The study was conducted in the Majang zone, Gambella People’s Regional State, southwestern Ethiopia. Approximately 620 kilometers separate Addis Abeba and the zonal capital of Meti. There are two Districts in the area: Godere and Mangeshi, with an agroecology that is considerably wider and elevations that range from 800 m to 2100 m above sea level. The region is situated between latitudes 7° 4′ 2.41″ and 7° 46′ 47.79″N and 35° 38′ 48.00″E to 34° 36′ 30.54″E. According to Mathewos & Bewuketu (2018), the zone has a total land area of 2252.79 km ². The zone is bordered by the Oromia Regional State in the north, the Southern Western Regional State in the east and south, Abobo in the northwest, and Gog and Jor in the west. The Zone has a total population of 89033, with 46119 men and 42914 women, as of the CSA’s (2013) projection for the 2022 census. Each of the 32 Villages administered by Godere and Mangeshi Districts is habituated with projected populations of 61079 and 27954, respectively. With an average of 5.3 people per household, 88% of the population lives in rural areas, and more than 60% of that population is under 20 years old.

The Majang zone is one of Ethiopia’s few areas endowed with abundant natural resources. The Majang Forest Biosphere Reserve is the name given to the forest and its resources, which serve as the community’s primary source of income, by the United Nations Organization for Science, Culture, and Education (UNESCO). Commercial farming (coffee), agriculture, honey production, fishing, hunting, foraging for fruits and spices in the forest, small and petty trades, and commercial farming are the main sources of subsistence in the study region. NTFPs, wood products, and traditional forest honey production provide for almost 87% of all households’ income. Raising livestock is another prevalent habit; however, it is not a significant source of income.

While the Godere district is dominated by other ethnic groups, including the Sheka, Kafa, Oromo, Amhara, Sheko, and Tigre, who are frequently referred to as the “highlanders” by native Majang populations, Mengeshi district is a Majang community-dominated woreda and is located in the lowland agro-climatic zone. Part of Godere district’s agro-climatic conditions are in the lowlands, but the majority are found in the mid- and high-altitude regions. The region has a hot, humid environment, and according to Ethiopia’s rainfall maps, it is the wettest place in the nation. Around 2100 mm and 20–33 °C are believed to be the mean annual temperature and rainfall, respectively. A flat to gentle slope, occasional rocky steep slopes, and deep valleys on the hills and along important streams define the environment. The majority of the streams are perennial and have relatively significant water discharges, but rural populations worry that deforestation is reducing the amount of water flow. The soils in the region are mostly dystric nitrate and range in color from red-brown to dark brown. Agriculture, grazing land, woodland, habitation, wetland, infrastructure, and wasteland are the main land cover types. Forestry and agriculture both cover the majority of the zone’s land.

The study was conducted in the Gambella People’s Regional State’s Majang Zone. Since combining qualitative and quantitative household data in a single research project enables a thorough and holistic understanding of the issue of interest, a mixed-methods design was utilized for the study ( Degefa, 2006). Furthermore, Ciani & Romano (2014) asserted that the pragmatic approach, which combines quantitative and qualitative approaches, is becoming increasingly popular in studies of resilience to food poverty. The Food and Agriculture Agency (FAO) was the first agency to use the idea in food security contexts and has extensive experience evaluating resilience using both quantitative and qualitative methodologies ( FAO, 2016a). The embedded design was picked for the study out of the six mixed designs that Creswell (2012) suggested. The embedded design allows for the simultaneous or sequential collection of both qualitative and quantitative data, but the quantitative method is employed most often, and the results from the qualitative data are used to support the findings of the quantitative analysis. Cross-sectional data types are best associated with embedded design.

The study households were chosen using a multistage selection technique. First off, since there are only two districts in the zone and they all have comparable livelihoods and administrative boundaries, Mengeshi and Godere were chosen on purpose. Second, out of the 32 villages, 10 villages—four in Godere and six in Mengeshi—were chosen systematically by random sampling under the premise that a high sampling ratio (about 30%) was thought to be adequate for small populations (1000). The sample villages were chosen based on the presumption that they were subsistence farmers, that the Majang community dominated, and that they were attached to a life reliant on the forest. To compute the respondent households from each village based on the proportion share of the total households, the 2022 estimated population (households) of each villages was used. The 10 Villages’ total population and households totaled 15826 and 3557, respectively. Finally, respondents were chosen at random using Cochran’s (1977) processes for large populations using the probability proportional to size technique. no = z 2 pq e 2 (1)

Z ² is the normal curve’s abscissa, which eliminates a region at the tails (1-α equals the appropriate confidence level), where no is the sample size. e is the desired level of precision, p (0.6) is the estimated proportion of an attribute or all forms of food-insecure households that are present in the zone’s population, and q is 1-p, as has been highlighted in the reports of Hailemariam (2011) and DRMFSS (2015). The 95% confidence interval is assumed for this research, and the Z table value equals 1.96. 369 households make up the sample size according to the formula.

The final sample size is calculated using Cochran’s (1977) formula for the sample size adjustment for sample sizes greater than 5% of the population. n 1 = no 1 + no / N (2)

Since the sample size is greater than 5% of the population, the ultimate sample size is n1, and N is the size of the population. The ultimate sample size is therefore [369/1+ (369/3557)] = 334. Data from a few families was omitted due to the incompleteness of some household data, and 320 sample households served as the sampling unit for the final analysis.

Prior to conducting the data collection work an official letter written by Addis Ababa University, the research hosting institution, was submitted to bureaus and offices, and briefings were held to create understanding on the overall objective of the study. A cross-sectional household data from primary and secondary sources was used in the study. The data was collected using questionnaires, FGDs, KIIs, and desk review from secondary sources. All methods adopted in the data collection were given consent of approval from institutional review board (IRB) of college of development studies, Addis Ababa University.

There are both closed-ended and open-ended questions in the survey data collection tool, with the closed-ended questions having ratings on ordinal, ratio, or categorical scales. The open-ended questions were designed to elicit more information for some of the closed-ended and description-needed questions. The questionnaire types are those developed by the authors and an institutionally approved one, and standardized questionnaires of the Food Consumption Score (FCS) and the Household Food Insecurity Access Scale (HFIAS). The data was sourced from 320 randomly chosen rural household respondents from 10 villages in the two districts. The questionnaires were administered to the household respondents guided by trained enumerators. The data gathered by this method is purely primary type. The data was collected using tablets and mobile apparatuses in which the questions in the questionnaires are filled into KoboToolbox data collection application.

The goal of the focus group discussion sessions was to gather primary data, which included descriptions of the top priority questions derived from the survey, assessments carried out in the research field, and literature consultation. The in-person discussion was framed with a comprehensive set of terms of reference questions that served as lead questions. Before starting the debate, the authors briefed the discussants about their goal of minimizing bias and establishing a shared understanding. Together with four hired data enumerators, the corresponding author—who is presently pursuing his PhD—collected the data. The corresponding author has worked as a postgraduate research advisor and has experience conducting and publishing research publications. The four male data collectors have all earned BSc degrees and have been employed by Ethiopia’s Central Statistics Authority (CSA) as contractual data collectors. A group of seven to ten individuals of both sexes was selected from the previously sampled household respondents, elderly persons, women, and youth groups at the village level for each of the ten FGDs that were held, one in each of the ten villages.

Similar to the other data-gathering methods, a clear understanding of the aim and merits of the research was made before interviews were conducted. To gather primary information relevant to the specified key explanatory questions, KIIs were done by interviewing people with varying levels of authority and responsibility, such as experts, officials, extension agents, and NGO staff. The primary data sources were specialists from the bureau and offices of development agents at the village level, agriculture, health, water, education, cooperatives, environment and climate, and disaster risk management. Within each purposively selected office, there were two individuals: a lead specialist and the office head as interviewees. The research’s corresponding author carried out the KIIs. This session has also included the implementation of the necessary procedural fulfillments and descriptions discussed in the FGDs.

The secondary data was gathered from the internet and desk review of paper works. Information on related literature reviews was gathered from published journal articles, books, chapters, reports of accredited sources, and from databases of recognized portals. Office paper works of organized plans, accomplished tasks and assessment reports, published bulletins, and brushers were used as secondary data sources of both numeric and narrative types. Access to documented data was requested by an official letter. The secondary data were collected to supplement and justify the data from the primary sources.

This study was granted ethical approval consent from the Institutional Review Board (IRB) of College of Development Studies (CoDS) of Addis Ababa University on 24/08/2023 and with Reference Number of spe/e/c/28/07/2023. The IRB letter is accessible at https://doi.org/10.20372/aau_rdm/EAOGKA ( Zerihun et al., 2023). Further detail on the letter can be requested from the IRB board via the email: cods.irb@aau.edu.et.

Informed consent was obtained from all subjects involved in the study. The household respondents and focus group participants were told the objectives and purpose of the study, that participation was voluntary, and were requested to all the questions. They were informed they were free to refuse to respond to any questions and withdraw at any time. The key informants were also briefed about the objective of the research and its merits towards contributing to the community, policy makers, researchers, and assured of nonexistence of any political or legal concerns. These participants were provided with mobile numbers and emails of immediate supervisors of the research lest they communicate for any inconveniency they encounter during the interview. Official letters of cooperation have been written to the concerned bureaus and offices to inform them of the objectives and intentions of the research.

In this study, the resilience of households to food insecurity was not measured as a capacity but rather as an indicator of food security, such that higher resilience scores are assumed to be indicative of better food security status. Both quantitative and qualitative analytical methods were used based on the characteristics of the variables investigated. Descriptive and inferential statistics were deployed using STATA software version 13 ( https://www.stata.com ). The descriptive results project percentages, means, and standard deviations, and the inferential statistics employ an econometric method applying the specified model.

The data collected using the survey method was first cured in Microsoft excel and then coded after importing into STATA software. To check the data normality, homogeneity and multicollinearity tests were conducted. Next, the data were subjected to their specific models for analysis and interpretation. Likewise, the qualitative data were analyzed such that the information gathered via FDGs and KIIs were transcribed into a written format, major themes were identified and the data were organized. Finally, the organized data were interpreted contextually to validate and substantiate the results from the quantitative analysis. The secondary data types were read, information extracted, and integrated into the study as supplementary information and reviews to justify the research objectives, methods, results.

The results of the descriptive statistics were utilized to determine the degree to which determinant components’ resilience affected the ability of families to withstand food insecurity as well as to shed light on various socioeconomic traits of the households. The drivers of resilience’s impact on food insecurity and their causal connections were examined using econometric analyses. Following the models described along the principal components, an ordinary least squares (OLS) regression model was used for the econometric study to look into the correlates of resilience status. The following mathematical illustration ( Equation 3) provides an estimation of the expected value of the dependent variable Y. Yij = β 0 + Σj = 1 . . p βjXij + Σi (3)

Where: Yij = Dependent variable “Resilience to food insecurity”

β0 = Constant,

β1, β2 … βi = Coefficients of variables,

X1, X2 … Xi = explanatory variables

εi = Error term

Fitting the resilience components into the model, the working equation takes the following form. RI = α + β 1 IFA + β 2 AC + β 3 ATA + β 4 AP + β 5 ABS + β 6 AC + β 7 S + ε i (4)

Where RI is household resilience; a is a constant; b1–7 are the coefficients of each variable (the component indices developed during the PCA analysis); and e is an error term.

In many food security-centered resilience analysis frameworks, like those developed by Alinovi et al. (2010a), the quantitative analysis of resilience identified seven or eight key dimensions of resilience that integrated capital and capacity approaches, including income and food access, access to essential services, assets, adaptive capacity, stability, adoption of agricultural technology, social capital, and/or social safety nets. To account for the study’s environment, the components that determined resilience capacity were stated as recommended by FAO (2016a) and ( Alinovi et al., 2010a). PSNP is substituted by its anonymous counterpart, Social Capital (SC), as there are no rural social safety net programs deployed in the zone and only urban safety net is just starting in the zone’s capital, Metti town. Guyu and Muluneh (2015) substituted the traditional SSNs that are inherent and used by the community in the study area for the SSNs as a way to help build social capital in places where the formal program is not implemented. The inclusion of observable factors under each component was made based on refined evidence from the literature and the researchers’ prior experience of the study area. The PCA creates uncorrelated indices or components from an initial set of n correlated variables (x1, x2, x3, …, xn), and each component is a linearly weighted combination of the initial variables. The mathematical equation takes the following form ( Equation 5): Yi = ai 1 x 1 + ai 2 x 2 + ai 3 x 3 … ainxn ;           i = 1 ; 2 ; … ; n ( 5)

Where Y _i is the household’s component score on the i ^th PCA, a _in represents the weighted value for the i ^th principal component and the n ^th variable.

Since the pillars are latent and cannot be directly quantified to indicate resilience at the household level, the resilience status of the households to food insecurity was examined using a two-stage analytical model. Each of the pillars can also be measured using socioeconomic and institutional characteristics that have been observed. Multivariate techniques are frequently used in research to quantify outcomes, and cross-sectional data from national demographic and household surveys, as well as individually designed and self-administered surveys, were employed for measuring the level of household resilience to food security ( Ansah et al., 2019; Beyene, 2016; Alinovi et al., 2010a). First, principal component analysis is used to estimate an index for each component individually across a collection of observable variables. The resilience index is then derived using a factor analysis on the interacting components estimated in the first stage, where the weights are the percentages of variance explained by each factor and the resilience index is a weighted sum of the factors created using Bartlett’s scoring method. According to Beyene (2016), who cited DiStefano et al. (2009), the Bartlett technique typically yields latent variable ratings that are univocal and unbiased.

To determine whether principal component analysis (PCA) is appropriate, model fitness and sample adequacy tests are computed. A KMO value of 0.5 in a decent model is required to pass the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy test. Bartlett’s test of sphericity was run concurrently to demonstrate the lack of multicollinearity or singularity issues. By examining the determinant of the R-matrix, which should be larger than 0.00001 as stated in Field (2009), multicollinearity or singularity concerns can be identified. Since it is analytically impossible to use these variables for factor analysis, or PCA, the best scaling technique was used to estimate the latent components that are related to the dummy or categorical variables by changing the observed variables. All variables in transformation are assumed to have a normal distribution with a mean of zero and a unit variance. Each observed variable is multiplied by a negative one to make it compatible with the meaning of the hidden variable when some of the seen variables are negative indicators of the latent variable.

Varimax, or orthogonal factor rotation, is used to separate the loadings amongst variables after PCA factor extraction. Varimax rotation is regarded as a good general strategy that makes factor interpretation simpler since it aims to maximize the dispersion of loadings within factors by heavily loading fewer variables onto each factor, leading to more interpretable clusters of factors. Field (2009) claimed that loadings with an absolute value of more than 0.3 were generally regarded as being important. However, he recommended that significant consideration be given to taking critical values of factor loadings of 0.722 for a sample size of 50 and that these values should be greater than 0.512, 0.364, 0.298, 0.21, and 0.162 for sample sizes of 100, 200, 300, 600, and 1000, respectively. The following equation represents the development of the household resilience index. Ri = ∑ WC Xi − Xj / Si ( 6)

Where Ri = estimated resilience index for household i, WC = the weight (factor score coefficient) for the X ^th indicator of resilience in the PCA, Xi = the i ^th household’s value for the X ^th variable (indicator) (Note: the X ^th’s in the current study’s context are the variables denoted by AP, ATA, ABS, S, SC, AC and IFA in equation 1 above) Xj = the mean of X ^th variable for overall households. Si = standard deviation of the X ^th variable for overall households.

The households combined index for each latent component (covariate of the overall resilience) is computed according to equation 5 below, as has been applied in a variety of research works ( Tekalign, 2022; Dhraief et al., 2019; Debessa, 2018; Beyene, 2016; Guyu & Muluneh, 2015). RIn = ∂ 1 Factor 1 + ∂ 2 Factor 2 + ∂ 3 Factor 3 + … + ∂n Factor n ( 7)

Where RIn is the resilience index of the nth household, ∂n is the variance explained by each factor, factors 1, 2, 3 … n are the respective factors generated by the factor analysis representing each latent dimension.

The next step is to assess each household’s overall resilience to food insecurity status after calculating the index for each latent variable for that particular household. The factor technique and principal component analysis are used to obtain the overall resilience index (RI) for each family. As a result, the households were divided into the four resilience categories that had been arbitrarily proposed ( Beyene et al., 2023). These groups were said non-resilient if the overall resilience index is (RI<0.00), moderately resilient (0.00≤RI<0.5), resilient (0.5≤RI <1.0), and highly resilient (1.0<RI). The categories for resilience were coded 1 through 4 in increasing order, with 1 being designated as susceptible and 4 being highly resilient.

A total of 320 rural households participated in this study, and the majority (90%) depend on agriculture to meet their dietary needs. The majority of households (79%) have a male head of family and are married (88%). 89.7% of households have agriculture as their sole source of income, 9.7% have agriculture plus trade, and 0.6% have income from other sources. The average household size is five family members, which is in line with the national average. The average age of the household is roughly 40 years, with the minimum and maximum ages being 18 and 75 years, respectively. The households had a mean total land holding of 2.63 ha, which is three times greater than the national and regional averages stated by CSA (2021) but less than 0.5 ha when the net land use is restricted to the production of food crops. According to calculations, the households’ average annual farm and non-farm earnings were 20273 and 495 birr, respectively. When expressed as a mean per capita yearly income, the figures for on- and off-farm incomes are 4054 and 100, respectively. In the research area, the average household dependency ratio was found to be greater (87.65%), with the highest dependency ratio approaching 400%.

According to the results of the descriptive statistics, 42.19% of the respondents had at least completed elementary school, while 57.81% had not attended any formal schooling. Additionally, almost three-quarters of the respondents stated that they were not receiving the services and technologies they needed to promote the expansion and productivity of their subsistence agriculture. Only 39% of them receive agricultural extension service contacts as a result, although FTCs are built in their Villages (60.62%), improved seeds are provided to them (25.62%), traditional irrigation systems are used to supplement crop production (12.81%), they are members of cooperatives and participate in cooperative activities (16.87%), they can obtain veterinary services for their livestock (20%), they can prepare and apply fertilizer (13.44%), they can use oxen to plow their agricultural land, and they are able to purchase and apply chemical pesticides (32.50%).

The analysis outputs of the observable variables explained their respective latent components at varying magnitudes and significance levels. The resilience dimensions used to estimate the resilience status of households include income and access to food, access to basic services, assets, adaptive capacity, stability, agricultural technology adoption, and social capital. The number and expected contribution of independent variables towards resilience to food insecurity under each pillar, the components retained, and the indices produced accordingly are portrayed below.

The income and food access (IFA) pillar is closely related to households’ ability to absorb shocks with what they have on hand and other food access options, which is why it has a positive link with resilience to food insecure circumstances. Income limits a household’s access to food by determining how much it can spend on food and how much food it can consume. The Household Food Insecurity Access Score (HFIAS) and the Food Consumption Score (FCS) of the households, both of which are proxy tools for measuring food access, were used to construct the index for IFA along with the household’s farm income, off-farm income, number of income sources, and amount of credit received. Except for the HFIAS, all the variables show a positive correlation because those scores rise when food security increases.

The findings of the main component analysis demonstrate that the variables being investigated are sufficient to explain the IFA because the model adequacy test, measured by the KMO measure of sampling adequacy (0.615), is significantly higher than the cutoff established by Kaiser’s rule (0.50). Chi2 = 538.113, df = 6, and p<0.01 values for Bartlett’s Test of Sphericity showed that there was no singularity or multicollinearity among the variables. The correlation results show a perfect connection between the converted data and the IFA latent variable, with an R-matrix value of 0.183. All six continuous-scale factors were loaded onto a single component with an eigenvalue of 2.31 and a total variance of 57.88%. Except for off-farm income, which loaded below the minimum and was eliminated from the analysis following the orthogonal rotation matrix factor loading retention requirement, all five variables sufficiently scored loadings above the minimum. The component loadings of the variables utilized in the model are shown in Table 1. Since each of the three variables is measured on a continuous scale, PCA is applied directly to them.

IFA = 0.5788 ∗ Factor 1 ( 8)

Assets are the capital that enables households to withstand challenges and shocks. When a household has varied assets on hand, they feel safe and protected from unforeseen events that may otherwise throw their life off balance. The two most important assets for rural households are land and livestock. Both productive and nonproductive assets act as buffer stocks under difficult circumstances as a short- and long-term way to avoid shocks and be robust. The household’s jewelry, telephone, television, plow, bike, table, and radio were all considered assets in some areas, although the definition of an asset depends on the context ( Boukary et al., 2016). Estimates of the agricultural asset components of resilience were based on land holding size, cattle ownership (TLU), beehive ownership, and possession of jewelry, radios, jewelry, televisions, satellite dishes, mobile phones, tables, sofas, beds with sponge mattresses, bicycles, and motorbikes. TLU was transformed using the conversion equivalents table for Sub-Saharan Africa published by Njuki et al. (2011). Presumably, each factor will increase the likelihood of resilience to food insecurity.

Before doing the PCA, the dummy variables underwent optimal scaling adjustment to be suitable for PCA. The variables’ suitability and model fitness for the principal component analysis were assessed after analysis, and the test result demonstrates that the variables are eligible with a KMO test value of 0.79. With χ ² = 916.335 and df = 55, the factor test was also significant (p<0.01), and the correlation matrix value of 0.054 showed that there were no collinearity issues among the variables. Using Kaiser’s formula to estimate the latent variable AP, four components were kept that together account for 64.34% of the overall variability. Five, two, three, and two-component loadings of more than 0.3 values, respectively, were applied to components 1, 2, 3, and 4 ( Table 2).

AP = 0.2959 ∗ Factor 1 + 0.1269 ∗ Factor 2 + 0.1135 ∗ Factor 3 + 0.1072 ∗ Factor 4 / 4 ( 9)

The third crucial component of resilience, adaptive capacity, describes how well a household can adjust to and respond to shocks. According to Dhraief et al. (2019), the phrase refers to the circumstances that make it possible for individuals to foresee and react to changes, minimize, cope with, and recover from the effects of change, as well as seize new possibilities. The latent pillar AC was established using the indicators of educational average, diverse sources of income, and household dependency ratio ( Baye, 2017). The adaptability or coping capability of households experiencing food insecurity difficulties is also influenced by the age, size, and marital status of the household. These observable factors, such as sex, age, marital status, family size, education, diversity of income sources, and the dependency ratio of households, were used to explain this pillar. Except for the dependency ratio, all of the variables used to index AC were thought to positively affect resilience to food insecurity. The marital status of household variables and the sex of the head of the household were transformed using optimal scaling statistics. The sample variables used to determine the AC pillar were found to be sufficient (KMO = 0.616), and the correlation matrix’s determinant of 0.096 verifies the model’s fitness in all necessary conditions for PCA to be performed without any concerns about multicollinearity or singularity. A Chi square value of 740.363 and a degree of freedom of 21 led to a significant (p<0.01) result for the Bartlett test of sphericity.

The results of the principal component analysis showed that every variable had a significant absolute value greater than 0.3, which indicated that it was important in explaining the AC pillar of resilience. Retained were two components with eigenvalues of 2.25 and 2.11, which together accounted for 32.31% and 30.12% of the overall variance (62.40%). The first component is heavily weighted with the factors of household size, age, and dependence ratio, while the second component is heavily weighted with the variables of household sex, marriage status, and years of education. The dependency ratio’s inverse relationship with adaptive capacity, which results in a household’s adaptive capacity eroding as a result of the considerably more non-working yet consuming household members, is what accounts for the dependency ratio’s negative loading value on component one ( Table 3).

The equation estimating the AC latent variable is: AC = 0.3231 ∗ Factor 1 + 0.3012 ∗ Factor 2 / 2 ( 10)

Infrastructure and institutions that offer services are essential to improving the resilience of rural households. Public service delivery is typically capital-intensive and exogenous to households’ dependence on governmental and non-governmental supporting organizations. To increase the effectiveness of a household’s access to assets, access to public services is essential ( Alinovi et al., 2010b). Infrastructures like roads, markets, water points, irrigation systems, and other services like farmers’ development agents, electricity, educational facilities, telecommunications, health facilities, credit institutions, and transportation all play crucial roles in supporting the effort to remain resilient. The distance to a source of drinkable water, the distance to the nearest highway, the distance to a school, and the distance to a medical facility are all used to calculate the latent variable ABS. In this analysis, accessibility and distance to public services are negatively correlated; thus, each has been multiplied in reverse to change the direction of influence. In addition to the variables, access to telephone services, electric service, and rural credit services were measured on a categorical scale and needed to be transformed into a ratio scale using the best possible scaling technique.

Following the execution of all statistical tests for PCA, it was discovered that the model is suitable and competent for significantly expressing ABS. The samples were sufficient to support factor analysis, as demonstrated by both the individual and aggregate (0.786) KMO tests. The R-matrix value of 0.058 is over the stipulated minimum cutoff level (0.00001), suggesting the absence of multicollinearity or singularity problems among variables. The factories or Bartlett’s Test of Sphericity result was significant (χ ² =898.787, df = 28, and p<0.01). Ten explanatory variables were used to determine ABS, but two of them—distance to health facilities and distance to a primary school—were removed due to their minimal loading or inadequate ability to explain the latent variable. The PCA identified three components that explained variances of 37.26%, 19.36%, and 14.56%, respectively, with eigenvalues of 2.981, 1.549, and 1.165 in ascending order. The total variance was explained by four factors loaded on Component 1, two loaded on Component 2, and two loaded on Component 3 ( Table 4). ABS = 0.3726 ∗ Factor 1 + 0.1936 ∗ Factor 2 + 0.1456 ∗ Factor 3 / 3 ( 11)

In least-developed countries, practically all rural households depend heavily on agriculture. Given that the agricultural sector is supported by enhanced agricultural inputs and technology that can increase crop and livestock production and productivity, food security, and therefore resilience to shocks and stresses, To measure the ATA latent component, sixteen variables were taken into account, including methods and tools for increasing agricultural and livestock output. Crop rotation, intercropping, improved seed, chemical pesticides, synthetic fertilizer, organic fertilizer application, oxen plowing, constructed crop storehouse or gotera, coffee drying on the bed, mulch application, irrigation, use of modern beehives, veterinary service, artificial insemination, and keeping livestock in separate homes are all examples of crop diversification practices. All the variables were transformed using optimal scaling approaches and run in PCA because they were all dummy variables. Four variables—use of synthetic fertilizer, usage of contemporary beehives, crop rotation, and coffee drying bed—were omitted from the analysis because they had singularity problems for the first two, to which the response was “no,” and because they had low factor loading for the latter two. The utilization of modern beehives and the total absence of chemical fertilizer use by the studied families gave rise to the singularity case.

Therefore, the PCA was performed using 12 indicator variables ( Table 5) to forecast how the ATA will affect households’ levels of resilience to food insecurity. To test the variables for variable correlation and collinearity and to make them suitable for PCA, an optimal scaling procedure was applied to them. The results that were subsequently produced showed that the variables are a linear combination of predicted factors to quantify the pillar for resilience measurement. The KMO statistics score of 0.843 indicates that the sample is adequate, and at the same time, Bartlett’s test rejects the null hypothesis that the original correlation matrix is not an identity matrix (p< 0.01, Ch2 = 1306.477, and df = 66). The correlation matrix’s determinant was found to be significantly over the minimum significance threshold of (0.016) in the test for the presence of multicollinearity or singularity. Variables with score loadings of 0.3 and components with eigenvalues larger than or equal to one were allowed to be retained by the factor analysis. Overall, three components (the first loaded with five variables, the second and third with four variables each) with eigenvalues of 2.946, 2.173, and 1.995 each explain variances of 24.55%, 18.11%, and 16.63%, respectively ( Table 5).

ATA = 0.2455 ∗ Factor 1 + 0.1811 ∗ Factor 2 + 0.1663 ∗ Factor 3 / 3 ( 12)

The sixth element of resilience is stability, which typically refers to a household’s options and ability to resist various socioeconomic and ecological circumstances that could impair a household’s ability to provide for its members. Since they are linked to shock and stress, the factors that are utilized to estimate this pillar typically have adverse effects. Most similar studies use variables like crop shocks, livestock shocks, human health shocks, climate-induced shocks, economic shocks, and so on to denote pillar S, showing how it has an inverse relationship to food insecurity resilience. Each observed variable for the relevant scenario was multiplied by -1 to make it comply with the stability pillar’s definition. Theoretically, stable households are thought to be more resilient to food insecurity. To calculate S, this study takes into account human health issues like serious illness, frequent medical visits, and having a family member with a disability; pre- and post-harvest crop shocks that are defined by disease and insect damage, weeds, wind and frost attack, and flood; climatic shocks like rainfall pattern or variability; socio-economic variations like frequency of DAs’ visits and support, commodity price variability, and latrine use; and lives. One of these observable variables, shown in Table 6, was employed in the study; the others were left out due to concerns about the variables’ multicollinearity and sample adequacy for PCA.

Since each of the five variables has a ratio scale, PCA is possible. The outcome of the factor analysis demonstrated that the variables are sufficient and that multicollinearity or singularity are not issues. At χ ² = 97.712, df = 10, the Kaiser-Meyer-Olkin Measure of Sampling Adequacy was found to be adequate (0.63), and the Bartlett test of sphericity was significant (p<0.001). The model is appropriate, as indicated by the correlation matrix’s (R-matrix) determinant of 0.734. According to Kaiser’s criterion, two components were kept, with their respective eigenvalues of 1.56 and 1.21 explaining variances of 31.15% and 24.13% for the two PCA measurement indicators. In addition, two variables associated with agricultural output were placed on the second factor, whereas the three variables related to human health were grouped under the first component and shared a significantly higher variance. Additionally, according to the component loadings ( Table 6), animal shocks are the most important factor in enhancing the study households’ ability to withstand food insecurity.

S = 0.3115 ∗ Factor 1 + 0.2413 ∗ Factor 2 / 2 ( 13)

Social capital networks are intra- or inter-social group ties that are based on a sense of community or belonging and actively support the resilience of both individuals and groups. They are governed by conventions, ideologies, and institutions developed locally to further common goals and interests. In communities, social capital is crucial because it offers support and backups that increase a household’s or a group’s resilience, especially in times of stress and shock. In more advanced situations, these platforms can be turned into chances for business ownership and wage employment, as well as official and informal loans, cash advances, inputs on credit, talents, and shared resources for production and marketing. Idir, members of religious organizations, Iqub, Debo, and people’s readiness to help are a few of the most popular ones.

The willingness of people to help one another out in difficult circumstances, Idir membership, and religious group participation were used to represent the SC component for this study ( Table 7). The variables underwent an optimum scaling adjustment before PCA because they are categorical in nature. The samples were found to be suitable in the subsequent PCA analysis, scoring a KMO value of 0.664, and the model was found to be appropriate and fit (R-matrix = 0.201) according to the significant Bartlett test of sphericity (p<0.01, χ ² = 506.984, df = 15). Since all six variables have component loading values greater than the criterion of 0.298, they are all statistically significant. The two components that were kept explained 63.33% of the total variance, with eigenvalues of 2.22 and 1.58.

SC = 0.3698 ∗ Factor 1 + 0.2634 ∗ Factor 2 / 2 ( 14)

The latent component indices from the principal component analysis of the latent components computed earlier in stage one are used to create the overall resilience index (RI) of the households. The indices of the seven pillars, ATA, IFA, ABS, AP, AC, SC, and S, were subjected to PCA at stage two of the resilience analysis under the assumption that all the blocs are normally distributed with mean zero and unit variance. At the second stage of the analysis, factor analysis using the principal component factor approach may have been used in several research articles ( Eze et al., 2021) that focused on resilience-centered topics. These variables’ suitability and appropriateness for the model were examined, and all necessary measures were determined to be met. Accordingly, the KMO measure of 0.859 indicated that the samples were enough, and the value of the correlation matrix’s determinant, which is 0.045, indicates that multicollinearity or singularity concerns are not present, becoming significant (p< 0.01) with a Chi square value of 977.339 and a degree of freedom of 21.

According to Kaiser’s criteria, one component explaining 53.07% of the total variance and having an eigenvalue greater than one (3.715) is kept. All variables had acceptable factor loadings except the variables measuring adaptability and social capital. The largest factor loadings were found for income and food access, followed by the adoption of agricultural technology, asset ownership, access to basic public services, and stability in that order, showing the relative contributions made by each block in helping households in the study area become resilient to food insecurity. The stability component, on the other hand, is loaded with a negative value ( Table 8). Observed variables are, in fact, an indicator of instability because of the nature of the variables used to evaluate stability, which requires that there be a negative association with resilience to food insecurity.

Hence the overall resilience index is generated as in equation 15 below: RI = 0.5307 ∗ Factor 1 ( 15)

Based on two analytical outputs, the determination of families’ food insecurity resilience statuses was carried out. In the first, the total resilience index of households was used to identify the categories or resilience status of households. The importance of the pillars in determining the household resilience capacity was covered in the second analysis, which was econometric.

Table 9 below lists the groups of households in the research area that are resilient to food insecurity. The findings showed that the population’s level of resilience is only moderately stable, with roughly 47.5% of families vulnerable and an additional 12.19% uncertain, bringing the percentage of weak resilience to almost 60%. The findings are in line with studies of nearly identical cont, respectively, with 23.13% and 17.19% each ( Table 9).

Additionally, as in the works of Debessa (2018), Ciani & Romano (2014), Alinovi et al. (2010a) and Alinovi et al. (2010b), the relative importance of the covariates in indicating the resilience capacity of households was done via the size of loadings of the components (pillars) as an alternative path to detect the substantive importance of each. According to Field (2009), the factor loadings in a particular analysis can take the form of either correlation coefficients or regression coefficients, providing two different avenues for analytical comparison. As a result, the factor loadings of the pillar components used to explain the combined resilience index ( Figure 2) showed diverse magnitudes and directions of effect, suggesting that one contributes considerably to building resilience but not significantly more so than the other. In light of this, ABS, AC, and SC played relatively few roles despite having a positive correlation, whereas IFA, ATA, and AP significantly and positively contributed to families’ resilience ability. In terms of relative relevance, it was determined that the IFA pillar contributed more than the ATA and AP pillars. The study area is heavily dominated by commercial crop production, such as coffee and apiary operations, and these considerably contribute to the well-being of the households. As a result, the income and food availability dimensions were shown to be crucial in the resilience-building process. The stability pillar of the resilience component showed a negative correlation with the outcome index, indicating that the research area had seen a high occurrence of food security shocks. Animal and crop-related biotic and abiotic problems account for the majority of the shocks, in particular. Both the key informants and the focus group participants saw how large crop shocks such as heavy rain, snow, strong wind, insect pest attacks, weeds, and illnesses have limited the amount of potential food that could have been produced from crops. Notable are also the shocks that have been identified as contributing to the decreased productivity of livestock output, including diseases such as trypanosomiasis, foot-and-mouth disease, and a lack of fodder and veterinary services. During discussion sessions the households claimed that they have to travel long distances to purchase medicines and medical equipment for their livestock without prescriptions but these aided their common understandings.

Identification of the relative significance of each pillar in establishing the resilience category for the families is equally important to understanding the status of the resilience of the households. The overall index was used as the dependent variable in the Ordinary Least Squares (OLS) regression analysis, while the indices of each pillar were used as the independent variables. To determine how much these pillars affected households’ resilience to food insecurity in the research area, they were regressed against a predetermined cutoff value of the overall resilience index (RCI). The outcomes of the OLS regression study are discussed below.

The OLS regression analysis result showed that all seven factors, except social capital, had considerably and highly affected the resilience status of households to food insecurity ( Table 10). The results of the econometric analysis, in contrast to the component loading-based conclusions, showed that adoption of agricultural technologies and asset ownership came in second and third place in terms of relevance. The findings show that, with all other variables maintained constant, a unit increase in agricultural technology availability increases households’ ability to withstand food insecurity by more than double (2.13 units), on average. The capacity for resilience will also improve by 1.697 units for every additional unit of assets owned by households. Similarly, a better intervention to foster access to services, building adaptive capacity and income and food access dimensions positively contributes to lifting of the resilience capacity by 0.648, 0.363 and 0.158 respectively. The social capital dimension remained insignificant, according to the group discussions result, social capital variables such as idir, equb, debo, and dado that could have positive contribution in building resilience have been little exercised particularly in the Majang community.