Market Participation Impacts of Improved
Wheat Varieties in Ethiopia:
Applications of Standard and Generalized
Propensity Score Matching Methods
Asfaw Negassa and Bekele Shiferaw
To be Presented at Wheat for Food Security in Africa Conference
UNECA Conference Hall
October 8-12, 2012
Addis Ababa, Ethiopia
Outline of Presentation
I. Background
II. Objectives of the Study
III.Empirical Model
IV.Data Source
V. Empirical Results
VI.Conclusions and Implications
I Background: Why wheat in Ethiopia?
● Wheat is among the very important staple food crops
grown in Ethiopia
More than 4 million farm households are directly dependent on wheat
production (CSA, 2011)
Wheat is the third most important sources of per capita calorie
supply next to maize and sorghum, accounts for more than 12% the
total food calorie supply (Berhane et al., 2011)
● Wheat consumption is increasing due to increase in
population, rise in urbanization and income growth while
increase in wheat price levels and variability have been
observed
● Given, wheat’s strategic importance in the national
economy, the Ethiopian government has been making large
investment in the development and extension of improved
wheat technologies—several improved varieties have been
released
I Background (Cont.)
● However, the market participation and commercialization
impacts of the adoption of improved wheat varieties has not
been explored so far
Lack of evidence regarding to what extent past research and
development efforts has helped the wheat producers to participate in the
market and in generating marketable wheat quantities—interaction
between technological change and market participation
● This has implications for the government’s effort to stimulate
wheat production through the adoption of improved wheat
varieties to generate increased marketed volume of wheat to
feed the growing urban population under the current conditions
of increasing wheat prices
II Objectives of the Study
The major objective of this study was to estimate the impact of adoption of improved wheat varieties on market participation and marketed volume of wheat for wheat producers in Ethiopia
Specific objectives:
1) To determine the difference in the effect of adoption of improved wheat varieties on likelihood of the farm households being in various net market positions (net buyer, autarkic, or net seller) of wheat and marketed volume of wheat
2) To determine the impact of area under improved wheat varieties on the extent of market participation and marketed volume of wheat among adopters
III Empirical Model
● The key challenge in empirical impact evaluation using
observational studies is how to obtain unbiased treatment
effect in the presence of confounding factors which could
affect both the chances of receiving the treatment and the
outcome itself
bias could arise when there are pre-treatment differences in observed
as well as unobserved covariates between control and treatment groups
as a result of non-random treatment assignment
Treatment Outcome
Confounding factors
III Empirical Model (Cont.)
● Quasi-experimental methods developed to provide adequate covariate balance between treated and control groups—create adequate counterfactual comparison groups for the treated groups so that any difference between the treated and control groups is due to the treatment effect
● Two methods used
Propensity score matching (PSM) method (Rosenbaum and Rubin, 1983) –to see the treatment effect difference between adopters and non-adopters
Generalized propensity score matching (GPSM) method (Imbens, 2000; Hirano and Imbens, 2004) —to see the treatment effect difference among the adopters due to differential levels of technology use
IV Data Sources
● Data: For this study, cross-sectional survey data involving
nationally representative 2096 sample farm households
randomly selected from eight major wheat growing agro-
ecological zones of Ethiopia
● Covariates:
Household head characteristics (age, sex and education)
Household characteristics (family size and dependence ratio)
Household resources (land and cattle)
Institutions (access to formal and informal financial services)
Agroecological zones
V Empirical Results
A Results of PSM ● PS Matching quality (adequacy of counterfactual
comparison group) T-test of mean difference for individual covariates between
treated and control groups before and after matching Before matching – significant in 5 of 20 cases After matching significant only in 2 of 20 cases
Overall covariate balance test
Criteria Before
matching
After matching
NNM with
caliper
KBM
Pseudo R2 0.043 0.005 0.017
LR χ2 97.64 14.2 25.73
P-value χ2 0.000 0.819 0.138
Mean bias 7.8 2.8 4.2
Percent bias reduction 64 46
Impacts of adoption of improved wheat varieties on market participation
Outcome variable by
matching algorithm
Estimated outcome Average treatment effect (ATT)
Treated Controls Point estimate 95% confidence
interval
Unmatched comparison
Net buyer (%) 7 8 -1 --
Autarky (%) 26 42 -15 --
Net seller (%) 66 49 16 --
Marketed volume (kg) 367 163 204
NNM method
Net buyer (%) 7 9 -2 (2) -5 - 2
Autarky (%) 26 38 -11(3)*** -19 - (-5)
Net seller (%) 66 52 14(4)*** 5 - 22
Marketed volume (kg) 360 166 194 (28)*** 139 - 250
KBM method
Net buyer (%) 6 10 -4(3) -9 - 1
Autarky (%) 26 38 -11(4)*** -20 - (-3)
Net seller (%) 67 52 15(5)*** 5 - 25
Marketed volume (kg) 355 162 193(43)*** 109 - 277
B Results of GPSM
● GPS matching quality
Covariate balance violated in 27% of the cases before
matching
Covariate balance violated in 11% of the cases after
matching
● Dose-response functions
● Treatment effect functions
Figure 1 Impact of adoption of improved wheat
varieties on farm households’ probability of being net
buyer of wheat
-.05
0
.05
.1
.15
.2
Pro
ba
bili
ty o
f be
ing
ne
t bu
yer
of
whe
at
0 1 2 3Area under improved wheat varieties (ha)
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = logit
Dose response function
-.3
-.2
-.1
0
.1
.2
Marg
inal c
han
ge
in p
roba
bili
ty o
f b
ein
g n
et buye
r o
f w
hea
t
0 1 2 3Area under improved wheat varieties (ha)
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = logit
Treatment effect function
Figure 2 Impact of adoption of improved wheat
varieties on farm households’ probability of being
autarkic in wheat net market position
0
.2
.4
.6
Pro
babi
lity
of b
eing
aut
arki
c
0 1 2 3Area under improved wheat varieties (ha)
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = logit
Dose response function
-.5
0.5
1
Mar
gin
al c
han
ge o
f pro
bab
ility
of b
ein
g a
utar
kic
0 1 2 3Area under improved wheat varieties (ha)
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = logit
Treatment effect function
Figure 3 Impact of adoption of improved wheat
varieties on farm households’ probability of being net
seller of wheat
.2
.4
.6
.8
1
Pro
babili
ty o
f bein
g n
et se
ller
0 1 2 3
Area under improved wheat varieties (ha)
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = logit
Dose response function
-1-.
50
.5
Marg
inal c
han
ge
in p
roba
bili
ty o
f b
ein
g n
et se
ller
0 1 2 3
Area under improved wheat varieties (ha)
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Probability of a positive outcomeRegression command = logit
Treatment effect function
Figure 4 Impact of adoption of improved wheat
varieties on Marketed volume of wheat
0
500
1000
1500
2000
Mark
ete
d v
olu
me o
f w
heat (k
g)
0 1 2 3Area under imporved wheat varieties (ha)
Dose Response Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Dose response function
-1000
0
1000
2000
3000
Marg
inal c
hange in
mark
ete
d v
olu
me o
f w
heat (k
g)
0 1 2 3
Area under improved wheat varieties (ha)
Treatment Effect Low bound
Upper bound
Confidence Bounds at .95 % levelDose response function = Linear prediction
Treatment effect function
VI Conclusions and Policy Implications
● Significant difference between adopters and non-
adopters in terms of their market participation and
marketed volume of wheat
● Increasing the adoption of improved wheat varieties
decreases the likelihood of farmers being net buyers,
decreases the likelihood of being autarkic and
increases the likelihood of being net seller of wheat
and increases the market supply of wheat
● The results provide strong evidence for positive but
heterogeneous effects of adoption of improved wheat
varieties on farm households net market position and
marketed volume of wheat
VI Conclusions and Policy Implications (Cont.)
● Thus, given the current level of adoption of improved wheat
varieties at less than 70% among the farm households and
actual wheat area under improved varieties is also low, there is
a need to improve the farm households’ level of adoption of
improved wheat varieties in Ethiopia
● This study also indicates that the binary variable treatment of
adoption status of improved wheat varieties in impact
assessment assumes that the adopters are homogeneous
group in terms of their adoption and leads to inaccurate impact
estimates and wrong conclusions and implications –impact
varies by adoption status and level of adoption (area of
wheat under improved wheat varieties)
Thank You