TABLE OF CONTENTS
2. REVIEW OF THE EMPIRICAL EVIDENCE
3. MATERIALS AND METHODS
3.1. Analytical Framework for Evaluation of Adoption of Wheat Variety Impact on Productivity
3.2. Data and Variables
4. RESULTS AND DISCUSSIONS
4.1. Descriptive Statistics
4.2. Propensity Scores Estimation using Probit Model
4.3. Assessing Matching Quality
4.4. Average Treatment Effects Estimation
5. CONCLUSION AND RECOMMENDATION
Appendix 1. Descriptive statistics of important variables used in the probit model
Appendix 2 . Propensity score matching quality test
Appendix 3 . Average treatment effects estimation using different propensity score matching estimators
Appendix 4. Distribution of propensity scores of full adopters and non/partial-adopters
First, We thank God for everything that He has done for us.
Next, We thank some of our former and current bosses-Mr. Kaleb Kelemu, Mr. Tesfaye Haregewoin, Mr. Eyob Bezabih, Mr. Yalew Mazengia and Mr. Mekonnen Hailu-for their advice, support and encouragement during the course of this work.
I, Mr. Fitsum Daniel, am also very grateful to my beloved wife, W/ro Chaltu Woldesenbet, to my 3 & 1/2 years old brilliant son, Natanim Fitsum, as well as to my 7 months old smart daughter, Bemnet Fitsum , for their understanding and support in the absence of which it might be difficult at best to complete the work by this time.
This list is not exhaustive and our thanks also go to our family, our wives' family as well as to all friends who supported us in one way or another and who wished us a success.
Like in many other sub-Saharan Africa countries, agriculture in Ethiopia is a basis for the entire socioeconomic structure of the country and has a major influence on all other economic sectors and development processes and hence it plays a crucial role in poverty reduction (Elias et al., 2013; GebreEyesus, 2015). Despite the marginal decline in its share of GDP in recent years, it is still the single largest sector in terms of its contribution to GDP as agricultural GDP constitutes 41% of total country's GDP (CSA, 2014/15).
As to Gebru 2006 citing CSA 2003, out of the total production of agriculture, about 70% comes from crop production. According to Abegaz 2011, cereal crops constitute the largest share of farming household’s production and consumption activities. Accordingly citing Alemayehu et al., 2009, only five major cereals (barley, maize, sorghum, teff and wheat) account for about 70% of area cultivated and 65% of output produced. Fertilizer use is also concentrated on cereals followed by pulses and oilseeds respectively according to Endale 2011 citing CSA 1995/96-2007/08. On the other hand, according to Endale 2011, data from the Ethiopian Seed Enterprise show that improved seeds are mostly used in wheat and maize cultivation with an average of 89 and 42 thousand quintal in the period 1994/95 to 2005/06, respectively. Moreover, Abegaz 2011 citing the Household Income, Consumption and Expenditure Survey of CSA indicated that the five major cereal crops account for 46% of household’s total consumption. Therefore, a closer look at what is happening in cereal production has an important welfare and policy implication in Ethiopia (Abegaz, 2011).
According to Ketema and Kassa 2016 citing Shiferaw et al. 2013, wheat contributes about 20% of the total dietary calories and proteins worldwide. Ethiopia is the second largest wheat producer in sub-Saharan Africa next to South Africa (Nigussie et al., 2015). Mann and Warner 2017 citing Minot et al. 2015 indicated that there are approximately 4.7 million farmers growing wheat on approximately 1.6 million hectares representing between 15 and 18% of total crop area and less than 1% of all wheat production takes place outside the four main regions of Ethiopia according to recent estimates. Wheat is one of the major staple crops in the country in terms of both production and consumption (Kelemu, 2017). According to Kelemu 2017 citing FAO 2014, it is the second most important food in the country behind maize in terms of caloric intake.
The Ethiopian agricultural sector, as to Gebru 2006 citing EEA 2004, is dominated by small-scale farmers cultivating about 96% of the total area under crop, producing more than 90% of total agricultural output and 97% of food crops. With these statistics, one can easily infer to what extent the small-scale farmers are the key element in strengthening the effort towards agricultural growth and consequently to the overall economic growth (Gebre-Selassie & Bekele).
On the other hand, most smallholder farmers (i.e. 59% of total cultivated area) reside in the moisture reliable cereal-based highlands among the five agro-ecological regions of Ethiopia distinguished by agricultural researchers (Taffesse et al., 2012). Accordingly, with farmers using virtually no irrigation, reliable rainfall is an important condition to achieve good agricultural productivity. In relation to this, as to the same source document, the Meher rainfall season is overwhelmingly important as it contributes about 96.9% of total crop production and 95.5% of total cereal production in 2007/08.
With respect to all these facts, it is not questionable that accelerated and sustained growth in the country’s agriculture in general and in the crop sub-sector in particular with special emphasis to the small-scale farmers will greatly help to achieve the various goals of the country (Gebru, 2006; MoFED, 2003; Gebre-Selassie & Bekele).
Moreover, food needs as well as the industrial demand for agricultural products increase due to population growth (Bor and Bayaner, 2009). All these needs, according to them, require an increase in the agricultural production. The growth in agricultural production in sub-Saharan Africa in the past was achieved by expanding the amount of land cultivated (Gebru, 2006). In relation with this, it is well known that in our country there are regions where there are large populations but limited land and vice versa (MoFED, 2003) . Accordingly, most of the lands available for settlement are found in the lowlands that lack basic infrastructural facilities and pose serious health hazards. With little suitable land available for expansion of crop cultivation, especially in the highlands, future cereal production growth will need to come from increasing land productivity mainly through the supply, duplication and diffusion of continuously improving technology and information (Ayele et al. 2006 citing Reardon et al 1996; Taffesse et al. 2012; Elias et al. 2013; Matsumoto and Yamano, 2010).
Cognizant of these as well as the fact that productivity is the major component of growth and a fundamental requisite in many form of planning irrespective of the stage of development and economic and social system as to Gebru 2006 citing Cheema 1978, the national wheat research program has released and disseminated a number of bread and durum wheat varieties since the 1950s and 1960s as to Ketema and Kassa 2016 citing Tesfaye et al. 2001. According to the same source citing CSA 2015b, a closer look at the proportion of the area covered by improved varieties of different crops showed that wheat took the second rank (7.4%) next to maize (46.4%) among cereals. Given the emphasis of increasing crop production through higher fertilizer use, import of chemical fertilizer augmented from 246,722 MT in 1995 to 375,717 MT in 2006 despite the removal of fertilizer subsidies since 1997/98 according to Endale 2011 citing MOARD 2007/08. In this regard, according to Ketema and Kassa 2016 citing CSA 2015b, wheat is the most fertilized crop (82%) among all crops and pesticide application is also most common on wheat as compared to that on other cereal crops.
Even though crop productivity and production remained low and variable in the 90s for the most part, there have been clear signs of change over the past decade (Abate et al., 2015). According to Kelemu 2017, the average level of wheat productivity for the period of 2000-2014 is about 1.73 ton/ha while the average growth rate in productivity is about 5.93%. During the same period, total wheat production has been increasing at 10.14% growth rate per annum (Kelemu, 2017).
As to Tsusaka and Otsuka 2013 citing FAO 2011, although the production of staple food has been increasing in sub-Saharan Africa, the rate of increase has not been high enough to outstrip its high population growth rate as a result of which per-capita agricultural production in the region has declined by about 10% since 1960. These all obviously calls for a further and a better growth in agricultural productivity as well as quality with minimum adverse impact on the environment. Kelemu 2017 citing Shiferaw and Okelo 2011 indicated that of the cereals whose production is soon likely to exceed domestic demand requirements, wheat is the commodity that will most easily find an export market to supply. In view of this prospect, according to him, the need for increasing productivity of wheat is very crucial.
Holistic and appropriate evaluation of the efforts and corresponding results as well as reasons/strengths and weaknesses/ of the past few decades in general and of the past recent years in particular is necessary in order to create a more fertile ground for the fast achievement of the aforementioned goal. In this regard, the role of historical data collected by different agencies like CSA as well as of different socio-economic studies carried out to provide vital policy and related recommendations is indispensable. Studies that assess the contribution of improved crop management practices information and technologies like improved crop varieties for the productivity growth of such important and widely cultivated cereals like wheat in Ethiopia in the past recent years are among studies that can be cited in relation to this. However, studies carried out in the country on this issue are not only few but also restricted to piece meal or location specific approach. As a result, the issue has not been satisfactorily and comprehensively assessed at a regional and national level.
Thus, the objective of this study is:
(1) to identify the impact of adoption of improved wheat varieties and information regarding improved wheat management practices on wheat productivity per unit of land cropped in Ethiopia as well as
(2) to identify the regional as well as agroecologic zonal disparity in the impact of adoption of improved wheat varieties and information regarding improved wheat management practices on wheat productivity per unit of land cropped among the four major administrative regions of Ethiopia which are also known to be the major wheat producing regions in the country.
2. REVIEW OF THE EMPIRICAL EVIDENCE
Tesfaye et al. (2016) conducted an impact assessment study by collecting data through multi stages purposive sampling techniques. In doing so, three districts of Arsi zone of Oromia Regional State, Ethiopia namely Arsi-robe, Digelu-tijo and Hetosa were purposively selected at the first stage based on wheat production potential and presence of wheat technology interventions. At the second stage, two Kebeles were randomly selected from each district and the probability proportional to sample size technique was then performed to ultimately select a sample of 177 farmers (consisted of 122 male-headed and 55 female-headed households). The researchers used age, education, farming experience, household head sex, family size, off-farm income, land fragmentation, livestock ownership, access to credit, and wheat disease as independent variables in the estimation of the propensity scores. After matching of observations from the treated and control groups was carried out based on their propensity scores, the effects of adoption of improved wheat varieties on productivity (the wheat production obtained from one hectare) and income of farm households were estimated through two different matching methods-the nearest neighbor and the kernel-based matching methods. Ultimately, a balance test was conducted to compare the similarities of the subsample of control cases with the treated cases. The causal effect of adoption of improved wheat varieties on wheat productivity is highly significant and is equal to about 10-11 quintals per hectare, which is the average wheat productivity difference between adopters and non-adopters, i.e. adopters were significantly (P<0.01) better than non-adopters by about 1 tonne ha-1 in wheat productivity. The study goes beyond the usual binary variable treatment of adoption status of improved wheat varieties in impact assessment. Accordingly, among the adopters, farmers who allocated less than 25%, 25-50%, 50-75% and 75-100% of their wheat area to improved wheat varieties obtained about 1.1, 2.3-2.5, 1 and 1-1.1 t ha-1 more than the non-adopter counterparts, respectively. The result shows that productivity gains from adoption of improved wheat varieties are the highest for those farmers who allocated from 25 to 50% their wheat farms to improved wheat varieties. The study also suggests that the effect of adoption of improved wheat varieties on farm household income was significant (P<0.1) and equal to 0.295-0.401, implying that the income of adopters was almost 30-40% higher than income of non-adopters. The sampled wheat farmers were similarly stratified by quintiles based on the proportion of the area under improved wheat varieties. Accordingly, those farmers who allocated 20-80% of their wheat area to improved wheat varieties earned almost two times more than the non-adopters and the impact was significantly higher in household heads who allocated 60-80% of their wheat farm to improved wheat varieties. However, the effect of technology adoption on household income for farmers who allocated less than 20% and 80 to 100% their total wheat area to improved wheat varieties was not significant, but positive.
Mulugeta T. and Hundie B. (2012) similarly conducted a study in Lode Hetosa district (woreda) of Arsi Zone which is administratively under Oromia Regional State, Ethiopia that aims to assess the impact of improved wheat technologies (improved wheat varieties and row planting method) on households’ food consumption level. Even though various types of crops, such as barley, tef, maize, horse beans, field peas, and various types of oil seeds, are cultivated and livestock such as cattle, sheep, goats, pack animals, and poultry, are important sources of livelihoods in the area, wheat is the major crop produced in the area as, for instance, it covered about 33% of the total cultivated land of Arsi Zone in 2007 (CSA 2007b). This might be because the weather condition of the district is suitable for wheat production as its temperature varies between 100C-250C, its annual rainfall ranges from 800mm to 1400mm and the average rainy days are about 120 days in a year. Their study relies dominantly on primary data which were collected through a household survey that employed a two-stage sampling technique in identifying the sample units/farm households. After purposively selecting five major wheat producer kebeles among the 19 rural kebeles of the district, 40 households were randomly selected from the member list of each kebele summing up to 200 households. They included more than a dozen of selected variables in the probit model to estimate propensity scores. Three commonly used matching algorithms, namely nearest neighbor matching, radius matching, and kernel-based matching, were also employed to assess the impact of improved wheat technologies. A matching quality check was also done in order to check that the distribution of variables are ‘balanced’ across the adopter and non-adopter groups. According to their analysis, adopters were found to be significantly superior to their counterparts, the non-adopters with respect to farm size (or land holding), livestock holding (TLU), participation in off-farm activities, access to institutional credit, and education. Moreover, adoption of improved wheat varieties planted in spacing was found to positively and significantly affect food consumption level of households with the increase in food consumption per adult equivalent per day ranging from 265 (11 %) kilocalories to 509 (23%).
Elias et al. (2013), on the other hand, conducted an impact assessment study using data obtained from a household survey undertaken in May and June 2012 in 3 randomly selected kebeles (i.e., Enerata, Kebi and Wonka) among the total 25 kebeles found in Gozamin woreda (district) of East Gojjam Zone in the Amhara Regional State, Ethiopia. The woreda, with average annual rainfall ranging from 1400-1800 mm and with an annual daily average temperature ranging between 11 and 25 degree C, covers 3 agro-ecological zones with 19% highland, 65% midland and 16% lowland. Main crops grown in the woreda in order of abundance include teff, wheat, maize, barely, check pea, soya bean, oats, niger seed (Neug) and lentil. They selected the woreda purposively for satisfying the following criteria: where crop production is widely practiced, where extension program have been implemented for relatively longer period of time, the availability of different agro-ecologies and its representativeness to the Ethiopian highlands as the Ethiopian highlands comprise nearly 45% of the total land area of 1.12 million square km and support over 85% of the country's 82 million population that are overwhelmingly rural. They employed a multi-stage stratified random sampling technique to select a total of 300 respondent farm households. In so doing, farmers in each selected village were first stratified into two groups as participant and non-participant of the extension program. Secondly, the two groups obtained from the first stage sampling were further stratified into male and female-headed households to ensure, as much as possible, representation of female-headed households in the sample. Unlike the others, this study used a combination of three methods (a benchmark Ordinary Least Square, Heckman’s Treatment Effect and Propensity Score Matching) to assess the effect of participation in agricultural extension program on farm productivity. In using propensity score matching (PSM) method, the propensity score (pscore) for each observation is first calculated using logit model for agricultural extension program (AE) participation. Moreover, in order to check the robustness of the ATT estimates, all the four matching algorithms-nearest neighbor matching, radius matching, kernel matching as well as stratification matching-were employed to pair each AE participant to similar non-participant using propensity score values. The result of the study from Heckman Treatment Effect Model (HTEM) shows that participation in AE increases farm productivity by about 20%. Unexpectedly the HTEM estimation for the effect of AE participation on productivity is higher compared to OLS estimation (6%), which was estimated without treating the endogenity of extension participation. All the matching estimators of the PSM method show that participation in agricultural extension program has a positive and statistically significant effect (ranging b/n 18 & 33%) on farm productivity. Generally, all the estimated results obtained from the different models confirm that AE participation in the study area have increased farm productivity. However, the overall level of farm productivity observed in this study for the three case study crops (teff, wheat and maize) is still low compared to the target yield set by the regional extension program based on farmers’ field conditions and research stations. However, the authors acknowledge that their results cannot be generalized to the national level since the sample was not representative of the entire country.
Endalkachew T. (2011) undertook a study through using a survey data collected by Environmental Economics Policy Forum for Ethiopia (EEPFE) in 2007 from two zones, East Gojam and South Wollo in Amhara Region, Ethiopia by which a total of 1,760 households who has 5,871 plots were interviewed. He implemented both non parametric (propensity score matching (PSM) method) and parametric methods (endogenous switching regression method) of analysis to study the impact of Soil and Water Conservation (SWC) on crop productivity. In the utilization of PSM in the study, the researcher first estimated a probit regression in which the dependent variable equals one if the plot adopted at least one SWC technology, zero otherwise and then checked the balancing properties of the propensity scores. Furthermore, he used the nearest neighbor matching (NNM) method and the radius matching (RM) method. The researcher also estimated endogenous switching regression model to control for unobservable selection bias and to assure the results of PSM are robust. The analysis revealed that adoption of SWC technologies has a significant negative impact on value of crop productivity. Accordingly, adoption of SWC had declined the value of crop productivity by about 16% for NNM, which is significant at 1% level of significance, and by 3.5% for RM, which is significant at 10% level of significance, on average compared to the non-adopters. This indicated that (assuming there is no selection bias due to unobservable factors) crop productivity for plots which adopted SWC technology is significantly lower than the non adopters. The result from the endogenous switching regression similarly indicated that the mean value of crop productivity of SWC adoption is statistically lower than had they not been adopt which is consistent with the result from propensity score matching. Accordingly, SWC adoption decreases crop productivity by about 15%. For non-adopters the mean crop productivity would have been decreased by 37% had they adopted SWC and the results are statistically significant at 1%.
Asfaw et al. (2010) tried to estimate the causal effect on marketed surplus of adopting improved chickpea varieties in three districts namely Minjar-Shenkora, Gimbichu and Lume-Ejere which are found in the central highlands of the country and are located around Bisheftu/Debre Zeit town. In doing so, the researchers used a data that originates from a survey conducted by the International Crop Research Institute for Semi-Arid Tropics (ICRISAT) and Ethiopian Institute of Agricultural Research (EIAR). After a reconnaissance survey was first conducted by a team of scientists to have a broader understanding of the production and marketing conditions in the survey areas and the findings from this survey were used to refine the study objectives, sampling methods and the survey instrument, the household survey was then carried out in March, 2008 by using a multi-stage sampling procedure to select districts, kebeles and farm households. In the first stage of this procedure, the three districts were selected from the major legume producing area based on the intensity of chickpea production, agro-ecology and accessibility as these districts represent one of the major chickpea growing areas in the country where improved varieties are beginning to be adopted by farmers. Finally, eight kebeles from each of Gimbichu and Lume-Ejere districts and ten kebeles from Minjar-Shenkora district were randomly selected from where a random sample of 700 households was selected for detailed household survey. The causal impact of technology adoption on market integration (marketed surplus) is analyzed by utilizing (1) two-stage standard treatment effect model, (2) regression based on propensity score as well as (3) matching techniques. In the 3rd model, four different methods for selecting matched non-adopters, namely stratification matching, nearest neighbour matching, radius matching and Kernel matching were used by the study. The marketed surplus was overwhelmingly explained by adoption of improved varieties as indicated by the positive and significant coefficient of adoption variable in the three econometric models employed by their research pointing to the robustness of the results. Ceteris paribus, adoption of improved technologies results in an increase in marketed surpluses by about 19% in the treatment effect model. In the case of the regression based on propensity score (model 2), two alternative specifications are estimated by them. First only the propensity score and the adoption variables are included in the equation and in the second part other control variables in addition to the propensity score are included. Both estimation results showed a positive and strong effect of adoption on marketed surplus. Using propensity score matching techniques (model 3) on only 448 observations, the estimated results based on the four matching algorithms showed that their ATT estimate is robust. The overall average gain in the percentage of total chickpea production sold ranges from 0.16 to 0.20 and the estimated gain was statistically significant at 1% for all the matching methods. This indicates that (assuming there is no selection bias due to unobservable farm and farmer specific factors) market integration level of smallholder farmers who adopted improved chickpea varieties is significantly higher than the non adopters. Generally, their results underscored that a household’s production technology choices fundamentally affect its level of market integration primarily by affecting its productivity.
Hailu et al. (2014) examined the impact of agricultural technology adoption on farm income by collecting data through semi-structured questionnaire administered on 270 randomly selected smallholder farmers from Raya-Azebo and Raya-Alamata districts of Tigray region, Ethiopia in 2013 cropping year. In doing so, they used a multi stage stratified random sampling method by which districts that are conducive for agriculture were first selected purposively by the overall researchers’ affiliation to the study area. Secondly, of the total 29 sub-districts, four sub-districts were selected randomly. Thirdly, eleven villages were selected proportionally where sub-districts with larger number of villages were given more weight. Fourthly, villages’ sample size was determined proportionally from the already defined sample size from which adopters and non-adopters were identified; and finally, final respondents were selected randomly from the list of the farm households from each targeted villages. By employing Ordinary Least Square (OLS) regression model, they found that chemical fertilizer adopters were much better to get birr 6672.022 than their non-adopter counterparts. Furthermore, HYV (High-Yielding Variety) adopters were found to be earners of birr 4717.575 much better than their counter parts.
Ketema M. and Kassa B. (2016) examined the impact of technological innovations on wheat production and also decomposed the total change in wheat output resulting from the introduction of new technologies into its constituent parts. For this purpose, a three-stage sampling technique was used to select sample respondents. In the first stage, out of seven wheat growing highland districts of Bale, two districts (Sinana-Dinsho and Gassera) with relatively higher number of improved wheat technology users were selected purposively. In the second stage, based on the proportion of the number of peasant associations in the selected districts, a total of 12 peasant associations were randomly selected. In the final stage, a total of 122 farm households (60 from old variety growers and 62 from new variety growers) were selected randomly using probability proportional to size technique which constituted a total of 114 new variety wheat plots and 84 old variety plots of the meher season and these plots are considered for econometric analyses. In order to investigate whether or not improved wheat technologies result in shift of the production function, output elasticities were estimated by Ordinary Least Squares (OLS) method by fitting Cobb-Douglas production function which was estimated separately for old variety plots, new variety plots, and for the pooled data in its log-linear regression.The output decomposition model developed by Bisaliah (1977) was also used in their study to decompose the total change in output into its constituent parts. The econometric results of the study indicated that, out of 55% of the observed productivity difference between old and new variety grown plots, technological change and change in associated input levels contributed about 24% and 31%, respectively and of the 31% increment attributed to input use levels, an increased use of herbicides and fertilizers caused the biggest jump in the productivity of improved wheat varieties (15.5% and 11% respectively).
D. Zeng et al. (2015), however, used data that came from a household survey conducted jointly by CIMMYT (International Maize and Wheat Improvement Center) and EIAR (Ethiopian Institute of Agricultural Research) in 2010 covering the four regions-Oromia, Amhara, Tigray, and Southern Nations, Nationalities, and People’s Region (SNNPR)-which together account for more than 93% of maize production in Ethiopia as to Schneider and Anderson (2010). The data was collected using a stratified random sampling strategy where strata are randomly selected “woredas” (districts) of high, medium and low maize yield potential. As a result, the data are nationally representative with regional differences in maize productivity accounted for and a total of 1,396 farm households from 30 woredas were surveyed; of these, 1,359 grow maize on 2,496 plots. Moreover, among the 1,359 households, 503 were adopters, 583 were non-adopters, and 273 were partial adopters of maize improved varieties. The researchers estimated a Cobb-Douglas production function via 2SLS, Probit-2SLS and GMM procedures to reveal yield ATT (average treatment effect on the treated). They also obtained alternative estimates under heterogeneity by the overall ATT evaluated using plot-specific marginal treatment effects (MTEs) and the results from the different estimation procedures are numerically close. Depending on the model, yield ATT is estimated to be between 47.6% and 63.3%. A flexible translog functional form is also estimated, and yield ATT is estimated as 53.5–61.6% whose closeness builds confidence in the estimates. Estimated yield MTEs were found to be highest among mid-low propensity scores which indicates negative selection: farmers are less likely to grow improved varieties on plots that are more likely to observe a higher yield gain. The researchers also implemented PSM (Propensity Score Matching) as a means of robustness check for the estimates of ATTs. Here, they first estimated a plot-level probit model to obtain propensity scores. After performing balancing tests which suggested no systematic differences in the distribution of covariates between treated and untreated groups, three matching techniques: nearest neighbor matching, radius matching, and kernel matching were employed by them. Accordingly, the yield ATT is estimated to be 43.4–48.9% (all with 1% significance) which was numerically close to the previous econometric estimates. Finally, they estimated the yield ATT using the subsample of 273 partial adopters with 772 plot-level observations as the difference in productivity between improved and traditional plots of the same farm household, and OLS regressions using Cobb-Douglas and translog specifications suggest 38.7% and 42.1% yield increases, respectively, both significant at 5% which again lend credence to their previous estimates.
Haregewoin T. et al. (2018), similarly, used data acquired from farm household survey undertaken during 2015/16 by Ethiopian Institute of Agricultural Research (EIAR) in collaboration with the International Maize and Wheat Improvement Center(CIMMYT). In collecting the data, a total of 837 farm households (of which 734 were adopters and 103 were non-adopters of improved wheat varieties) living in major wheat producing areas of 11administrative zones (provinces), 27 districts and 65 “kebeles”/villages/local councils in Oromia Regional State were interviewed and a multi-stage stratified sampling procedure was used to select villages from each agroecology, and households from each “kebele”/village. In doing so, first, agro-ecological zones that account for at least 3% of the national wheat area were selected from the major wheat growing areas of Oromia Regional State, Ethiopia and second, based on proportionate random sampling, up to 21 villages in each agro-ecology, and 15–18 farm households in each village were randomly selected. The researchers used PSM (Propensity Score Matching) in order to identify the attributable impact to the improved wheat varieties adoption. In so doing, they first estimated a plot-level probit model to obtain propensity scores. After performing balancing tests which suggested no systematic differences in the distribution of covariates between treated and untreated groups, all the four matching techniques: nearest neighbor matching, stratification matching, caliper/radius matching, and kernel matching were employed by them. Accordingly, their result revealed that on the average adopters of improved wheat varieties get a significantly higher productivity, ranging from 34-38% (all with 1% significance), than their counterparts, the non-adopters.
Daniel F. (2018) also used data acquired from farm household survey undertaken during 2015/16 by Ethiopian Institute of Agricultural Research (EIAR) in collaboration with the International Maize and Wheat Improvement Center(CIMMYT) covering seven major wheat growing agro-ecological zones that accounted for over 85% of the national wheat area and production distributed in the four major administrative regions of Ethiopia-Oromia, Tigray, Amhara as well as South Nations, Nationalities and Peoples (SNNP). In collecting the data, a total of 1,611 (837 from Oromia, 85 from Tigray, 509 from Amhara and 180 from SNNP) farm households (of which 1349 were adopters and 262 were non-adopters of improved wheat varieties) living in major wheat producing areas of 27 administrative zones (provinces), 61 districts and 123 “kebeles”/villages/local councils in the four major administrative regions of the country were interviewed and a multi-stage stratified sampling procedure was used to select villages from each agroecology, and households from each “kebele”/village. In doing so, first, agro-ecological zones that account for at least 3% of the national wheat area were selected from the major wheat growing regional states of the country mentioned above and second, based on proportionate random sampling, up to 21 villages in each agro-ecology, and 15–18 farm households in each village were randomly selected. The researcher used PSM (Propensity Score Matching) to identify the attributable impact to the improved wheat varieties adoption at national, administrative regional as well as agroecologic zonal levels . In doing so, he first estimated a plot-level probit model to obtain propensity scores. After performing balancing tests that suggested no systematic differences in the distribution of covariates between treated and untreated groups, all the four matching techniques: nearest neighbor matching, stratification matching, caliper/radius matching, and kernel matching were employed by him. Accordingly, his result indicated that on the average adopters of improved wheat varieties get a significantly higher rate of growth in productivity, ranging from 18-22% (all with 1% significance), than their counterparts, the non-adopters at national level. On the other hand, his result indicated that adoption of improved wheat varieties doesn't have homogeneous positive and significant impact on productivity growth in all of the administrative regions and agroecologic zones considered. Accordingly, improved wheat varieties adoption has a positive and significant impact on productivity growth in only two of the four regions considered, namely Oromia (ranging from 28-43%) and Tigray (ranging from 23-24%) while it has a negative and significant impact on productivity growth in SNNP (ranging from -71 to -41%). Improved wheat varieties adoption also has a positive and significant impact on productivity growth in only three of the nine agroecological zones considered, namely Tepid sub-moist mid highlands (ranging from 31-52%), Tepid humid mid highlands (ranging from 30-50%) and Cool moist mid highlands (ranging from 25-45%).
Abate G.T. et al. (2016) examined the impact of the Wheat Initiative technology package promoted by the research and extension systems in Ethiopia on wheat growers in the highlands of the country where the package includes improved wheat seed, a lower seeding density, row planting, fertilizer recommendations, and marketing assistance. For this purpose, they used 36 experimental “kebeles” spanning on 18 “woredas” in the Oromia, Amhara, and Tigray regions of Ethiopia and they compared three groups of farmers-full package, marketing only, and control. In the Benchmark or full package group, farmers benefitted from the full promotional ATA (Ethiopia’s Agricultural Transformation Agency) wheat package (inputs, extension and awareness of the Ethiopian Grain Trade Enterprise market opportunities). In the Market group, farmers did not benefit from extension and input support, but were made aware of the of the guaranteed market opportunity offered by Ethiopian Grain Trade Enterprise. Farmers in the Control group did not receive extension or input support, nor were they made aware of the Ethiopian Grain Trade Enterprise market opportunity. In their study, it is assumed that market and control farmers plant wheat following the existing or traditional production practices, although they were not precluded from adopting parts of the package at their own costs. Their sample design followed a three-stage approach. In the first stage, 18 “woredas” that were able to send a list of 14 farmers by “kebele” were selected for the evaluation; each of these “woredas” constitutes between 4 and 10 “kebeles”. In the second stage, 2 “kebeles” per “woreda” were randomly selected. In the third stage, the 14 farmers were randomized into benchmark farmers, market farmers, and control farmers (504 wheat farmers in total) in order to create three otherwise similar groups, stratified by model, non-model, and female farmers in order to ensure that the mix of farmers targeted by the program was preserved within each “kebele”. The researchers estimated the effect of the full wheat package and the market-only aspect of the package based on different equations that included different sets of dependent variables as well as interaction terms. Finally, they found that the full package intervention had a 14 percent increase on wheat yields, measured with both crop-cuts and farmer predicted yields, once they control for the farmer type (like model farmers or female farmers) and some household and plot characteristics that are unlikely to change over time including the age of the household head, its education level, the landholding size, the household size, the soil quality, and the distance to plot from home. They also noted that the measured yield difference may underestimate the true yield difference associated with the technology because of incomplete adoption of the recommended practices by intervention farmers and adoption of some practices by control farmers.
In a nutshell, as most of past studies undertaken to assess impact in the Ethiopian context made use of primary data collected by the researchers themselves, the studies are not expected to face limitation with respect to data availability which can be considered as one major strength of those studies. On the other hand, almost all studies employed propensity scores matching (PSM) technique as one preferred method of evaluating the impact of an intervention (such as adoption of some improved technology) on outcome of interest assuming there is no selection bias due to unobservable factors. In so doing, most studies used most/all of the available matching algorithms-nearest neighbor matching, stratification matching, radius matching, and kernel-based matching and they also necessarily undertook matching quality test. Moreover, the use of alternative methods of impact evaluation (like the endogenous switching regression model) by some of the studies also yields consistent results with that of the PSM method. Thus, this study can draw important lessons with respect to these methodological aspects. However, almost all such studies in the past were undertaken by considering one/a few specific location(s) or district(s) only. Besides, most studies were biased towards those locations that had high/better suitability and/or preference for the production of the specific crop considered. Thus, a nationally or regionally representative data could not be collected for the studies and the conclusions drawn so far would have low probability of influencing national and regional policies. Moreover, the focus of most studies was measuring the impact of a single improved agricultural technology or information rather than of a package of agricultural technologies and information.
3. MATERIALS AND METHODS
3.1. Analytical Framework for Evaluation of Adoption of Wheat Variety Impact on Productivity
The correct evaluation of the impact of a treatment like adoption of a technology will require identifying the “average treatment effect on the treated” defined as the difference in the outcome variables between the treated objects like farmers and their counterfactual. A counterfactual is defined as “knowledge of what would have happened to those same people if they simultaneously had not received treatment” (Olmos A., 2015 citing Shadish et al., 2002). In this context, as to González et al. 2009, if Y represents the outcome variable and if D is a dummy variable that takes the value of 1 if the individual was treated and 0 otherwise, the “average treatment effect on the treated” will be given by:
Abbildung in dieser Leseprobe nicht enthalten
However, accordingly, given that the counterfactual (E[Y (0) / D = 1]) is not observed, a proper substitute has to be chosen to estimate TATT. Using the mean outcome of non-beneficiaries-which is more likely observed in most of the cases-do not solve the problem given that there is a possibility that the variables that determine the treatment decision also affect the outcome variables. In this case, the outcome of treated and non-treated individuals might differ leading to selection bias (González et al., 2009). To clarify this idea, the mean outcome of untreated individuals has to be added to (1) from which the following expression can be easily derived:
Abbildung in dieser Leseprobe nicht enthalten
Here E [ Y (0) / D = 1]− E [ Y (0) / D = 0] represents the selection bias which will be equal to zero if treatment was given randomly which can be achieved through the use of experimental approach.
The experimental approach, according to Olmos A. 2015, has two characteristics: (1) it manipulates the independent variable, that is, whether an individual receives (or not) the intervention under scrutiny and (2) individuals are randomly assigned to the independent variable. The first characteristic does not define the experimental approach: most of the so-called quasi-experiments also manipulate the independent variable. What defines the experimental method is the use of random assignment (Olmos A., 2015). However, due to ethical or logistical reasons, random assignment is not possible as to Olmos A. 2015 citing Bonell et al. 2009. Moreover, accordingly, equivalent groups are not achieved despite the use of random assignment which is known as randomization failure. Usual reasons why randomization can fail are associated with missing data which happened in a systematic way and sometimes can go undetected (Olmos A., 2015).
As a consequence of randomization failure, or because of ethical or logistical reasons, in a very large number of real-world interventions, experimental approaches are impossible or very difficult to implement. However, if we are still interested in demonstrating the causal link between our intervention and the observed change, our options become limited. Some options include regression discontinuity designs which can strengthen our confidence about causality by selecting individuals to either the control or treatment condition based on a cutoff score. Another alternative is propensity scores matching technique. Propensity scores matching is a statistical technique that has proven useful to evaluate treatment effects when using quasi-experimental or observational data (Olmos A., 2015 citing Austin, 2011 and Rubin, 1983). Some of the benefits associated with this technique, accordingly, are: (a) Creating adequate counterfactuals when random assignment is infeasible or unethical, or when we are interested in assessing treatment effects from survey, census administrative, or other types of data, where we cannot assign individuals to treatment conditions. (b) The development and use of propensity scores reduces the number of covariates needed to control for external variables (thus reducing its dimensionality) and increasing the chances of a match for every individual in the treatment group. (c) The development of a propensity score is associated with the selection model, not with the outcomes model, therefore the adjustments are independent of the outcome. According to Olmos A. 2015, propensity scores are defined as the conditional probability of assigning a unit to a particular treatment condition (i.e., likelihood of receiving treatment), given a set of observed covariates:
Abbildung in dieser Leseprobe nicht enthalten
where z = treatment, i = treatment condition, and X = covariates. In a two-group (treatment, control) experiment with random assignment, the probability of each individual in the sample to be assigned to the treatment condition is: (z = i │X)=0.5. In a quasi-experiment, the probability (z = i │X) is unknown, but it can be estimated from the data using a logistic regression model, where treatment assignment is regressed on the set of observed covariates (the so-called selection model). The propensity score then allows matching of individuals in the control and treatment conditions with the same likelihood of receiving treatment. Thus, a pair of participants (one in the treatment, one in the control group) sharing a similar propensity score are seen as equal, even though they may differ on the specific values of the covariates (Olmos A. 2015 citing Holmes 2014).
3.2. Data and Variables
The data utilized for this study is acquired from farm household survey undertaken during 2015/16 by Ethiopian Institute of Agricultural Research (EIAR) in collaboration with the International Maize and Wheat Improvement Center (CIMMYT). The sampling frame covered seven major wheat growing agro-ecological zones that accounted for over 85% of the national wheat area and production distributed in the four major administrative regions of Ethiopia- Amhara, Oromia, Tigray as well as South Nations, Nationalities and Peoples (SNNP). A multi-stage stratified sampling procedure was used to select villages from each agro-ecology, and households from each “kebele”/village. First, agro-ecological zones that account for at least 3% of the national wheat area each were selected from all the major wheat growing regional states of the country mentioned above. Second, based on proportionate random sampling, up to 21 villages in each agro-ecology, and 15 to 18 farm households in each village were randomly selected. The data was collected using a pre-tested interview schedule by trained and experienced enumerators who speak the local language and have good knowledge of the farming systems. Moreover, the data collection process was supervised by experienced researchers to ensure the quality of the data.
Productivity stands for the productivity of wheat per unit of land cropped measured in kilogram per hectare.
LnProductivity stands for the natural logarithmic transformation of Productivity.
HHAGE stands for the age of a household head.
HHSEX is a dummy variable indicating the sex of a household head where HHSEX = 1 if the head is male and 0 if otherwise.
FAMILY_SIZE stands for size of a household.
HHEDU is a dummy variable indicating whether a household head is literate where HHEDU = 1 if the head is literate/able to read and write/ and 0 if otherwise.
CREDIT is a dummy variable indicating household's access to credit where CREDIT = 1 if the household has got the credit it needed in 2013 and 0 if otherwise.
LANDHOLDING_SIZE stands for size of the land holding of a household measured in hectare.
DSTMNMKT stands for distance to the nearest main market from residence measured in kilometer.
OXEN stands for the total number of oxen owned by a household.
TNOTRAREDS stands for the total number of traders known by a household who could buy the produced grain.
EXCONTACT is a dummy variable indicating whether a household had contact with government extension workers where EXCONTACT = 1 if the household had got contact with government extension workers and 0 if otherwise.