average productivity by 9.73
quintal/hectare (See Figure 3.1 below). However, the study also found that the variance in
productivity within the adopter category was relatively higher than the non-adopter category.
This suggests that the adoption of IBWS alone, without complementary inputs and agronomic
practices included in the IBWP (Improved Bread Wheat Production) did not bring substantial
productivity improvement.

25
 Productivity (qt/ha)
Figure 3. 1: Frequency Distribution of productivity between adopters and non - adopters

To explore the relationship between productivity and adoption of IBWP components, the study
divided sample farmers into adoption index continuums, ranging from 0 (unimproved practices)
to 1 (the whole package is practiced). The results showed that only 4% of the study respondents
had applied the whole package, and their associated productivity was well above 47
quintal/hectare. Additionally, only 19.3% of respondents had an adoption index score greater or
equal to 0.8 (80%), while 51.6% of farmers had an adoption index of less than 0.5, indicating
that farmers did not make more than 50% of the recommendations included in the IBWP. The
study also found that productivity declined with a declining adoption index, indicating the need
for extension advisory services to assist farmers in adopting the whole components of the IBWP.
The study highlights that the productivity improvement attributable to the adoption of multiple
components of a grain production package is well documented (Teklewold et al., 2013, Bezabeh
et al., 2014, Kebede et al., 2017, Timu et al., 2014).

3.5. Productivity Effects of Improved Bread Wheat Seed Adoption

3.5.1. Checking the Balance in the Confounders

26
3.5.1.1. Standardized difference between confounders

The quality of the PSM analysis depends on the extent of the balance in the confounders of the
control and treatment groups (Cragg, 1971, Gujarati and Dawn C., 2009, Johnson et al., 2003,
Morgan, 2018). Before going deep into further analysis, comparison for all confounders between
the treated (adopters) and the untreated (non-adopters) was made. With the intention of checking
the existence of differences in the confounders, their mean and standard deviation was calculated
(See Table 3.1). It is evident from the same table that there is a difference in the confounders for
the adopters (treated) and non-adopters (untreated) groups. Yet, because the units of
measurement for the confounders is dissimilar, it is difficult to gauge whether if the difference in
one variable is more important than the others. One can go for a significance test. However, a
significance test is contingent upon sample size, and it may not indicate how big is the difference
between adopters and non – adopters. Therefore, we opt for the dissimilarity check based on the
standard deviations (standardized differences). The standardized difference in the mean value of
the confounders was calculated and presented in Appendix Table A3.

A simple comparison of the standardized difference between the treated and the untreated groups
do not indicate any notable difference in terms gender, grade level, frequency of extension
contact, and amount of credit. But, there was a conspicuous standardized difference in other
variables between the two groups in terms of age, farm experience, family size, dependency
ratio, Tropical Livestock Unit, plot size, seed recycling frequency, sowing method, weeding
frequency, spacing between rows, cooperative membership, amount of fertilizer, amount of
pesticide, and man-days. The output in the standardized difference between confounders show us
the treated and the untreated groups differ by the standard deviations indicated in the Appendix
Table A3. This implies the treated and the untreated tend to be similar in terms of gender than
they are in other confounders.

3.5.1.2. Matching confounders using Propensity Scores

The existence of standardized difference between the mean values of the confounders is not
sufficient. These calls for matching based on Propensity Score (PS). We have employed the
binary logistic regression model in estimating the propensity scores of the variables considered

27
in this study. The distribution of the propensity scores inclined to be higher for the treated than
the untreated as indicated in Annex Graph 1 (A). Because the value of the propensity score lies
in between 0 and 1, both distributions are skewed. Under such scenario, it is recommended to
make use of the log of the odds of the PS (often referred as linear predictor) against the normal
PS itself as it bestow the distribution to be more normally distributed as indicated in the Annex
Graph 1 (B) (Bryson et al., 2002). The difference between the treated and the untreated ended –
up to be statistically significant that make the matching process manageable by creating
homogenous classes. Details of the comparison of observed and predicted associations between
confounders and the treatment is clearly depicted in Annex Graph 2.

Moreover, assessing the balance attained after the PS matching requires consideration of several
other factors that are likely to provide additional insights into the matching quality and reliability
of the treatment effect estimates (Gujarati and Dawn C., 2009, Johnson et al., 2003, Morgan,
2018). The covariate overlap, statistical significance, the PS distribution, sub-group analysis, and
the common support were considered in this study to check the quality of the matching and
reliability of the treatment effect estimations. As indicated in Annex Graph 2, a sufficient overlap
was observed in the PS between adopters (treated) and non - adopters (non-treated) groups
indicating the existence of meaningful comparable units. Besides, from the distribution of the PS,
it was observed that the PS for the treated and the untreated groups is similar. This implies the
balance in the matching of the two groups. Besides, it was found that the treated and the
untreated groups shared a wider range of the PS indicating the fact that there are units in the data
that have a non-zero probability in either of the treatments (adopter or treated versus non –
adopters or untreated) as indicated in Annex Graph 3.

3.5.2. Effect of IBWS adoption on Productivity

Based on the information provided, the study employed the Regression adjustment treatment
effect model to estimate the average treatment effect of adopting Improved Bread Wheat Seed
(IBWS) while controlling for various variables such as gender, age, farm experience, education
level, family size, dependency ratio, TLU, land size, amount of capital invested, seed recycling
frequency, seed rate, hand weeding frequency, tilling frequency, type of sowing method, row

28
 spacing, frequency of extension contact, amount of credit, and membership in seed
multiplication and marketing cooperatives.

The study found that the average productivity of farmers relying on local and unimproved bread
wheat seed was lower by 8.24 quintal/hectare compared to those who adopted IBWS. However,
the study was also interested in knowing the average amount by which adopters' bread wheat
productivity increased because of adopting IBWS. The study used the Potential Outcome Mean
(POM) to estimate the expected average bread wheat productivity for both non-adopters and
adopters. The POM for non-adopters was 40.21 qt/ha, which means that if none of the farmers
relied on unimproved/local seeds, the expected average bread wheat productivity would have
been 40.21 qt/ha. On the other hand, the POM for non-adoption of IBWS was 31.96 qt/ha,
indicating that non-adopters produced 8.25 qt/ha less than adopters. Therefore, the study found
that adopting IBWS increased bread wheat productivity by an average of 8.25 quintal/hectare
compared to not adopting IBWS. This result suggests that IBWS adoption can significantly
increase bread wheat productivity and may be a promising solution for improving agricultural
productivity. Several empirical studies are in line with the current finding (Alston and Pardey,
2017, Alemu and Bishaw, 2015, Etwire et al., 2016, Kebede et al., 2017).
Table 3. 3: Estimation of the Effect of Adoption Status on Productivity
Productivity Robust std.err
POmeans for adopter groups Coefficien z p>|z| [95% Conf. Interval]
t
Adopters 31.963 .9863 32.41 .0000 30.0301 33.8964
Non-Adopters 40.21 .0000 .0000 .0000 .0000 .0000

4. Conclusion and Policy Implications

4.1. Conclusion

From the probit and truncated model results, it can be concluded that adoption decision and
adoption intensity (how much to adopt) are influenced by a mix of variables considered in the
study. Older farmers with a greater number of farming experience, higher number of family size
and land size, higher frequency of extension contact, membership in seed multiplication and
marketing cooperatives had positively contributed to farmers’ adoption decision. On the

29
contrary, higher dependency ratio and seed rates had discouraged farmers adoption of IBWS.
Albeit, farmers gender, grade level, and amount of credit did not show any marked effect on
adoption decision. One can also deduce that a substantial portion of the sample respondents
rarely adopt the whole components of the recommendations included in the IBWP and that is
associated with their poor productivity. It is also evident from the truncated model that being
male, older farmers with more years of farming experience, higher number of Tropical Livestock
Unit, land size, family size, and the application of different agronomic practices such as sowing
method and row spacing have positively affected how much of the components included in the
IBWP should be adopted. This can be interpreted as these variables encourage farmers to adopt
all or most of the components included in the calculation of the AI. The variance in the adoption
index score within the adopter groups was found higher than the variance within the non –
adopters’ group. This implies that adopters make partial adoption of the IBWP at the expense of
several complementary recommendations of the IBWP. Moreover, one can safely conclude that
the adoption of multiple components of the recommendations in the IBWP enhances farmers’
bread wheat productivity than the adoption of fewer components of the IBWP. From the POM
productivity difference between adopters and non – adopters of the IBWS calculated using the
Regression Adjustment it can be inferred that non – adoption of IBWS has a productivity
thwarting effect compelling non – adopters to produce 8.25 qt/ha less than adopters.

4.2. Policy Implications

The researcher makes the following recommendations with a strong conviction that upon their
implementation, the adoption intensity of IBWP will be considerably improved:
 Age and experience in making use of improved bread wheat production practices had
extremely enhanced the probability of adoption and adoption intensity. Thus, establishing
a platform through which farmers will learn from other farmers is strongly recommended
to boost both the decision to adopt and how much to adopt. Besides, farmer-to-farmer
experience sharing will ease extension agents' work burden in convincing farmers to
adopt the IBWP.
 Higher dependency ratio has significantly reduced farmers' adoption decision and their
AI. Thus, it is highly suggested that farmers with a significant proportion of dependent
family members should be supported with laborsaving technologies to improve their

30
 adoption and AI of the different labor-intensive components of the IBWP. Besides, the
design and implementation of alternative social welfare programs and projects that will
support farmers with higher dependency ratios are highly recommended.
 Lack of sufficient labour and inability to employ hired labour were reported as pushing
factors to stick to traditional bread wheat production practices such as broadcasting
sowing method and minimum tillage. The introduction of laborsaving and improved
sowing machineries and their supply with affordable prices is, therefore, strongly
recommended.
 Farmers in the study area were exposed to a single bread wheat variety. This implies the
narrow bread wheat variety supply problem. Thus, this study strongly recommends the
creation of wider bread wheat varietal options for farmers in the study area.
 A higher amount of TLU and land size had enhanced the probability of adoption and AI
of IBWP. Thus, this study highly recommends expanding farmers' access to diversified
asset ownership by implementing additional income-generating alternative livelihood
strategies.
 An entire bread wheat package adoption leads to an 8.25 quintal increment in
productivity than non – adoption. Thus, the agricultural extension advisory service should
focus its effort on enhancing an entire package adoption than partial adoption.

31
References
ABEBE, A. & LIJALEM, K. 2011. The recent development in seed systems of Ethiopia. . In: D.
ALEMU, S. K., AND A. KIRUB, (ed.) Improving farmer's access to seed empowering
farmers innovation. Addis Ababa, Ethiopia: JICA.
ABEYO, B., BADEBO, A., GEBRE, D. & LISTMAN, M. 2020. Achievements in fast-track
variety testing, seed multiplication and scaling of rust resistant varieties: Lessons from
the wheat seed scaling project, Ethiopia. CDMX, Mexico, and Addis Ababa, Ethiopia:
The International Maize and Wheat Improvement Center (CIMMYT) and the Ethiopian
Institute of Agricultural Research (EIAR).
ALEMU, D. & BISHAW, Z. 2015. Commercial behaviours of smallholder farmers in wheat seed
use and its implication for demand assessment in Ethiopia. Development in Practice, 25,
798-814.
ALEMU, L. & ABERA, D. 2013. Comparison of chemical fertilizer, fish offal’ s
fertilizer and manure applied to tomato and onion. African Journal of Agricultural
Research, 8, 274-278.
ALI, A., RAHUT, D. B., BEHERA, B. & IMTIAZ, M. 2015. Farmers' access to certified wheat
seed and its effect on poverty reduction in Pakistan. Journal of Crop Improvement, 29,
247-265.
ALSTON, J. M. & PARDEY, P. G. 2017. Transforming traditional agriculture redux.
ALTAYE, S. & MOHAMMED, H. 2013. Linking seed producer cooperatives with seed value
chain actors: Implications for enhancing the autonomy and entrepreneurship of seed
producer cooperatives in southern region of Ethiopia. International Journal of
Cooperative Studies, 2, 61-65.
ATA 2015. Agricultural Transformation Agency. A Training Manual on Cereal Production.
Addis Ababa, Ethiopia: Agricultural Transformation Agency.
BECKER, S. O. & ICHINO, A. 2002. Estimation of average treatment effects based on
propensity scores. The stata journal, 2, 358-377.
BEKELE, S., SMALE, M., BRAUN, H.-J., DUVEILLER, E., REYNOLDS, M. & MURICHO,
G. 2013. Crops that feed the world 10. Past successes and future challenges to the role
played by wheat in global food security. Food Security, 5, 291-317.
BEZABEH, E., HAREGEWOIN, T., HAILEGIORGIS, D., DANIEL, F. & BELAY, B. 2014.
Change And Growth Rate Analysis In Area, Yield And Production Of Wheat In Ethiopia.
. Int. J. Dev. Res. , 4(10), 1994 - 1995.
BISHAW, Z. & ATILAW, A. 2016. Enhancing agricultural sector development in Ethiopia: The
role of research and seed sector. Ethiopian Journal of Agricultural Sciences, 27, 101-130.
BISHAW, Z., STRUIK, P. C. & VAN GASTEL, A. J. G. 2012. Farmers' seed sources and seed
quality: 1. physical and physiological quality. Journal of Crop Improvement, 26, 655-
692.
BRYSON, A., DORSETT, R. & PURDON, S. 2002. The use of propensity score matching in the
evaluation of active labour market policies.
BUREAU OF FINANCE AND ECONOMIC DEVELOPMENT 2013. Development Indicators
of the Amhara Region. . In: DEVELOPMENT., B. O. F. A. E. (ed.). Bahir Dar: Amhara
National Regional State, .
CIMMYT 2012. Wheat for food security in Africa. Africa‘s food security and the changing
demand for wheat over the coming decades.

32
CRAGG, J. G. 1971. Some statistical models for limited dependent variables with application to
the demand for durable goods. Econometrica: Journal of the Econometric Society.
Econometrica: Journal of the Econometric Society, 829-844.
CSA 2018. Agricultural Sample Survey. Report on Area and Production of Major Crops (Private
Peasant Holdings, Meher Season). Addis Ababa, Ethiopia.
DAWIT, T. 2012. Profitability of Contractual Bread Wheat Seed Production in Mecha District of
Amhara Region, Ethiopia. DOI: https://doi.org/10.5513/jcea.v13i1.1077. Journal of
Central European Agriculture, 2012, 13(1), p.142-149, 13, 142-149.
DEREJE, D., DEGEFA, T. & ABRAHAM, S. 2021. Analyzing the contribution of crop
diversification in improving household food security among wheat dominated rural
households in Sinana District, Bale Zone, Ethiopia. Agriculture & Food Security, 10, 1-
15.
DEREJE, D., T. DEGEFA, AND S. ABRAHAM 2021. Analyzing the contribution of crop
diversification in improving household food security among wheat dominated rural
households in Sinana District, Bale Zone, Ethiopia. Agriculture & Food Security, 2021.
10(1): p. 1-15., 10, 1-15.
ETWIRE, E., ARIYAWARDANA, A. & MORTLOCK, M. Y. 2016. Seed delivery systems and
farm characteristics influencing the improved seed uptake by smallholders in Northern
Ghana. Sustainable Agriculture Research, 5.
FINGER, R. & EL BENNI, N. 2013. Farmers adoption of extensive wheat production and
Determinants and implications. Land Use Policy, 30, 206-213.
GASHAW, T., TULU, T., ARGAW, M., WORQLUL, A. W., TOLESSA, T. & KINDU, M.
2018. Estimating the impacts of land use/land cover changes on Ecosystem Service
Values: The case of the Andassa watershed in the Upper Blue Nile basin of Ethiopia.
Ecosystem Services, 31, 219-228.
GEBRE, G. G., ISODA, H., AMEKAWA, Y. & NOMURA, H. Gender differences in the
adoption of agricultural technology: The case of improved maize varieties in southern
Ethiopia. Women's studies international forum, 2019. Elsevier, 102264.
GHIMIRE, R. & HUANG, W.-C. 2015. Household wealth and adoption of improved maize
varieties in Nepal: a double-hurdle approach. Food Security, 7, 1321-1335.
GOODWIN, K. B., OFFENBACH, A. L., CABLE, T. T. & COOK, S. P. 1993.
Discrete/Continuous contingent valuation of private hunting access in Kansas. . Journal
of Environmental Management. , 38 (1), 1 - 12.
GRILLI, L. & RAMPICHINI, C. 2021. Propensity scores for the estimation of average treatment
effects in observational studies. Training sessions on causal inference, Bristol, 28-29.
GUJARATI, D. N. & DAWN C., P. 2009. Basic Econometrics McGraw-Hill International
Edition, 5th ed., Boston,, 269 - 261.
GUJARATI, D. N. & DAWN, C. P. 2009. Basic Econometrics McGraw-Hill International
Edition 5th ed., Boston., 260 - 261.
HECKMAN, J. 1979. Sample Specification Bias as a Selection Error. . Econometrica., 47 (1),
153 - 161.
IDRISA, Y. L., SHEHU, H. & NGAMDU, M. B. 2012. Effects of Adoption of Improved Maize
Seed on Household Food Security in Gwoza Local Government Area of Borno State,
Nigeria. . Global Journal of Science Frontier Research, 12 (5).

33
JOHNSON, N. L., LILJA, N. & ASHBY, J. A. 2003. Measuring the impact of user participation
in agricultural and natural resource management research. Agricultural systems, 78, 287-
306.
KEBEDE, D., KETEMA, M. & DECHASSA, N. 2017. Disparity in adoption of wheat
production technology packages in eastern ethiopia. Review of Agricultural and Applied
Economics (RAAE), 20, 22-29.
LI, S., VLASSIS, N., KAWALE, J. & FU, Y. 2016. Matching via Dimensionality Reduction for
Estimation of Treatment Effects in Digital Marketing Campaigns. IJCAI, 3768-3774.
MANDA, J., ALENE, A. D., GARDEBROEK, C., KASSIE, M. & TEMBO, G. 2016. Adoption
and impacts of sustainable agricultural practices on maize yields and incomes: evidence
from Rural Zambia. . J. Agric. Econ.,, 67 (1), , 130 - 153.
MEHMOOD, Y., RONG, K., ARSHAD, M. & BASHIR, M. K. 2020. Do liquidity constraints
influence the technical efficiency of wheat growers: evidence from Punjab, Pakistan. The
Journal of Animal and Plant Sciences, 27, 667-679.
MESAY, Y., TESAFYE, S., BEDADA, B., FEKADU, F., TOLESA, A. & DAWIT, A. 2013.
Source of technical inefficiency of smallholder wheat farmers in selected waterlogged
areas of Ethiopia: A translog production function approach. African Journal of
Agricultural Research, 8, 3930-3940.
MIDEKSA, T., LETTA, T., BAYISA, T., ABINASA, M., TILAHUN, A., HUNDIE, B.,
ALEMU, W. & ABERA, M. 2017. Bread wheat varietal development and release in
Southeastern Highlands of Ethiopia. American Journal of Biological and Environmental
Statistics, 4, 15-19.
MORGAN, C. J. 2018. Reducing bias using propensity score matching. Springer.
NAZZIWA-NVIIRI, L., VAN CAMPENHOUT, B. & AMWONYA, D. 2017. Stimulating
agricultural technology adoption: Lessons from fertilizer use among Ugandan potato
farmers, Intl Food Policy Res Inst.
OGADA, M. J., MWABU, G. & MUCHAI, D. 2014. Farm technology adoption in Kenya: a
simultaneous estimation of inorganic fertilizer and improved maize variety adoption
decisions. Agricultural and food economics, 2, 1-18.
RAGASA, C., DANKYI, A., ACHEAMPONG, P., WIREDU, A. N., CHAPOTO, A.,
ASAMOAH, M. & TRIPP, R. 2013. Patterns of adoption of improved rice technologies
in Ghana. International Food Policy Research Institute Working Paper.
REED, M. S., FERRÉ, M., MARTIN-ORTEGA, J., BLANCHE, R., LAWFORD-ROLFE, R.,
DALLIMER, M. & HOLDEN, J. 2021. Evaluating impact from research: A
methodological framework. Research Policy, 50, 104147.
ROSENBAUM, P. R. & RUBIN, D. B. 1983. The central role of the propensity score in
observational studies for causal effects. Biometrika, 70, 41-55.
SISAY, D. T., VERHEES, F. J. H. M. & VAN TRIJP, H. C. M. 2017. Seed producer
cooperatives in the Ethiopian seed sector and their role in seed supply improvement: A
review. Journal of crop improvement, 31, 323-355.
TAREMWA, N. K., MACHARIA, I., BETT, E. & MAJIWA, E. 2022. Determinants of access to
agricultural credit among smallholder rice and maize farmers in the eastern and western
provinces of Rwanda. Agro-Science, 21, 1-11.
TEKLEWOLD, H., KASSIE, M., SHIFERAW, B. & KÖHLIN, G. 2013. Cropping system
diversification, conservation tillage and modern seed adoption in Ethiopia: Impacts on

34
 household income, agrochemical use and demand for labor. . Ecological Economics, , 93,
85 - 93,.
TESFAYE, S., BEDADA, B. & MESAY, Y. 2016. Impact of improved wheat technology
adoption on productivity and income in Ethiopia. . Afr. Crop Sci. J. , 24, 127.
TESHOME, A. & ABATE, E. 2013. Wheat technologies from where to where? The case of east
Gojam zone of Amhara region, Ethiopia. Journal of Agricultural Economics and
Development, 2, 226-236.
TIMU, A., MULWA, R., OKELLO, J. & KAMAU, M. 2014. The role of varietal attributes on
adoption of improved seed varieties: the case of sorghum in Kenya. J. Agri. and Food
Sec, 3 (9).
TIRUNEH, S., YIGESU, A. Y. & BISHAW, Z. 2015. Measuring the effectiveness of extension
innovations for out-scaling agricultural technologies. African Journal of Agricultural
Science and Technology (AJAST), 3, 316-326.
TOBIN, J. 1958. Estimation of relationships for limited dependent variables. Econometrica:
journal of the Econometric Society, 24-36.
WONDIMU, T., TESHOME, A. & GEZAHEGN, B. Y. 2015. Economic Analysis of Tef Yield
Response to different Sowing Methods: Experience from Illuababora Zone, Ethiopia. .
Journal of Economics and Sustainable Development, , 6 (1).
YAMANE, T. 1967. Statistics: An Introductory Analysis, 2nd Edition, New York: Harper and
Row.

Appendix Tables
Appendix Table A1: Distribution of respondents by Categorical variables
Variables Adopter Non- 𝛘2 p-value
adopter
Gender (Men) 289 119 .781 .377
Female 27 15
Total 316 134
Sowing Methods
• Broadcasting 23 116 277.11 .000
• Row Planting 293 18
Total 316 134
Cooperative Membership
• Member 117 0 67.046 .000
• Non-member 199 134
Total 316 134

Appendix Table A2. Results of Multicolinearity Test

Variables VIF 1/VIF
Dependency ratio 6.50 0.153883
Land covered by IBWSs 5.77 0.173236
Family size 5.37 0.186080
Tropical Livestock Unit 4.64 0.215538
35
 Row spacing 4.56 0.219074
Tilling frequency 4.02 0.248609
Certified seed use experience 3.30 0.302754
Urea rate 3.05 0.327377
Herbicide rate 2.71 0.368436
Seed recycling frequency 2.50 0.400732
Frequency of extension 1.68 0.595698
contact/month
Seed Rate 1.64 0.609766
NPS Rate 1.63 0.615190
Grade level 1.10 0.907768
Credit 1.05 0.950411
Mean VIF 3.26

Appendix Table A3: Mean and standardized difference of confounders in the treated (Adopters) and untreated (Non - adopters)

Confounders/Variables Mean in Treated Mean in Untreated Standardized difference

Gender 0.90 0.93 -0.087
Age 43.71 27.33 2.222
Farm experience 10.94 3.24 1.707
Grade level 1.75 2.00 -0.092
Family size 5.98 2.45 1.830
Dependency ratio 0.51 0.84 -1.524
Tropical Livestock Unit 7.32 0.25 2.253
Land Size 0.78 0.00 1.523
Seed recycling frequency 3.50 11.90 -2.647
Tilling frequency 3.73 1.31 2.562
Weeding frequency 1.75 0.45 1.966
Sowing method 1.84 1.00 3.215
Spacing between rows 18.08 5.24 1.970
Fertilizer rate 141.66 33.88 1.743
Frequency of extension contact 1.62 0.74 0.973
Credit (Birr) 4149.03 2592.86 0.337
Cooperative membership 0.42 0.05 0.981
Seed rate 180.82 326.19 -1.514
Pesticide amount 0.60 0.09 1.374
Amount of labour (Man-days) 90.72 35.40 3.167
Warning: Significant imbalance exists in all variables except gender

36
 kdensity pscore

kdensity lp
.25

.15

.05
8

.2

.1
Annex Graph 1: Distribution of the Propensity Score

0
0 .2 .4 .6 .8 1 -5 0 5 10
x x

kdensity pscore kdensity pscore kdensity lp kdensity lp

(A) Normal Propensity Score (B) Logs – odd Propensity Score

37
 Annex Graph 2: Comparison of observed and predicted associations between confounders and treatment
.65

.55
.7

.6

.5

.8

.6

.4

.2
1

-.5
.8

.6

.4

.2
1

.5
1

0
0 .2 .4 .6 .8 1 0 5 10 15 0 5 10 15 20 20 40 60 80
Sex of the household head Family size of the respondent Tropical livestock unit Age of the hh

.8

.6

.4

.2
.85

.75

.8

.6

.4

.2
lowess TREATMENT gender lowess pscore gender .9 lowess TREATMENT FAMSIZE lowess pscore FAMSIZE lowess TREATMENT HHTLU lowess pscore HHTLU lowess TREATMENT HHAGE lowess pscore HHAGE

.8

.7

1
-.5
.5
1

0 5 10 15 20 0 2 4 6 8 10 0 1 2 3 4 0 10 20 30
Bread Wheat Seed recycling frequency Grade level of the hh Land covered with CBWS Certified bread wheat seed use experience

lowess TREATMENT SRFREQNCY lowess pscore SRFREQNCY lowess TREATMENT HHGRADE lowess pscore HHGRADE lowess TREATMENT LACBWS lowess pscore LACBWS lowess TREATMENT CBWSEXPE lowess pscore CBWSEXPE

.8

.6

.4

.2
1
.8

.6

.4

.2
.8

.6

.4

.2
1
.8

.6

.4

.2
1

0 .2 .4 .6 .8 1 1 2 3 4 5 0 .5 1 1.5 2 1 1.2 1.4 1.6 1.8 2

Dependency ratio Tilling frequency Frequency of hand weeding Sowing method practiced

lowess TREATMENT DERATIO lowess pscore DERATIO lowess TREATMENT TILLFREQ lowess pscore TILLFREQ lowess TREATMENT HWEEDFRE lowess pscore HWEEDFRE lowess TREATMENT SOWINGMT lowess pscore SOWINGMT

38
 .85

.75

.65
.9

.8

.7
.8

.6

.4

.2
1

.8

.6

.4

.2
1

0
0 100 200 300 400 500 0 50 100 150 0 .2 .4 .6 .8 1
Amount of wheat seed used in kg Labour employed for wheat production in Mandays 0 = No & 1 = Yes

.85

.75

.65
.8

.6

.4

.2

.8

.7
lowess TREATMENT MSEEDCOP lowess pscore MSEEDCOP

1
lowess TREATMENT seedrate lowess pscore seedrate lowess TREATMENT manday lowess pscore manday
.8

.6

.4
1

5 10 15 20 25 30 0 5000 10000 15000 20000

0 2 4 6
Frequency of extension contact Row Spacing Credit amount in ETB
.9

.8

.7

.6

.5

lowess TREATMENT ROWSPACIG lowess pscore ROWSPACIG lowess TREATMENT CREDITAMT lowess pscore CREDITAMT
1

lowess TREATMENT FREXTCON lowess pscore FREXTCON

.8

.6

.4

.2
1

0 .5 1 1.5 2 0 200 400 600

Amount of herbicide applied per hectare in liters Amount of NPS used per hectare

lowess TREATMENT herbph lowess pscore herbph lowess TREATMENT npspha lowess pscore npspha

39
Annex Graph 3: Common support for propensity score estimation with a kernel matching technique

40