The impact of adopting artificial cattle insemination technology by smallholder farmers on their wellbeing: The case of Yem Special District, Ethiopia

Abstract This study investigated the impact of adopting artificial cattle insemination technology (ACI) by smallholder farmers on their well-being in the case of Yem special district. Using three-stage sampling techniques, cross-sectional data were collected from 361 household head that include adopter and non-adopter of ACI. The propensity score matching method was used to estimate the impact of adopting ACI on the welfare of households. The method was checked for covariate balancing with a standardized bias, t-ratio, and joint significance-level tests. Moreover, sensitivity analysis of the estimated effect of adoption to unobserved selection bias was checked using Rosenbaum bounds procedure. We employed STATA Version 13 software to run the estimation and diagnostic tests. The results revealed that ACI adoption practices have a positive impact on the wellbeing of households. The results indicate that the adoption of ACI technology on average increases income from milk, livestock, and the total consumption of households by 62.742, 31.215, and 11,325.694 Birrs per year, respectively, compared to non-adopters at a 1% significant level. The sensitivity analysis shows that the estimated impact was insensitive to unobserved selection bias.


Introduction
The economic and social importance of livestock is known both at the national and household levels. The sub-sector contributes 12% of Gross Domestic Product (GDP) and over 45% to agricultural GDP (Muuz, 2018). The Ethiopia livestock master plan brief underlined that livestock is ABOUT THE AUTHOR Endalkachew Kabtamu Mekonen is a Lecturer at Wolkite University, department of Economics. He attended his undergraduate study in Economics at Addis Ababa University, school of Economics. He also pursued his postgraduate study at School of Economics, Addis Ababa University with concentration in Economics Policy Analysis. In addition to teaching, he has been actively engages in research and community services activities. His research interests include poverty, urban economics, labor economics, financial economics, human resource economics, macroeconomic and microeconomic issues.
The wellbeing of livestock growing households depends on their income, production and employment levels, savings, and other household needs. In various studies, such as Diro Chelkeba et al. (2016), the impact of technology adoption on household well-being can be measured based on different aspects of well-being indicators that are associated with an improved standard of living, such as food, clothing, shelter, health care, education, and recreation. However, in this study income from milk, and animal sales, and total consumption/food and non-food/were used to measure the impact of artificial cattle insemination on the wellbeing of households. In Yem's special district, livestock is the pillar for the livelihood of the majority of the population by giving draft power, sources of food like meat, milk, yogurt, and source of cash income. Despite their benefit in the district, its productivity has remained low, and it is dominated by local breeds with low productivity. Several studies confirmed that adoption of artificial insemination boost livestock production that inspires transition from low production to high productivity that insures household wellbeing. For instance, Amanuel et al. (2018); Melesse and Jemal (2012), and L. Baker (2018) showed that adopting dairy technology like artificial insemination improve productivity and increase income earnings. Adoption of artificial insemination is considered as a key element in structural changes in the livestock industry as it directly affects the performance of dairy cows (Gillespie et al., 2004) and leads to income generation and improvement in the wellbeing of user households (Olynk & Wolf, 2008).
Artificial cattle insemination is implemented in the Yem special district since 2006 to scale up dairy production and generate higher income from the sector. However, no studies have been conducted to examine the impact of adopting artificial insemination technology on the well-being households in the district at all. According to Yem special district agriculture office (2021) report indicates that no effort had been made to evaluate the impact of the program; hence creating an information gap that needed to be filed. Several questions require rigorous assessment, which adheres to the impact of adoption of improved technology on the wellbeing of households. Thus, the study is intended to examine the impact of adopting ACI technology for cattle breeding on the wellbeing of smallholder farmers. Moreover, it is designed to bridge the existing knowledge and information gap related to the impact of ACI on household wellbeing. The successfully of the ACI program depend on the several factors, such as genetic, reproductive condition, environment, sperm separation between X and Y (Kusumawati et al., 2019;Susilawati et al., 2020). Recognizing this, adopting artificial insemination technology is broadly considered to address household wellbeing (Yohannes, 2014). Therefore, this study is aimed to examine the impact of adopting artificial cattle insemination technology on household wellbeing in Yem special district.
The output of the study would have theoretical and empirical contributions. It can show how the adoption of artificial insemination technology affects the well-being of cattle growers. In addition, the finding of the study is expected to benefit policymakers and planners as a source of information to formulate policies concerning farmers' technology adoption and related development issues. This study could also support the government and stakeholders in terms of providing critical insights and knowledge to design future development programs and projects supporting farmers' wellbeing. It can serve as literature for researchers who want to undertake further study on a similar topic. The study is organized into five sections. The first section presents the introduction, the second section focuses on reviews of the related literature, the third section introduces the methodology of the study, the fourth section presents result and discussion of the study and the last section presents conclusions, and recommendations.

Review of related literature
Various literatures indicated that improving dairy productivity plays a noteworthy role in the economic, social, and cultural status of households, and it improves the income and well-being of the farm family (Elisa et al., 2015). Livestock improves food supply, nutrition, income, savings, soil productivity, livelihoods, transport, agricultural diversification and sustainable agricultural production, family and community employment, ritual purposes, and social status (Moyo & Swanepoel, 2010). Artificial insemination is the most utilized technology widely applied to increase livestock production and productivity that generates household income and improve their wellbeing (Gordon, 2004). Adopting ACI plays a vital role to increase the yield of cows and it is the cheapest way of genetic improvement (Effa, 2011). The widespread application of ACI in countries, such as the USA, has resulted in a steady improvement in the genetic quality of dairy animals and a doubling of milk yields during the past years (Khanal, 2010). In countries, such as India, governments were able to support artificial insemination breeding programs with the semen of exotic breeds like Holstein-Friesian, Brown Swiss, and Jersey (Gordon, 2004). Dehinenet et al. (2014), it is confirmed that there is a difference between dairy technology adopters and non-adopter in terms of production and productivity that can bring a different well-being status between user and non-user households. Adopter smallholder farmers could get more milk production on average than non-adopter farmers and could have better household well-being. Various cross-country studies have confirmed the higher income and nutritional security effect of commercial dairy farming that results from the adoption of crossbreeding technology. For instant, studies carried out by Alary et al. (2011, Melesse andJemal (2013), Quddus (2012), and Udo and Steenstra (2010) have found a positive impact of artificial insemination breeding technology adoption on income and nutritional security of dairy farm households. Adoption of improved bred cattle can, thus, be a suitable option for improving farmers' wellbeing through enhancement of livestock-level income and higher milk consumption due to a rise in production leading to household's better nutrition. A study by Quddus (2012) confirmed that the adoption of dairy farming technologies by small farm holders, particularly ACI for cattle breeds, and the age of the farmer was interrelated with technology adoption.

As indicated in
A study conducted in Brazil about ACI revealed that the direct and indirect impacts of increasing ACI adoption and genetically superior replacement bull utilization shows that the value of these actions is remarkable. Any increase in the use of superior animals will cause very significant economic effects in the Brazilian beef industry, reaching values as high as US$ 342 million with only a 200% of increment, which, with the fast growth of AI, is going to be reached in near future. A study by Valergakis et al. (2007) indicates that daughters of AI sires were producing almost 900 kg of extra milk per lactation than daughters of natural service bulls. Another report from the USA showed a difference of more than 1000 kg of milk per lactation on farms using ACI (mith et al., 2013). Using AI for cattle breeding plays an important role in enhancing animal productivity, especially milk yields, in developing countries that have a well-defined breeding strategy and a sound technical base to absorb and adapt the technology to meet their needs and get better welfare (BBC Historic Figures, 2015). In a nutshell, from the above literature, it can be understood that most of the studies focused on either identifying the importance of using AI for improving livestock breed or discussing the way of adopting artificial insemination; but none of them have identified and evaluated the impact of adopting artificial insemination on the well-being of user households. Thus, this study has designed to examine the economic impact of adopting AI on the well-being of user households.
Literature gap: From the above literatures, it can be understood that most of the studies focused on either identifying the importance of using ACI for improving livestock breed or discussing the way of adopting artificial insemination; but none of them have identified and evaluated the impact of adopting artificial insemination on the well-being of user households. Adequate attention is ignored to assess its impact on the household well-being. So this study is designed to examine the economic impact of adopting ACI on the well-being of users' households. Therefore, it was intended to contribute and fill the knowledge gap regarding its impact on household well-being and adoption practices.

Research methodology
The study was conducted in the Yem special district, located in the north-western tip of the Southern Nations, Nationalities, and Peoples Regional State of Ethiopia. The administrative center of the Yem special district is Saja, 247 Kilometers far from Addis Ababa in the Southwest direction. Yem is bordered on the west and north by the Oromia region and separated from Gurage on the northeast and Hadiya on the east by the Gibe River. The total population of Woreda as per the 2007 population census is estimated to be 80,647 of which 50.3% are male and 49.7% female (CSA/Central Statistical Agency, 2010/2011). Different economic indicators are related to farming and non-farming activities in the districts. Livestock is the backbone of the livelihood of the majority of the population by giving draft power supply for crop production and transport, as a source of meat, milk, egg, and source of cash income. The administrative map of Yem special district is shown as Figure 1.
According to the Yem Special District Agriculture and Natural Resource Development Office (2019) report, there are 32 rural kebeles and 5 urban centers in the district. Among 32 kebeles, only 25 are adopting artificial insemination for the dairy cow. Three-stage sampling was employed in the study. First, 8 kebeles were selected purposively from 25 kebeles based on the number of adopters in the kebeles. These sample kebeles are Fofa, Saja Ketama, layigna kesheli, tachi kesheli, Nubba, Gessi, Tobba, and Deri tegu. Secondly, from eight kebeles, households were grouped into two clusters, adopters, and non-adopters of artificial insemination technology. Lastly, simple random sampling was applied to select respondents from the two strata. A representative sample for each stratum was selected proportionally to each category as presented in Table 1. The sample households were selected at ± 5% precision level and 95% confidence interval. Based on Kothari (2004), the sample size is determined as (equation 1).
Where, n is the desired sample size; Z is the standard normal curve, which is equal to 1.96 for 95%); e is the desired level of precision+5%, p is the estimated proportion of an attribute that is present in the population, and q is 1-p.
Both primary and secondary sources were used in the study. The primary data was collected from sampled households, ACI technicians, Woreda agricultural offices, development agents (DA), and others who have adequate information about ACI. A structured questionnaire was employed to collect primary data. The questionnaire includes information on respondents' demographic characteristics, socioeconomic and institutional, behavioral, situational, social factors, and communication factors related information. On the other hand, secondary data was collected from records of the Woreda agricultural offices and reports of government and non-governmental organizations.
The study employed both descriptive and inferential data analysis. The descriptive analysis involves the use of mean, standard deviation percentage, and frequency distribution. In addition, the Propensity Score Matching (PSM) model was used to examine the impact of ACI technology adoption on household wellbeing. As indicated in Caliendo and Kopeinig (2005), the implementation of PSM involves five steps. These are PSM estimation, selection of matching algorithm, checking for common support, estimating treatment effect and assessing matching quality, and finally sensitivity analysis. To attain the intended objectives, the PSM method was used to know the impact of adopting AI on different outcome variables. It is preferred to other non-experimental methods because it does not require baseline data, the treatment assignment is not random, and considered the second-best alternative to experimental design in minimizing selection biases (J. L. Baker, 2000). Estimating the effect of adopting artificial insemination on a given outcome (Y) is specified as (equation 2).
Where Ti is the treatment effect (effect of implementing AI), Yi is the outcome on household i, Di is whether a household i has got the treatment or not (i.e., whether a household use AI service for breeding cows or not). However, Y (Di = 1) and Y (Di = 0) cannot be observed for the same household at the same time. As a result, estimating individual treatment effects, T i is not possible but shifts to the average treatment effects of the population than an individual one. Two treatment effects are usually estimated in empirical studies. The first one is the household Average Treatment Effect (ATE), which is simply the difference between the expected outcomes by considering users and non-users specified in Equation (3).
Equation (3) answers the question of how much households benefit by participating in the program compared to what they would have experienced without participating in the program. The outcome of individual households from treatment and comparison groups would differ even in the absence of treatment leading to a self-selection bias. However, by rearranging and subtracting E[Y (0) D = 0] from both sides of equation 3, ATT can be specified as equation 4.
In equation (4)  i.e., when there is no self-selection bias. To solve the problem of self-selection, the PSM method has two underlying assumptions. These are conditional independence and common support assumptions. The conditional Independence Assumption (CIA) is given as equation (5).
Where ⊥ indicates independence, X is a set of observable characteristics, Y 0 is the control household, and Y 1 is the treated household. Given the set of observable covariates (X) which are not affected by treatment (AI), potential outcomes (household well-being) are independent of treatment assignment (independent of how households decide to use AI). That means the selection is based on observable characteristics (X), and variables that influence treatment assignment and household well-being are concurrently observed (Bryson et al., 2002;Caliendo & Kopeinig, 2008). After adjusting for observable differences, the mean of potential outcomes is the same for D = 1 and D = 0 and E (Y0│D = 1, X) = E (Y0│D = 0, X).
The common support is the region where the balancing score has positive density for both treatment and comparison units. This assumption rules out perfect predictability of D given that 0 < pr (D = 1│X) < 1. No matches can be formed to estimate the parameters when there is no overlap between the treatment and comparison groups. It also guarantees an individual with observable characteristics has a positive probability of belonging both to the participants and control group (P. Rosenbaum & Rubin, 1983). Given the above assumptions, the PSM estimator of ATT can be written as equation 6.
Where p(x) is the propensity score computed on the covariates X. Equation (6) is explained as PSM estimator of the mean difference in outcomes over common support, appropriately weighted by the propensity score distribution of participants. The propensity score is obtained using either logit or probit models. According to (Gujarati, 2004), both provide similar results. For simplicity logit model is used to estimate propensity scores using households' pre-intervention characteristics (P. Rosenbaum & Rubin, 1983). A dependent variable (using ACI), Y, is a binary variable taking the value 1 indicating adopting ACI. The independent variables (sex, age, family size, education level, and land size, having a mobile phone, Knowledge of heat detection, off-farm income, farm income, and timeliness of ACI service, perception, and distance to AI center) are used to measure the probability of the variable.
In the estimation of propensity score, we are not interested in the effects of covariates on the propensity score because the purpose is to evaluate the impact of using ACI on smallholder households' well-being. However, the choice of covariates to be included in the first step is a vital issue. Heckman et al. (1998), argue that omitting important variables can increase the bias in the estimation. Variables that determine households' decision to adopt AI for cattle breeding could also affect outcome variables. The pre-intervention characteristics, which bring variation in outcomes of interest among users and non-users, will be used. There are no general rules for which variables to include in the model (Andersson et al., 2009). However, the present study is guided by economic theory and empirical studies to know which observable variables affect the outcomes interest (Bryson et al., 2002).
To measure the impact of adopting ACI on the wellbeing of households' three indicators were used. These are income from sales of animals, milk income, and total consumption. Livestock income is a continuous variable that represents households' income from sales of live animals and livestock products. It is because we expect users to have higher livestock income than non-users. It is measured by calculating the total annual livestock income of households from the sale of live animals. Income from milk is a continuous outcome variable. In this case, AI-user households own genetically superior dairy cows, and they are expected to produce more volume of milk than nonuser households and they earn more amount of money. It is measured by using annual milk income during one production year. Household annual consumption is a continuous variable. It is the total household spending on different goods and services, such as total food and non-food expenditure on an annual basis. Generally, the measurement of the variables is presented in Table 2.

Results and discussions
In this section, the first sub-section presents the description of households' characteristics and outcome variables are analyzed with mean, standard deviation, and percentages. However, the second sub-section is the estimation results of the propensity score matching, treatment effect, and sensitivity analysis.

Descriptive statistics of household characteristics
This sub-section describes household characteristics that provide information on demographic and socio-economic characteristics. The summary of socioeconomic features of the household along with the mean difference test (t-test) of continuous variables is presented in (Table-3). After estimating the mean values, the significance of the mean difference test was undertaken by a two-group mean comparison test. The distribution of the categorical variables related to the adopter and the non-adopter household was presented in Table 4. The proportion of the respondents falling into these categories is given, and the differences in the proportion across adopter and non-adopter households were tested by using the chi-square test.

Age of the household head (logAge)
The mean age of the sample households in the study area was 3.7156 years with minimum and maximum ages of 2.94 and 4.45 years. Whereas the mean age of non-adopters was 3.76141 years with minimum and maximum ages of 3.04 and 4.45 years, respectively, and that of the adopter was 3.61997 years, with minimum and maximum values of 2.94 and 4.06 years, respectively. The result revealed a significant difference in the age of household heads between users and non-users of the technology at a 1% level of significance (Table 3). The result indicated that the age of non-  (2022) adopter household heads was higher as compared to adopter household heads in the study area as was expected.

Family size (logFAMSIZ)
The mean family size of the total sample households in the study area was about 1.720134, with minimum and maximum family sizes of 0.6931472 and 2.639057. Whereas the mean family size of non-adopter households was 1.648373 with a minimum and maximum of 0.6931472 and 2.639057, although the mean of adopter households was 1.86979 with a minimum and maximum family size of 0.6931472 and 2.397895, (values are in logarithm form). The descriptive analysis implies that there was a significant difference in the family size of households between adopter and non-adopter households in the study area at a 1% level of significance.

Non-farm income (SQR_NFAINC)
The mean annual non-farm income of the sample households in the study area was birr 90.29427, with minimum and maximum annual non-farm income of birr 0 and 189.842, respectively. But the mean annual non-farm income of a non-adopter household was birr 84.94783 with minimum and maximum annual non-farm income of birr 0 and 177.1384 respectively, whereas that of the adopter household is birr 101.4441, with minimum and maximum annual non-farm income of birr 0 and 189.842, respectively. The descriptive analysis revealed that there was a significant difference in the annual non-farm income of households between adopters and non-adopter in the study area at a 1% significance level.

Livestock holding of sample households (SQR_TLU)
The mean value of livestock of the total sample household, which is measured in the Tropical Livestock Unit (TLU) is 3.460221 with a minimum and a maximum number of 1.414214 and 5.196152, Livestock respectively, and that of adopter respondents was 3.761268, with minimum and maximum values of 2 and 5.196152, respectively, and that of non-adopter households is 3.315867, with minimum and maximum mean values of 1.414214 and 5, respectively, all the values here are in square root form. Households that have a large amount of livestock have participated in the technology more than those who have smaller amount of livestock which is consistent with the hypothesized sign.

Distance from AI Center (SQR_DAIC)
The mean distance from the artificial insemination center to the sample household which is measured in k/m is 1.415 with a minimum and a maximum distance of 0.442136 and 3.162278 k/m, respectively, and the mean distance from adopter households to the artificial insemination center was 1.383667, with minimum and maximum mean values of 0.4472136 and 2.828427, respectively, and that of non-adopter households to the AI center is 1.65134, with minimum and maximum mean values of 0.4472136 and 3.162278, respectively (all the values were transformed to square root form). The descriptive analysis indicates that there is a significant difference in distance between the user and non-user households to the artificial insemination center.

Education level
The proportional years of the education level of the sample households in the study area were primary education of schooling, whereas the adopter households had a maximum and minimum education level of primary and illiterate. And the proportional education level of non-adopter households had a maximum and minimum education level of primary and College & above schooling. The chi-square value, which is 36.74 indicates there is a significant difference in education level between user and non-user household heads.

Timely availability of AI Service (TIMAI)
Out of the total sample respondents about 78.95% of them reported the service is available on time, while 21.05% of them reported timely unavailability of the service, whereas the proportion of timely availability of service reported by a user and non-user respondents was about 88.89% and 82.38%, respectively, and the proportion of timely unavailability of service reported by user and non-user households were about 11.11% and 17.62%, respectively. The chi-square test on this variable indicates there was a significant difference between adopter and non-adopter households at a 1% level of significance.

Knowledge of heat detection
In this specific study from the total sampled households, about 85.60% know how to detect heat periods while 14.40% of them do not know about detecting the heat period, as the same time 92.31% of users households know how to detect heat period but only 7.69% of them have no skill to detect the heat period, and the proportion of non-user households who have the skill and have no skill to detect heat period were about 82.79% and 17.21%, respectively. The chi-square test of this variable shows that there is a statistically significant difference between the two groups at a 1% probability level.

Perception about the AI service
Mostly the entire user households perceive that it is an important technology and 84.8% of the non-adopter households also responded that AI is important, and 15.16% of them felt that it is not important. There is a statistically significant difference in perception of the importance of AI between the user and non-user households at a 1% probability level.

Holding mobile phone
Out of the total sample respondents, about 53.46% of them have a mobile phone while 46.54 % of them have no mobile phone, whereas the proportion of adopter and non-adopter respondents holding a mobile phone were about 66.67% and 47.13%, respectively, and the proportion of not holding mobile phone adopter and non-adopter households were about 33.33% and 52.87%, respectively. The chi-square test result on this variable indicates that there was a significant difference between adopter and non-adopter households at a 1% level of significance.

Access to grazing land
Regarding the accessibility of grazing land out of the total sampled households, about 69.25% of the total household heads have access to grazing land, whereas the proportion of adopter and non-adopter households with access to grazing land was about 83.76% and 62.30%, respectively, which accounts 69.25% of the total household heads. Whereas the proportion of adopter and nonadopter household heads that lack access to grazing land was 16.24% and 37.70%, respectively, this is 30.75% of the total sampled household heads. The chi-square test result on this variable shows that there was a significant difference between adopter and non-adopter households at a 1% significant level.

Descriptive statistics of outcome variables
The outcome variables of this study are income from milk sale, livestock income, and total household expenditure. The difference between the two groups (adopter and non-adopter) concerning the outcome variables are presented at Table 5.

Milk income (SQRMLINC)
It is a continuous variable that represent households' income from milk. The mean value of annual milk income of the sample households in the study area was birr 50.9256, with minimum and maximum annual milk income of birr 0 and 209.7618, respectively. But the mean annual milk income of non-adopter household was birr 32.10991 with minimum and maximum annual milk income of birr 0 and 109.5445, respectively, whereas that of the adopter household is birr 90.16516, with minimum and maximum annual milk income of birr 0 and 209.7618, respectively, all the values here are presented in square root forms. The descriptive analysis revealed that there was a significant difference in the annual milk income of households between adopter and nonadopter households in the adoption practice. The mean difference of those was significant at a 10% significance level. This shows that the milk income of the adopter household was higher than the non-adopter household.

Livestock income (SQR_LVSINC)
Livestock income includes the sum of income from the sale of live animals and animal products. The mean annual livestock income of the sample households in the study area was birr 130.6529, with minimum and maximum annual livestock income of birr 22.36068 and 236.453, respectively. However, the mean annual livestock income of a non-adopter household was birr 117.6178 with minimum and maximum annual livestock income of birr 22.36068 and 224.722, respectively, whereas that of the adopter household is birr 157.8371, with minimum and maximum annual livestock income of birr 81.24039 and 236.453, respectively, all values are in square root form. The descriptive analysis revealed that there was a significant difference in the annual livestock income of households between adopter and non-adopter households in the study area. The mean difference of those was significant at a 10% significance level. This implies that the livestock income of the adopter household was higher as compared to the non-adopter household.

Household total annual consumption/expenditure
It is another and the last outcome variable, which is measured in Ethiopian birr in this study. It is the total household spending on different goods and services (food and non-food). The mean annual consumption of the sampled households in the study area was birr 42,461.76 with minimum and maximum annual consumption of birr 10,010 and 75,482, respectively. But the mean

13.024
Source: Survey result (2022), *** means significant at 1% Lemma et al., Cogent Social Sciences (2023) annual consumption of nonuser households was birr 37,610.75, with minimum and maximum annual consumption of birr 10,010 and 68,225, respectively, whereas that of the artificial insemination technology user households was birr 52,578.39, with minimum and maximum annual consumption of birr 32,851 and 75,482, respectively. The descriptive analysis revealed that there was a significant difference in the annual consumption of households between artificial insemination technology users and non-user households in the study area. The mean difference between the groups was significant at a 10% significance level. This implies that the consumption of the user household was higher as compared to non-user households in the study area.

Empirical result
Before proceeding with impact estimation Variance Inflation Factor (VIF) was conducted to test for the presence of multicollinearity problem among explanatory variables. As presented in Table 6, there was no explanatory variable dropped from the estimated model since no serious problem of multicollinearity was detected in the VIF test, because it is less than the cut-off point of 10.
Similarly, the heteroscedasticity test was checked by using Breusch-Pagan test. This test resulted in the rejection of the existence of the heteroscedasticity hypothesis and there was no need to make standard error robust since there is no heteroscedasticity problem. In addition, a model specification was checked by using the Ramsey RESET test. The result indicates that the p-value was 0.1733 which is greater than 0.05, therefore we accept the null hypothesis meaning there is no model specification error indicating that no variable is dropped from the model. Finally, the Goodness-of-Fit test for the Logistic Regression model was conducted by using Hosmer-Lemeshow Test. The result of the test indicates that our model fits reasonably well. The logistic regression model was applied to estimate the propensity score matching for the ACI adopter and non-adopter households. The logit regression result presented in Table 7 revealed a fairly low pseudo-R 2 of 0.3812. A low R 2 value shows that program households do not have many distinct characteristics and finding a good match between adopter and non-adopter households becomes easier. The logit estimate shows that the adoption of artificial insemination is significantly influenced by eight explanatory variables, while the remaining five variables were not significant.
The distribution of the propensity score for each household included in the treated and control groups was computed based on Table 7, participation model to identify the existence of common  (2022) support. Figure 2 depicts the distribution of the household concerning the estimated propensity scores. The figure shows that most of the treatment households were found in the middle and partly on the right side while most of the control households are partly found in the center and partly on the left side of the distribution. It also reveals that there is a wide area in which the propensity score of both treatment and control groups are similar.

Matching adopter and non-adopter group
As indicated in Table 8, the estimated propensity scores for ACI adopter household vary between 0.0033287 and 0.9915505 (mean = 0.6155176) and 0.0000106 and 0.8501441 (mean = 0.1843625) for non-adopter (control) households. The common support region, therefore, lie between 0.0033287 and 0.8501441 and the balancing property is satisfied to the final number of block 5. The number of blocks ensures the mean propensity score is not different for treated and controls in each block. This means households whose estimated propensity scores are less than 0.0033287 and larger than 0.8501441 were not considered for the matching purpose. As a result of this restriction, 50 households were discarded. It is good that abundant respondents from the sample in computing the impact estimator were not dropped.

Choice of matching algorithm
Different matching estimators were used in matching the treatment and control households in the common support region. The final choice of a matching estimator was guided by different criteria such as the equal means test or balancing test (Dehejia and Wahba, 2012), pseudo-R 2 , and matched sample size. The equal means test (the balancing test) suggests that a matching estimator balance all explanatory variables (insignificant mean differences between the two groups) after matching. In case of the pseudo-R 2 , the smallest value is preferable. Also, matching estimator that results in the largest number of matched sample sizes is preferred. Therefore, a matching estimator that balances all explanatory variables, with lowest pseudo-R 2 and, produces a large matched sample size is preferable. Looking into the result of the matching quality presented in Table 9, it is shown that kernel matching with a bandwidth of 0.1 was found to be the best for the data we have at hand. Hence, the estimation results and discussion for this study are the direct outcomes of the kernel matching algorithm with a band of 0.1. Finding a consistent estimate of adoption impact on household well-being necessitates controlling for all such confounding factors. In doing so, propensity score matching has resulted in 87 adopter households being matched with 224 non-adopter households (one-to-many matched) after discarding 30 participants and 20 control households whose values were out of the common support region.

Testing the balance of propensity score and covariate
The balancing powers of estimations were ensured by different testing methods. Reduction in the mean standardized bias between the matched and unmatched households, and equality of means using t-test and chi-square test for joint significance of variables were employed. The fifth and sixth columns in Table 10 show the standardized bias before and after matching, and the total bias  reduction obtained by the matching procedure, respectively. The standardized difference in covariates before matching is between 3.7% and 71.9% in absolute value, whereas standardized difference of covariates for almost all covariates lies between 0.2% and 12.4% after matching. This is fairly below the critical level of 20% suggested by (Rosenbaum and Rubin, 1985). Therefore, the process of matching creates a high degree of covariate balance between treatment and control samples that are ready to use in estimation procedure. Similarly, t-values in Table 10 show that before matching all the chosen variables except farm income and sex of household head exhibited statistically significant differences. However, after matching all covariates became insignificant and the variance ratio for all covariates after matching are less than the critical value, which is two indicating covariates are fully balanced.
The low pseudo-R 2 , the insignificant likelihood ratio tests, the B-value less than 25, and the R-value lies between the critical point and the mean biases less than 20 support the hypothesis that both groups have the same distribution in covariates X after matching (Table 10).
As the tests presented in Table 11 indicated, the matching algorithm used is relatively best for the data. This indicates that using this result we can examine the impact of adopting artificial cattle insemination by smallholder farmers on their milk income, livestock income, and total consumption among groups of households having similar observed characteristics. Thus, we can proceed to estimate the average treatment effect on the treated (ATT) for the sample households.

Estimating treatment effect on treated (ATT)
To attain the stated objectives the following impact indicators (presented in Table 12) of the treatment effect have been performed using the PSM model.  (2022), * denotes the number of explanatory variables with no statistically significant mean differences between the matched groups of the adopter and non-adopter households.  As presented in Table 12, the impact of artificial cattle insemination technology on milk income, livestock income, and total household expenditure are evaluated on households after preintervention differences were controlled. The result presented in Table 12 shows that income from milk for ACI adopter households is greater than that of non-adopter households. The ATT for the treated is 93.99 while that of control is 31.25. The result is statistically significant at 1% (t = 10.02) level. Many literatures indicate that income from milk is one of the most widely used proxy variable to measures household wellbeing. After controlling for differences in characteristics of the adopter and non-adopter households, it was found that, on average, the adopter household has increased annual income from milk by 62.7418143 (66.75%) as a result of the adoption of artificial insemination. Artificial insemination service users (treated) groups yield a higher volume of milk than non-users (control) groups so they earn more income from milk in agreement with the hypothesized sign. The finding is also consistent with the findings of Diro Chelkeba et al. (2016) and Yohannes (2014).
Similarly as shown in Table 12, there is difference in livestock income among ACI technology adopter and non-adopter households. The impact of adopting artificial cattle insemination on livestock income shows the presence of significant difference between the treated and control groups. The sum of live animal sales and animal products except milk are included in the livestock income. Users earn higher income from all the included livestock income sources and it shows a positively significant impact. This indicates that the adoption of artificial insemination has a positive influence on livestock income of adopter households. The study shows that the livestock income of adopter households is greater (153.86) than that of non-adopter (122.65) and the artificial insemination adoption practice has positive and statistically significant at 10% (t = 5.88). The finding of the present study is consistent with the findings of Diro Chelkeba et al. (2016) which shows the positive effect of adopting ACI technology by households compared non-adopter households.
The result presented in Table 12 also shows that the consumption of ACI technology adopter households significantly increased compared to non-adopter households. The estimation result presented shows that the total consumption of adopter households is greater (ATT = 52790) than that of non-adopter households (ATT = 41464.53). Artificial cattle insemination technology adoption practice has positive and statistically significant effect at 10% (t = 6.57). As adopter households have better income opportunities they spend also more. This means the average treatment effect of total household expenditures also shows significant difference between the two groups in the study area. After controlling for differences in characteristics of the adopter and non-adopter households, it was found that, on average, the adopter household has more expenditures or increased total expenditures by 11,325.6945 birr than non-adopter households due to using artificial insemination technology. The finding is consistent with the finding by Hana (2019).

Triangulation with Qualitative information.
In addition to the quantitative analysis, the result is also validated by key-informant interview and focus group discussion (FGD). Most of the FGD discussants and key informants testified that the ACI technology has significantly increased the income of the beneficiary households. From FGD and key informants, it has been confirmed that adopter households have improved the breed of cows by using ACI technology and have achieved high results in their species. They also provided evidence that adopter households had a significant increase in their milk income, livestock income, and annual consumption, consistent with the empirical results stated above. Therefore, it is confirmed that the findings from the key informant interview and FDG were reliable with the empirical findings shown in Table 12.
Two key-informants from Fofa Kebele and Toba city development agents confirmed that ACI adopter farmers have more income and better life than the non-adopters. These two informant interviewees described that the adoption of the technology as facilitating food and non-food consumption, creating employment opportunities, paying children's school fees, covering medical expenses, purchasing agricultural inputs, paying for social spending, and other necessary expenses. Finally, consistent with the survey results, the findings from FGD and key informant interviews clearly show that the adoption of artificial insemination technology has a significant and positive impact on the adopters' wellbeing in the study area.

Sensitivity Analysis.
To control for unobservable biases, Table 13 shows the result of the sensitivity of artificial insemination effects on different outcome variables. There may be hidden biases against the result of matching estimators and hence testing the robustness of the result is recommended. As it is not possible to estimate the magnitude of selection bias with nonexperimental data, the problem can be addressed by using a sensitivity test. The basic issue in testing sensitivity is to check whether the treatment effect is due to unobserved factor or not. P. R. Rosenbaum (2002) proposes using Rosenbaum bounding approach to check the sensitivity of the estimated ATT.
Where, e γ (Gamma) is the log odds of differential due to unobserved factors and where Wilcoxon significance for continuous variable and level for each significant outcome variable is calculated. Table 13 denotes the critical level of e γ (first row), at which the causal inference of significant artificial insemination effect has to be questioned. As noted by Hujer et al. (2004), sensitivity analysis for insignificant effects is not meaningful and is therefore not considered here. Given that the estimated artificial insemination adoption effect is positive for the significant outcomes, the lower bounds under the assumption that the true treatment effect has been underestimated were less interesting (Becker and Caliendo, 2007) and therefore not reported in this study. Rosenbaum bounds were calculated for artificial insemination adoption effects that are positive and significantly different from zero. The first column of the table shows those outcome variables which bears statistical difference between treated and control households. The rest of the values which correspond to each row of the significant outcome variables are p-critical values (or the upper bound of Wilcoxon significance level-Sig+ for continuous outcome variable and mantel hevens (mhbound) upper bound significance level for categorical variable) at a different critical value of e γ . In this specific study, all outcome variables are continuous, and no need to conduct mantel hevens (mhbound).
The result shows that the implication for the effect of the artificial insemination technology interventions is not changing, though the participants and non-participant households have been  (2022) allowed to differ in their odds of being treated up to 200% (e γ ¼ 3) in terms of unobserved covariates. That means for all outcome variables estimated at various levels of critical value e γ , the p-critical values are significant, which indicates that we have considered important covariates that affected both participation and outcome variables. We couldn't get the critical value e γ , where the estimated ATT is questioned even if we have set e γ largely up to 3, which is a larger value compared to the value set in different literature, which is usually 2 (100%). Thus, we can conclude that our impact estimates (ATT) are insensitive to unobserved selection bias and are a pure effect of artificial insemination technology adoption.

Conclusion and recommendations
The objective of the study is to analyze the impact of adopting artificial cattle insemination by smallholder farmers on their wellbeing in the case of Yem special district. The study applied PSM technique which has become the most widely applied non-experimental tool for impact evaluation. The findings of the study showed that adopting artificial insemination technology has positive effects on the welfare of adopter households. As a result, ACI adopter households have better wellbeing. The impact estimate indicates that there are significant differences in wellbeing between treatment and comparison groups, which are attributable to ACI adoption practice. The effect of the artificial insemination adoption on household milk income, total expenditure, and livestock income is higher for the adopter households. In a nutshell, it is concluded that adopting artificial insemination brought significant impact on the improvement of the wellbeing of rural farm households by creating opportunities to increase their overall income. This leads to the conclusion that adopting technology is one of the basic instruments to enhance the leaving standard of farming communities in most rural parts of Ethiopia. Based on the findings, the following recommendations are forwarded. It is better to encourage the adoption of artificial cattle insemination technology because it increases both income and consumption of households. ACI centers should be opened at the nearest areas of households.