“Analyzing the determinants of rural household welfare in Ethiopia: Does agricultural technology adoption matters”: Using seemingly unrelated multi-variate probit model”

1.1 Motivation of the study

As elucidated in numerous scholarly investigations, Ethiopian citizens continue to endure diminished welfare, as evidenced by indicators such as livelihood conditions, food security, and poverty levels (Sebsibie, Asmare, & Endalkachew, 2014). A prime example of this is the elevated incidence of poverty and household food insecurity in rural Ethiopia, which constitutes a formidable obstacle in the pursuit of sustained economic growth and development (Zegeye et al., 2022). The interim report further underscores that rural poverty remains disproportionately higher than its urban counterpart across all temporal intervals (Abate, 2024a). Notably, the nation-wide reduction in poverty is predominantly attributed to the alleviation of urban poverty, although both rural and urban areas have witnessed a general decline in poverty (measured in both incidence and depth). Nevertheless, the income distribution among the rural poor has shown negligible improvement (Ayenew et al., 2020).

In contrast, the prevailing drivers of hunger and impoverished living conditions in Ethiopia are widely considered to be a confluence of factors, including the widespread devastation wrought by the COVID-19 pandemic, civil unrest, internal strife, and constrained agricultural productivity. These elements are encompassed within the scope of this study (Bekana & Takele, 2020). To ameliorate the prevailing low welfare and living standards, it is imperative to analyze both the extent and the determinants of welfare, alongside its intricate relationship with the adoption of agricultural technology. Thus, the overarching aim of the present study is to explore the nexus between rural household welfare and the uptake of agricultural technology within the Ethiopian context. Specifically, the study seeks to probe the interdependencies between various welfare metrics and examine the influence of agricultural technology adoption on rural household welfare.

This research contributes to the extant body of literature in three distinct ways. First, it utilizes the seemingly unrelated multivariate probit model to investigate the impact of agricultural technology adoption on household welfare (Atinafu et al., 2022). This methodological innovation provides new insights by simultaneously considering three welfare dimensions uni-dimensional poverty, multidimensional poverty, and food insecurity rather than relying on a singular measure, as is commonly the case in much of the existing research (Mulugeta Habtewold & Heshmati, 2023).

Secondly, this study considers the effect of recurrent internal conflicts and wars on the welfare of households, which was not previously considered. Thirdly, this study analyzes the interdependence of rural household welfare measurements instead of using only one proxy variable to indicate the welfare of a society. This means this study is the first to check whether households facing multidimensional poverty are also facing food insecurity.

The rest of the paper follows this structure, and the next section presents a concise review of the theoretical literature on the factors that affect rural household welfare, with special emphasis on agricultural technology adoption. It also includes empirical literature on the determinants of households’ welfare, both in Ethiopia and globally. Section III outlines the methodology, followed by a discussion of the estimation results in Section IV. The final section provides conclusions and recommendations.

3.1 Data and sampling procedure

The data used in this study is secondary data obtained from ESS (Ethiopian Socio-economic survey). The Ethiopian Socioeconomic Survey (ESS) data was collected by the Central Statistical Agency (CSA) of Ethiopia in collaboration with the Living Standards Measurement Study Integrated Surveys on Agriculture (LSMS-ISA) of the World Bank. The data collection process included the following steps. The ESS sample was designed to be representative of Ethiopia’s small towns and rural areas. The sampling frame included households from the Amhara, Oromia, Southern Nations, Nationalities, and Peoples’ Region (SNNP), Tigray, Afar, Benshangul Gumuz, Gambella, Harari, and Somali Regional States, as well as the city government of Dire Dawa. The survey involved a two-stage stratified sampling technique. In the first stage, enumeration areas (EAs) were selected from each region based on probability proportional to size. In the second stage, households within these EAs were randomly selected. A total of 4,870 households were included in the sample for the study. This sample size was determined to ensure that the survey results would be statistically significant and representative of the broader population. Data was collected through household interviews using structured questionnaires. The survey covered various socio-economic indicators, including household demographics, education, health, employment, agriculture, and income. To ensure the quality and accuracy of the data, field supervisors conducted regular checks and validation of the collected data. The data collection process also involved training enumerators and supervisors on the survey instruments and data collection procedures.

3.2 Methods of analysis

Both qualitative and quantitative methods of analysis with descriptive and econometric applications are used. Descriptive analysis is used to investigate the extents of poverty and food insecurity in the study area while econometric analysis is applied to identify the determinants of rural household’s welfare by considering agricultural technology adoption as the major factor to affect welfare of the society. Uni-dimensional Poverty (Wossen, Alene, Abdoulaye, Feleke, & Manyong, 2019) can be measured using per adult consumption expenditure per households and individual household is uni-dimensional poor if they are living under the absolute poverty line of the area. Household with per adult consumption expenditure represented by “Y” and poverty line by “Z”, and then poverty is defined by $Y^{\ast }=\mathrm{Z}-\mathrm{Y}$ , hence household is poor if $Y^{\ast }$ > 0.

Multidimensional poverty measure non income poverty by considering deprivations of households for basic needs. In this study, the UNDP Multidimensional poverty index is developed using ten indicators categorized under three dimensions namely health (nutrition and child mortality), education (school enrollment, years of schooling and living standard with indicators of access to water, electricity, sanitation, cooking fuel, floor and, assets (Kovacevic & Calderon, 2014).

Principal component analysis (PCA) was used to estimate the household food security index. Similarly, PCA was employed in empirical research to create indices of household food security (Hawas & Degaga, 2023). Researchers created a food security index for this analysis using food consumption score, dietary diversity, consumption expenditure, calorie intake, and hunger months. The PCA linearly divides the indicator variables into smaller components while retaining most information from the original indicators (Massresha et al., 2021). Stated mathematically, from an initial set of correlated indicator food security variables (X₁, X₂, X₃…, X _n), PCA creates uncorrelated indices or components whereby each component is a linear weighted combination of the initial variables as follows:

where FSI. j the food security index follows a normal distribution with a mean of 0 and a standard deviation of 1. F _i is the weight for the ith variable in the PCA model, X _ji is the jth ’ ‘household’s value for the ith variable, and X _i and S _i are the mean and standard deviations for all households.

3.3 Model specification

Seemingly unrelated multivariate probit model is applied for this research. The seemingly unrelated regression model is preferable (Baltagi, 2012) and (Pan et al., 2020) when there are more than one regression models and the dependent variables are affected by similar co-variate or interdependent variables (Kim & Cho, 2019). Rural household’s welfare for this study is measured using two dummy variables of multi-dimensional poverty and food insecurity status when those variables are determined by similar co-variate. The regression model is developed as follows:

Yi = Rural household welfare and has a multiple category of 1 for unidirectional poverty, 2 for Multidimensional poverty and 3 for Food insecurity

Where y1 = 1 for uni-dimensional poor, y2 = 1 for multidimensional poor and y3 = 1 for food in-secured and the variance of the co-variate error is given by

My merging uni-dimensional and multi-dimensional poverty in to one variable (to be poverty) from the above multi-variate probit model, the bi-variate probit regression model is also specified as follows;

Yi = Rural household welfare and has two categories of 1 for Multidimensional poverty and 2 for Food insecurity (1 is assigned for poor and food in secured)

Where y₁ = 1 for poor and y₂ = 1 for food in-secured and the variance of the co-variate error is given by

3.4 Description of variables and expected signs

Welfare as measured by three proxy variables of uni-dimensional poverty, multi-dimensional poverty and food insecurity is expected to have specified relationship with explanatory variables as presented in the following Table 1.

4.1. Descriptive analysis

Under this descriptive analysis part, the issue of gender is addressed and what proportion of female headed farmers are food-insecure and poor is analyzed. For this sake, the bar graph presented below shows the proportion of females from the total sample and how many of those households are living under low welfare as measured by poverty and food insecurity.

Out of a total sample of the study, 47.61% (2322 out of 4877) are adopting agricultural technologies in many forms. The remaining 52.39% are not using modern agricultural inputs and irrigation to increase the productivity of their agricultural outputs. Among those their welfare status is analysed based on their gender as presented below ( Figure 2).

Figure 2. Bar chart of gender and food insecurity.

Out of a total sample of 4887 rural household farmers in Ethiopia, more than 3787 farmers are male headed and the remaining 1090 households are female headed ( Figure 1). Out of male headed households, about 68% are food secured and the remaining 32% are food insecure households. When we see their female headed counter parts, out of a total of 1090 female headed rural households, about 70% are food secured and the remaining 30% are food insecure This shows us there is no significant difference of food security status based on their gender.

The below Figure 2 presents the male female composition of the sample and the numbers households living under multi-dimensional poverty. About 67% of male headed households are facing multi-dimensional poverty and about 71% of female headed households are multi-dimensional poor. From this result we can observe that majority of the population is facing multi-dimensional poverty either they are male headed or female headed regardless of their gender differences.

Moreover, the provision of credit across the male and female-headed households are analyzed here below using pie chart ( Figure 3). Majority of female-headed households accessed credit provision when compared with their male-headed counter parts.

Figure 3. Multi-dimensional poverty across gender.

As seen from the above pie chart, 36% of female-headed households accessed credit even though female-headed household’s accounts for about 40%of the total sample and the remaining 4% are not accessing credit. From those no credit accessed categories, 96.32% are male-headed households and hence we can say that more priority is given for females in the area of credit provision from governmental and non-governmental donors ( Figure 4).

Figure 4. Gender and access to credit.

After analyzing the data obtained from the focused group discussion, the following findings had been drawn relating to the impact of internal conflict on rural household’s agricultural activities and their welfare. Internal conflicts and wars have profound and far-reaching impacts on rural household welfare, particularly in agricultural contexts such as Ethiopia. These impacts can be broadly categorized into several areas:

Displacement and migration: Conflicts often lead to the displacement of populations, forcing rural households to abandon their homes and farmland. This displacement disrupts agricultural activities, leading to loss of crops, livestock, and income sources. Migration also increases competition for resources in safer areas, exacerbating food insecurity and poverty.

Destruction of infrastructure: Wars and conflicts typically result in the destruction of critical infrastructure such as roads, irrigation systems, storage facilities, and markets. The damage to infrastructure hampers the ability of farmers to access markets, sell their produce, and obtain necessary agricultural inputs like seeds and fertilizers.

Loss of agricultural productivity: The direct impacts of conflict, such as destruction of crops and livestock, combined with the disruption of farming activities, result in significant losses in agricultural productivity. Farmers may miss planting and harvesting seasons, leading to reduced yields and food shortages.

Economic instability: Conflicts lead to economic instability, characterized by inflation, loss of income, and decreased investment in the agricultural sector. The uncertainty and risk associated with conflict discourage both local and foreign investment, reducing the financial resources available for agricultural development.

Food insecurity: The combination of displacement, infrastructure destruction, and reduced agricultural productivity culminates in heightened food insecurity. Rural households struggle to produce and purchase sufficient food, leading to malnutrition and increased vulnerability to diseases.

Social disruption: Conflicts erode social cohesion and trust within communities. Traditional support systems, which are crucial for coping with economic and environmental shocks, are weakened. This social fragmentation makes it more difficult for rural households to recover and rebuild their livelihoods post-conflict.

Loss of human capital: Wars and conflicts result in the loss of human capital through death, injury, and the disruption of education. Now currently, internal conflict is experienced in many parts of the study area and it leads to devastating human capital level. The absence of able-bodied family members reduces the household labor force, impacting agricultural productivity. Additionally, children’s education is often interrupted, limiting future opportunities and perpetuating the cycle of poverty.

Psychological impact: The trauma and stress associated with living in conflict zones have long-term psychological impacts on individuals and communities in the study area. Mental health issues can reduce the ability of individuals to engage in productive activities, further diminishing household welfare.

4.2 Econometrics analysis

Under this section, the effect of agricultural technology adoption had been analyzed using econometric results obtained from bi-variate probit regression model. First results obtained from the bi-variate probit (Table 2) is used to analyze the relationship that exists between Welfare and its determinant with special emphasis of the effect of agricultural technology adoption of rural household’s welfare. With this regard, overall significance test of the model is applied using Chi-square test and individual variables statistical significance is tested using simple students t- test with the application of general rule of thumb (the null hypothesis H0: estimated coefficients are zero can be rejected when t-statistics values are greater than two).

The maximum likelihood estimation result of the bivariate probit model gives us a chi-squared probability value of zero (0.000), indicating the overall significance of the model by rejecting the null hypothesis that states the coefficients of all variables are zero. In addition to the overall significance test, all independent variables are tested to determine if they are significant using Z and probability values. Variables with Z values greater than two or P-values less than 0.05 (Baltagi 2012) are statistically significant in affecting the level of rural welfare as measured by poverty and food insecurity.

Negative coefficients for poverty and food insecurity suggest that adopting the specified agricultural technologies is associated with a lower likelihood of being in each respective state of poverty or food insecurity. It’s important to consider the statistical significance of these coefficients (t-values) and their practical significance in the context of the study.

As seen from Table 2, all independent variables included in the model, except for access to credit, education, and access to pesticides, are statistically significant in affecting both poverty and food insecurity. Accordingly, among indicators of agricultural technology adoption, access to irrigation (Tilahun, Bantider et al. 2023), access to improved seeds (Mutungi, Manda et al. 2023), and use of fertilizers (Ayenew, Lakew et al. 2020) significantly and positively affect the status of being non-poor and food-secure households (Biru, Zeller et al. 2020). More specifically, the results of the bivariate probit model state that having access to irrigation and improved seeds is associated with respective probabilities of 0.014 and 0.049 of being under the non-poor and food-secured category when compared with households without access to irrigation and improved seeds. Adopting improved seeds is associated with a statistically significant reduction in the likelihood of poverty. The negative coefficient indicates that, holding other variables constant, adopting improved seeds is linked to a decrease in the probability of being poor. Adopting irrigation has a strong and statistically significant negative association with poverty. Farmers using irrigation are much less likely to be poor compared to those who do not adopt irrigation. This result is supported by many research findings such as (Legesse, Srivastava et al. 2019; Belay and Mengiste 2023; Atinafu, Lejebo et al. 2022; Korir, Manning et al. 2023). Existing research often supports the idea that adopting agricultural technologies like improved seeds leads to increased crop yields and, subsequently, improved welfare of farmers. This is consistent with findings from (Biru, Zeller et al. 2020; Asrat, Anteneh et al. 2022; Habtewold 2018) that state improved seeds are associated with a lower likelihood of poverty and food insecurity. This result contradicts the findings of (Wossen, Alene et al. 2019; Wordofa and Sassi 2020; Tilaye, Delele et al. 2023) who state that their results deviate from the theory due to the target population and method of sampling used in the study.

Additionally, using one kilogram of fertilizer is associated with increasing the probability of being non-poor by 0.007. Adopting fertilizers is associated with a statistically significant decrease in the likelihood of poverty, as stated by Mulugeta Habtewold and Heshmati (2023), Mogess and Ayen (2023), Legesse, Srivastava et al. (2019) Gebre, Amekawa et al. (2023), though the effect is not as strong as with improved seeds. The negative coefficient suggests a reduction in the probability of poverty when farmers adopt fertilizers.

In this study, the age of households and dependency ratio are associated with a high probability of being poor and food-insecure households. When the age of the household head increases by one year, the probability of being poor and food insecure increases by 0.013 and 0.067, respectively. The remaining variable affecting poverty and food insecurity status is the number of livestock owned by households (Mekonnen 2017; Korir, Manning et al. 2023). One-unit increase in livestock is associated with increasing the probability of being non-poor and food-secure by 0.137. Hence, we can say that the number of livestock is a key determinant of household welfare, and an increased number of livestock has the power to reduce the level of poverty and food insecurity (Shikur 2023). Existing research often supports the idea that adopting improved seeds leads to increased crop yields and, subsequently, improved economic outcomes for farmers. This is consistent with the findings of (Bekana and Takele 2020; Belay and Mengiste 2023; Mogess and Ayen 2023), who also state that the adoption of improved seeds is associated with a lower likelihood of poverty and food insecurity.

Variables such as access to credit, use of pesticides, and education level are statistically insignificant (Gebrehiwot 2015; Habtewold 2018), and access to agricultural extension service is negatively related to rural household welfare as measured by proxy variables of poverty and food security, in contrast to the results of (Nonvide 2024; Kebede 2022; Hawas and Degaga 2023). These results are somewhat unexpected and deviate from theories. Researchers analyzed the reason behind these results as follows:

Access to credit had an insignificant impact on rural household welfare in the study area because many agricultural credits are provided by credit associations to farmers with high interest rates, creating difficulty in repaying the credit. Instead of improving the welfare of rural households, high-interest payments create tension, and farmers are obliged to sell their assets to repay the credited money with high interest, negatively affecting household welfare. Agricultural extension service support is not improving the welfare of households, and farmers with extension support are relatively under low welfare status, necessitating a re-evaluation of the relevance of agricultural extension. Additionally, the education level is statistically insignificant since almost all farmers in rural Ethiopia are either illiterate or have only completed the first cycle of elementary education, resulting in no significant difference in welfare associated with their levels of education. In addition to the bi-variate model, the seemingly unrelated multi-variate probit model with few selected independent variables of the research are presented in the table below (Table 3) and results are interpreted and analyzed as follows.

The overall significance of the model, with a p-value less than 5% (Table 3), showed that the model is developed with significant co-variates since the null hypothesis, which states that all variables coefficients are zero, can be rejected. With this in mind, we proceed to the interpretation of statistically significance variables by examining individual hypothesis testing results for all variables.

According to the estimation result from the multi-variate probit model ( Table 3), the size of cultivated land owned by household farmers significantly affect their welfare. Land size is positively related with food security but inversely related with uni-variate and multi-variate poverty. This shows, when household’s cultivated land size increased by one hectare, the probability of being food secured increased by 0.047, while increased land size is associated with a decrease in the probability of being multidimensional and multidimensional poor by 0.006 and 0.0055 respectively. This means that a large size of cultivated land may help households to be food secured (Shita et al., 2023), but households with increased land size are living under uni dimensional and multi-dimensional poverty. This result aggres with the findings of Mekonnen (2017) and Habtewold (2021). Additionally, provision of agricultural extension services, numbers livestock are significance variables affecting rural households, welfare as measured by three proxy variables of food insecurity, uni dimensional poverty, and multi dimensional poverty. This result is supported by the bi-variate probit model regression result analyzed previously ( Table 2). This result is also similar to the findings of Gelata et al. (2024), Atinafu et al. (2022) Habtewold (2018), Gebrehiwot (2015) and Wordofa & Sassi (2020). From the probit regression model, we observed the interdependence of three measures of welfare, all of which are significantly affected by similar explanatory variables. This means that individual household farmers which are facing food insecurity are also facing both uni-dimensional and multi-dimensional poverty. Farmers living under uni-dimensional poverty are also facing multidimensional poverty and food insecurity. Finally, multi-dimensional poverty is occurred in line with food insecurity and uni-dimensional poverty. This result is supported by the findings of Merga et al. (2023).

5.1 Conclusion

This study analyzed the determinants of rural household welfare in Ethiopia, with a special emphasis on the effect of agricultural technology adoption on improving the welfare of rural household farmers. The analysis using the bivariate probit regression model revealed several key findings regarding the relationship between agricultural technology adoption and rural household welfare in Ethiopia. From descriptive summary, it is noted that out of a total sample of the study, 47.61% (2322 out of 4877) are adopting agricultural technologies in many forms more than 70% of female headed households are food secured and no poor and 68% of male headed households are food secured and nun poor. The overall significance test of the model indicated its relevance in explaining the determinants of welfare, particularly the impact of agricultural technology adoption. The results showed that variables such as access to irrigation, improved seeds, and fertilizers were statistically significant in reducing the likelihood of poverty and food insecurity. These findings are consistent with previous research, suggesting that adopting these technologies can lead to increased crop yields and improved welfare for farmers. However, variables like access to credit, education, and access to pesticides were found to be statistically insignificant, indicating the need for further investigation into their impact on welfare outcomes.

Moreover, the seemingly unrelated multivariate probit model provided additional insights into the relationship between selected independent variables and welfare indicators. The results highlighted the significance of variables such as cultivated land, agricultural extension services, and livestock ownership in affecting welfare outcomes. Larger cultivated land size was associated with increased food security but decreased uni-dimensional and multidimensional poverty, suggesting a complex relationship between land size and welfare. Additionally, the provision of agricultural extension services and livestock ownership were found to have a positive impact on welfare, consistent with the findings from the bivariate probit model.

5.2 Recommendations

Based on the analysis of the relationship between agricultural technology adoption and rural household welfare in Ethiopia, the following recommendations can be made to improve welfare outcomes:

Promotion of Agricultural Technologies: Policymakers should prioritize the promotion and adoption of agricultural technologies such as irrigation, improved seeds, and fertilizers. This can be done through targeted extension programs, subsidies, and incentives to encourage farmers to adopt these technologies. Providing access to high-quality and affordable agricultural inputs can significantly improve crop yields and enhance food security.

Enhanced Access to Credit: Although access to credit was found to be statistically insignificant in this study, efforts should be made to improve access to credit for rural farmers. This can be achieved by establishing rural credit institutions, providing credit guarantees, and offering low-interest loans to farmers. Access to credit can help farmers invest in agricultural inputs and technologies, leading to increased productivity and improved welfare.

Improvement of Education: Although the education level was found to be statistically insignificant in this study, improving access to education can have long-term benefits for rural households. Policymakers should focus on improving the quality and accessibility of education in rural areas, including investing in school infrastructure, providing scholarships, and promoting adult education programs. Education can help improve livelihood opportunities and empower rural communities to make informed decisions about their welfare.

Targeted Poverty Alleviation Programs: Based on the findings regarding age and dependency ratio, policymakers should consider implementing targeted poverty alleviation programs for vulnerable groups such as the elderly and households with high dependency ratios. These programs can include social safety nets, cash transfer programs, and food assistance to help improve the welfare of these groups.

Further Research and Monitoring: Finally, further research is needed to explore the reasons behind the insignificant impact of certain variables and to identify additional factors that may affect rural household welfare. Continuous monitoring and evaluation of welfare programs can help policymakers tailor interventions to address specific challenges and improve overall welfare outcomes in rural areas.