Abstract: Background and Objective: Tef is the most important cereal crop and the main staple food for more than 70 M people. The average productivity of the crop is very low compared to other cereal crops and the spatial variability of the productivity of the crop is very high across different districts and peasant associations within the region and different regions of the country. Therefore, this research was intended to assess the impacts of the adoption of improved and high-yielding tef varieties on the improvement of household income in the Dendi District. Materials and Methods: This study used 210 sample households from five peasant associations in the Dendi District. Descriptive and econometric data analyses were done. The propensity score matching method and logistic regression model were used for econometric data analysis. Results: The result revealed that household heads who are using improved and high yielding tef technologies on average get more income of 7943 birrs compared to household heads that are non-users of tef technologies. Conclusion: Based on the result of this research, improving the awareness of tef farmers towards adoption of high yielding improved tef technologies will contribute more to improving the household income and their livelihood specifically and also contribute to improving national income generally.
INTRODUCTION
Tef is the most important indigenous cereal crop in Ethiopia, which is the leading crop in terms of the area of production, is about 3 million hectares and the second in total production next to Maize CSA cited in Nigus et al.1. Tef serves as the main staple food for more than 70 million people and its straw is highly utilized as livestock feed2,3. It is also getting popularity across the globe as it is a gluten-free and healthy food. Tef is grown for food and animal feed in different countries like Australia, the United States, Israel, the Netherlands, Eritrea, India, Pakistan, South Africa, Uganda, Mozambique and Kenya4,5.
Tef is a nutritious crop containing essential nutrients like protein, carbohydrates, ber, fat and minerals6. It is also rich in some minerals like iron which is significantly higher than the amount that we can get from bread wheat7. But, the average productivity of tef is lower compared to other cereals as different factors are contributing to this low productivity. Tef production in Ethiopia is facing immense production constraints that affect the yield potential of the crop, including lodging, low inputs, inappropriate sowing method, post-harvest losses and using low yielding local varieties8,9.
The agricultural sector of Ethiopia in general is mainly characterized by small-scale and subsistence, which is inevitably affected by different factors like extreme weather, poor agronomic practices, lower rates of inputs and low qualities of inputs. To overcome these problems and improve production and productivity, the Ethiopian government is recently implementing cluster farming as a new farming approach in which the leading and the coordination role was given to agricultural research centres, regional state, governmental and non-governmental organizations. Technical and material support like provision of pieces of training, seed, fertilizer and types of machinery, certification of product quality and facilitation of market linkage. This approach is expected to accelerate technology dissemination, enhance information about production and marketing and also enhance the efficiency of farm households through the diffusion of best practices across individual members.
Tef (Eragrostis tef ) is the most important cereal crop serving as the staple food for the majority of Ethiopian people that contributing more to improving food and nutritional security and serving as a source of income for small-holder farmers to cover their expenses. Its cultivation area is expanding from time to time over many years and continued to date10. But, the productivity of tef is lower compared to other cereals and its spatial variability across the regions of the country and different zones within the regions is high for different reasons. Among the contributing factors, using low yielding local varieties, drought stress, lodging effects, shattering and poor agronomic practices are reported to be the most significant factors11. The existence of such factors significantly affects farmers’ efforts to improve production, productivity, income and food security.
According to CSA cited in Nigus et al.1 tef productivity is showing wider spatial variabilities across different regions of the country. For example, during the 2020/21 main production season, the regional average of tef productivity varied from 19.31 quintals per hectare in the Oromia Region to 15.19 quintals per hectare in the Benishangul Gumuz Region. Similarly, there are also spatial variabilities across different zones, in which it varied from 15.84 (in the Southern) to 17.67 quintals per hectares (in the Northern) zones of the Tigray Region, from 10.09 (in Wag Hemra) to 22.65 quintals per hectare (in East Gojjam) zone of Amhara Region, from 15.50 (in East Bale) to 21.26 quintals per hectare (in South-West Shewa) zones of Oromia Region, from 10.53 (in Mao-Komo) to 15.44 quintals per hectare (in Metekel) zones of Benishangul Gumuz Region and from 12.74 (in Segen people) to the 18.10 quintals per hectare (in Sidama) zone of Southern nations and nationalities region.
Based on the information listed above, adoption of the new tef technology may be one of the sources of variation and similarly, there may be income differences between tef technology adopters and no-adopters. Therefore, this research was intended to assess the impact of tef technology adoption on the household income in the study area.
MATERIALS AND METHODS
Description of the study area: The study was conducted in Dendi District, West Shewa Zone, Oromia Regional State, Ethiopia during 2017/18. The District lies between 38°10'54'' East and 9°01'16'' North and is also found at 80 km to the West of the Capital City, Addis Ababa. The District is bordered by Dawo and Wenchi Districts on the South, by Ambo and Ilfata on the West, by Jaldu on the North and by Ejersa Lafo on the East. The District has 79,936.29 hectares of land (39,227.5 cultivated, 14,912.36 grazing, 7,925.93 forest and 14,829.5 uncultivated and 3,041 homestead and others), 38 kebeles (35 rural and 3 urban), 200715 population (42953 urban and 157762 rural), 19231 households (85.6% male and 14.4% female). The mean annual rainfall of the district is 1094 mm (ranging from 750-1170 mm), mean temperature of 16.30°C (ranging from 9.30 to 23.80°C) and mean altitude of 2200 m.a.s.l. Tef, wheat, barley, maize and sorghum are major cereals crops grown in the District12.
Sampling technique and sample size determination: Multi-stage sampling technique was employed. Dendi district was purposively selected based on its tef production potential. Of 38 kebeles in the District, 24 tef producing kebeles were identified. Then, five kebeles were randomly selected from a total of 24 tef producing kebeles. Finally, a total of 210 sample households were randomly selected. The sample size of 210 was determined using Yamane cited in Adam13, which can be expressed as follows:
| (1) |
Where:
| n | = | Sample size |
| N | = | Population size |
| e | = | Level of precision |
The sample households from each kebeles were selected using the proportional sampling method of:
| (2) |
| ni | = | Sample from the ith kebele |
| Ni | = | Total population in the ith kebele |
| ∑Ni | = | Population of the five sample kebeles |
| n | = | Total sample from the district |
Data type, source and method of analysis: Both primary and secondary data were used in this research. The primary data was collected from sample households using structured and semi-structured questionnaires. The collected data were analyzed using descriptive, inferential and econometric data analysis methods. Mean, percentage, minimum and maximum were used to report descriptive data analysis. Similarly, the t-test and chi-square test were used to infer the mean difference of descriptive data analysis. For econometric data analysis, the propensity score matching method and logistic regression model were employed.
In a quasi-experiment, the independent variables are manageable, but the problem is selection bias as program participants were not randomly selected to the conditions. In such cases, the propensity score matching method can be used to reduce selection bias. The propensity score is the probability of taking treatment and it controls the selection bias of the included independent variables, while not controlling the variables not measured (unobserved bias). It represents the included variables by a single value, which is a propensity score value.
Statistical analysis: Propensity scores can be estimated using different statistical methods. Logistic regression is the most frequently used method. In this method, estimation of the propensity score is done using logistic regression. The logistic regression model can be expressed as follows14:
| (3) |
Where:
| Pi | = | Probability of adopting the technology for the ith participant |
| βi | = | Are the model parameters |
| Xi | = | Vector of the explanatory variable |
The equation of the probability of adoption can be simplified as:
| (4) |
Using the probability of adoption, the probability of non-adoption can be derived as:
| (5) |
The odds ratio, the ratio of the probability of adoption to non-adoption, can also be derived as:
| (6) |
Taking the natural logarithm of the odds ratio, it can be further simplified as:
| (7) |
Where:
| Li | = | Natural logarithm of the odds ratio in favor of adopting tef technology |
| β's | = | Are the parameters to be estimated |
| Xi’s | = | Are the vectors of explanatory variables |
| ɛ | = | Is the error term |
RESULTS AND DISCUSSION
Descriptive results: Access to credit services showed positive and significant relation with households’ decision to adopt tef technologies. Table 1 showed that 84% of households having access to credit services are adopters of tef technologies while only 38.5% of households without access to credit were adopters of tef technologies and the chi-square test also revealed that the mean difference was significant at 1%. This result is similar to the one reported by Habtewold15.
Access to off-farm income also showed a positive relationship with the household decision to adopt tef technologies. According to the result in Table 1, 72% of households having access to off-farm income were adopters of tef technologies, while 60.7 of the households without access to off-farm income were adopters of tef technologies. From the chi-square test, the mean difference is statistically significant at the 10% probability level. This is in line with the finding reported by Milkias and Muleta16.
According to the result, households having access to extension services are more adopters of tef technologies. As displayed in Table 1, the majority of the sample households 156 (74%) have access to extension services out of which 112 (72%) were adopters of tef technologies. The chi-square test also showed that the mean difference of extension was significantly different at 1% for the adoption of tef technologies. This result is similar to the one reported by Wossen et al.17.
Based on the result in Table 1, 113 (54%) of the sampled households were members of cooperatives, while 97 (46%) were not. This shows that households that are members of cooperatives are more likely to adopt tef technologies. The result of the chi-square test also showed that there is a significant mean difference in the adoption of tef technologies between cooperative members and non-members. This result is similar to the one reported by Wossen et al.17.
According to the result in Table 2, the mean age of non-adopters is 46.5 years, while that of adopters is 41.4.
| Table 1: | Descriptive results for dummy variables by the adoption of tef | |||||
|
Adoption of tef technologies |
||||
| Variables |
No |
Yes |
Total |
Χ2 |
| Sex of the household head | ||||
| Female |
17 |
21 |
38 |
2.3 |
| Male |
55 |
117 |
172 |
|
| Access to credit services | ||||
| No |
66 |
105 |
171 |
7.6*** |
| Yes |
6 |
33 |
39 |
|
| Access to off-farm income | ||||
| No |
42 |
62 |
104 |
3.4* |
| Yes |
30 |
76 |
106 |
|
| Access to extension contact | ||||
| No |
28 |
26 |
54 |
9.9*** |
| Yes |
44 |
112 |
156 |
|
| Cooperative membership | ||||
| No |
45 |
52 |
97 |
11.7*** |
| Yes |
27 |
86 |
113 |
|
| Access to information | ||||
| No |
20 |
34 |
54 |
0.2 |
| Yes |
52 |
104 |
156 |
|
| *,***Significant levels at 10 and 1%, respectively and Source: Own computation from survey data | ||||
| Table 2: | Descriptive results of continuous variables by the adoption of tef | ||||||||||
|
Adopters |
Non-adopters |
Combined |
|||||
| Variables |
Mean |
Std. dev |
Mean |
Std. dev |
Mean |
Std. dev |
t-test |
| Age of the household head |
41.4 |
11.3 |
46.5 |
11.1 |
43.2 |
11.4 |
-3.1*** |
| Education status of the head |
4.9 |
3.7 |
2.8 |
2.5 |
4.2 |
3.5 |
4.4*** |
| Tef farming experience |
20.3 |
10.4 |
22.3 |
10.3 |
20.9 |
10.4 |
1.4 |
| Livestock holding (TLU) |
5.9 |
2.7 |
4.1 |
2.1 |
5.4 |
2.6 |
5.1*** |
| Family size (ME) |
2.9 |
1.3 |
2.8 |
1.1 |
2.8 |
1.2 |
-0.9 |
| Land owned (hectares) |
1.3 |
0.6 |
0.9 |
0.5 |
1.2 |
0.6 |
3.7*** |
| Market distance (KM) |
1.1 |
0.5 |
1.4 |
0.6 |
1.2 |
0.5 |
-3.3*** |
| Distance from FTC (min) |
17.7 |
16.6 |
21.5 |
16.9 |
19.0 |
16.7 |
-1.6* |
| *,***Significance levels at 10 and 1%, respectively and Source: Own computation from survey data | |||||||
| Table 3: | Estimation of the propensity score for impact assessment | |||
| Adoptions |
Coefficients |
Standard error |
t-value |
p-value |
| Sex of the head |
0.217 |
0.477 |
0.46 |
0.649 |
| Age of the head |
-0.126 |
0.041 |
-3.04 |
0.002 |
| Education of head |
0.058 |
0.067 |
0.88 |
0.380 |
| Family size (ME) |
-0.061 |
0.134 |
-0.46 |
0.648 |
| Farm experience |
0.106 |
0.046 |
2.32 |
0.021 |
| Livestock holding (TLU) |
0.159 |
0.102 |
1.55 |
0.120 |
| Land owned |
0.372 |
0.393 |
0.95 |
0.343 |
| Access to information |
0.044 |
0.393 |
0.11 |
0.911 |
| Coop membership |
0.031 |
0.420 |
0.07 |
0.941 |
| Market distance |
-0.412 |
0.363 |
-1.14 |
0.256 |
| Distance from FTC |
-0.006 |
0.012 |
-0.49 |
0.625 |
| Off-farm income |
0.923 |
0.407 |
2.27 |
0.023 |
| Extension contacts |
0.840 |
0.421 |
2.00 |
0.046 |
| Credit access |
0.504 |
0.530 |
0.95 |
0.342 |
| Constant |
1.888 |
1.302 |
1.45 |
0.147 |
| Mean dependent variables |
0.656 |
SD dependent variable |
0.476 |
|
| Pseudo r-squared |
0.223 |
Number of observations |
209.000 |
|
| Chi-square |
41.529 |
Prob>chi2 |
0.000 |
|
| Source: Own computation from survey data | ||||
This shows that the age of the household head is negatively related to the household’s decision to adopt tef technologies and the mean difference is statistically significant at a 1% significance level. This result is similar to the one reported by Shita et al.18.
The educational status of the household head is positively and significantly related to the adoption of tef technologies. From the result in Table 2, the mean educational level for adopters was 4.9 schooling years, while that of non-adopters was 2.8 schooling years. The t-test also showed that the mean difference was statistically significant at 1% probability. This result is similar to the findings reported by Jaleta et al.19.
Livestock holding also positively and significantly affected households’ decision to adopt tef technologies. According to this result, the mean livestock holding for adopters and non-adopters were 5.9 and 4.1, respectively. This is to mean that households having a greater number of livestock tend to adopt tef technologies compared to those households having a smaller number of livestock. The t-test result showed that the difference is statistically significant at 1% probability. This result is in line with the finding reported by Ayenew et al.20.
Land ownership is positively related to sample households’ tef technology adoption decisions, to means that household heads having larger land sizes are more likely to adopt tef technologies compared to those household heads owning less land. From the result in Table 2, the tef technology adopters own 1.3 ha of land while those non-adopters own 0.9 ha. The t-test result also revealed that the mean difference was statistically significant at 1%. This is contrary to the finding reported by Ayenew et al.20.
Distance from the main market and distance from farmers’ training centre both showed negative and significant relation with households’ decision to adopt tef technologies. Table 2 showed that the mean distance from the market and farmers’ training centre was 1.1 km and 17.7 min, respectively for adopters, 0.5 km and 16.6 min for non-adopters. From this result, households living nearer to the main market and farmers’ training centre are more likely to adopt tef technologies. The t-test result also revealed that the mean difference was significantly different for both. This result is similar to the one reported by Zegeye21.
Econometric results: Before going for the econometric model, Variance inflation factor (VIF) and multicollinearity tests were done. The results of both tests confirmed there were no problems with multicollinearity. Therefore, an estimation of the propensity score was done and the result is presented in Table 3.
Age of the head, farming experience, access to off-farm income and access to extension contact were the variables significantly affected households’ decision to adopt tef technologies. Farming experience, access to off-farm income and access to extension contact positively and significantly affected households’ decision to adopt tef technology, while the age of the household head negatively and significantly households’ decision to adopt tef technology.
Using the estimated propensity score, the common support region was restricted. According to the result in Table 4, the common support region is the region between 0.0759814 and 0.9219624. Based on this result, 33 households, (3 non-adopters and 30 adopters) were excluded from the model as they were out of the common support.
The estimated propensity score can also be visualized as presented in Fig. 1. The propensity score values greater than 0.0179827 and less than 0.0759814 were the values excluded from the control group as it is not in the region of common support and the values of the propensity score that is greater than 0.9219624 and less than 0.9871869 were excluded from the treatment group as it is not in the region of common support.
| Fig. 1: | Propensity score for adopters and non-adopters on and off support region |
| Table 4: | Restriction of the common support region | ||||
| Variable |
Observation |
Mean |
Standard deviation |
Minimum |
Maximum |
| P score (0) |
72 |
0.482196 |
0.2337602 |
0.0179827 |
0.9219624 |
| P score (1) |
137 |
0.7465831 |
0.1990134 |
0.0759814 |
0.9871869 |
| Source: Own computation from survey data | |||||
| Table 5: | Choosing the matching algorithm |
| Matching methods |
Matched samples |
Balanced variables |
Pseudo R2 |
| Nearest neighbor matching | |||
| Nearest neighbor (1) |
178 |
10 |
0.060 |
| Nearest neighbor (2) |
178 |
14 |
0.038 |
| Nearest neighbor (3) |
178 |
14 |
0.008 |
| Nearest neighbor (4) |
178 |
14 |
0.011 |
| Caliper matching | |||
| Caliper (0.01) |
77 |
14 |
0.103 |
| Caliper (0.10) |
108 |
14 |
0.030 |
| Caliper (0.25) |
120 |
14 |
0.087 |
| Caliper (0.50) |
129 |
5 |
0.495 |
| Radius matching | |||
| Radius caliper (0.01) |
134 |
13 |
0.049 |
| Radius caliper (0.10) |
178 |
14 |
0.012 |
| Radius caliper (0.25) |
178 |
14 |
0.022 |
| Radius caliper (0.50) |
178 |
11 |
0.084 |
| Kernel matching | |||
| Kernel bandwidth (0.01) |
134 |
12 |
0.056 |
| Kernel bandwidth (0.10) |
178 |
14 |
0.013 |
| Kernel bandwidth (0.25) |
178 |
14 |
0.012 |
| Kernel bandwidth (0.50) |
178 |
12 |
0.050 |
| Source: Own computation from the survey result | |||
After the restriction of the common support region, choosing the best matching algorithm was done using nearest neighbor matching, caliper matching, radius matching and kernel matching as presented in Table 5.
From the four matching algorithms tested, the nearest neighbour matching of the third neighbour showed good matching properties (smallest value of pseudo-R-square (0.08), all covariates balanced (14) and the largest number of observations matched (178)) compared to other methods and chosen for this specific study.
A balancing test was also conducted to check the matching quality. After a matching is done, there should not be a significant difference between the group (treatment and control).
| Table 6: | Covariates balancing test (testing for matching quality) |
|
Mean |
T-test |
|||||||
| Variables | Unmatched/matched |
Treated |
Control |
Bias |
Reduct |bias| (%) |
t |
p>|t| |
V (T)/V (C) |
| Sex of the head | Unmatched |
0.84672 |
0.76389 |
20.9 |
1.48 |
0.141 |
. |
|
| Matched |
0.83333 |
0.85494 |
-5.5 |
73.9 |
-0.44 |
0.663 |
. |
|
| Age of the head | Unmatched |
41.212 |
46.528 |
-47.8 |
-3.28 |
0.001 |
1.02 |
|
| Matched |
42.444 |
42.855 |
-3.7 |
92.3 |
-0.29 |
0.771 |
1.42 |
|
| Education of the head | Unmatched |
4.9927 |
2.8333 |
67.7 |
4.39 |
0.000 |
2.29* |
|
| Matched |
4.1481 |
4.0216 |
4.0 |
94.1 |
0.32 |
0.750 |
2.24* |
|
| Family size ME | Unmatched |
2.9504 |
2.7889 |
13.8 |
0.92 |
0.357 |
1.38 |
|
| Matched |
2.9657 |
2.9926 |
-2.3 |
83.4 |
-0.17 |
0.864 |
1.51* |
|
| Farm experience | Unmatched |
20.153 |
22.333 |
-21.1 |
-1.45 |
0.149 |
1.01 |
|
| Matched |
20.88 |
21.111 |
-2.2 |
89.4 |
-0.17 |
0.866 |
1.10 |
|
| Livestock TLU | Unmatched |
5.989 |
4.1317 |
76.7 |
5.09 |
0.000 |
1.59* |
|
| Matched |
5.4617 |
5.3935 |
2.8 |
96.3 |
0.22 |
0.825 |
1.74* |
|
| Land owned | Unmatched |
1.2971 |
0.98958 |
55.8 |
3.73 |
0.000 |
1.45* |
|
| Matched |
1.2046 |
1.1684 |
6.6 |
88.2 |
0.51 |
0.609 |
1.35 |
|
| Access to info | Unmatched |
0.75182 |
0.72222 |
6.7 |
0.46 |
0.644 |
. |
|
| Matched |
0.74074 |
0.76543 |
-5.6 |
16.6 |
-0.42 |
0.676 |
. |
|
| Cooperative membership | Unmatched |
0.62044 |
0.375 |
50.4 |
3.46 |
0.001 |
. |
|
| Matched |
0.53704 |
0.5216 |
3.2 |
93.7 |
0.23 |
0.821 |
. |
|
| Market distance | Unmatched |
1.1153 |
1.3494 |
-45.5 |
-3.23 |
0.001 |
0.66* |
|
| Matched |
1.1817 |
1.1709 |
2.1 |
95.4 |
0.16 |
0.871 |
0.79 |
|
| Distance from FTC | Unmatched |
17.708 |
21.583 |
-23.1 |
-1.59 |
0.113 |
0.97 |
|
| Matched |
18.065 |
18.972 |
-5.4 |
76.6 |
-0.43 |
0.670 |
1.67* |
|
| Off-farm income | Unmatched |
0.55474 |
0.41667 |
27.7 |
1.90 |
0.058 |
. |
|
| Matched |
0.46296 |
0.54012 |
-15.5 |
44.1 |
-1.13 |
0.259 |
. |
|
| Extension contacts | Unmatched |
0.81022 |
0.61111 |
44.8 |
3.19 |
0.002 |
. |
|
| Matched |
0.76852 |
0.75309 |
3.5 |
92.2 |
0.26 |
0.792 |
. |
|
| Credit access | Unmatched |
0.24088 |
0.08333 |
43.6 |
2.82 |
0.005 |
. |
|
| Matched |
0.16667 |
0.17593 |
-2.6 |
94.1 |
-0.18 |
0.858 |
. |
|
| *If variance ratio outside (0.71, 1.40) for unmatched and (0.68, 1.46) for matched and Source: Own computation from survey data | ||||||||
| Table 7: | Joint significance test for covariate balancing |
| Samples | Pseudo R2 | LR chi2 | p>chi2 | Mean bias | Med bias | B | R | Variables (%) |
| Unmatched | 0.223 | 60.15 | 0.000 | 39.0 | 44.2 | 121.8* | 1.04 | 50 |
| Matched | 0.008 | 2.33 | 1.000 | 4.6 | 3.6 | 20.7 | 1.53 | 50 |
| *If B>25%, R outside (0.5, 2) and Source: Own computation from survey data | ||||||||
| Table 8: | Estimation of average treatment effect (ATT) |
| Variable | Samples | Treated | Controls | Difference | Standard error | T-stat |
| Income | Unmatched | 38742.38 | 25296.81 | 13445.57 | 2683.48 | 5.01 |
| Matched | 38880.46 | 30937.38 | 7943.100 | 4824.40 | 1.65 |
Table 6 confirmed that there is no significant difference between treatment and control after matching that making ease of comparison between participants and non-participants.
The joint significance test in Table 7 was our guarantee that we can estimate the average treatment effect on the treated since the pseudo R2 reduced from 0.223 to 0.008, the likelihood ratio reduced from 60.15 to 2.33 and the mean bias also reduced from 39.0 to 4.6.
Finally, the average treatment effect on the treated (ATT) was estimated. The result in Table 8 revealed that the average treatment effect was 7943 Birr. According to this result, households who were adopters of tef technology earn 7943 Birr more income on average compared to household heads that are not adopters of teff technology. The t-test result also showed that the mean income difference for adopters and non-adopters is statistically significant at a 5% probability level.
To check whether the estimated average treatment effect was the pure effect of the adoption of tef technologies, a sensitivity analysis was conducted. The result proved that the estimated treatment effect was the pure effect of tef technology adoption as it was insensitive to unobservable bias if the gamma value increased to 3.
CONCLUSION
Based on the result of this research, household heads who adopted high yielding improved tef technologies earned a higher income of 7943 Birr than non-adopter household heads on average. From the t-test result, the mean income difference between adopters and non-adopters was statistically significant. Therefore, the adoption of tef technologies significantly improves the household income in the study area.
SIGNIFICANCE STATEMENT
This study discovered that the adoption of high yielding improved tef varieties directly contributes to improving household income and indirectly contribute to reducing spatial variability of tef production and productivity that prevails within and across different regions of the country.