Predictors and regional prevalence of food insecurity in Ethiopia during COVID-19: a multilevel analysis

Abstract

Introduction

Food insecurity is one of the most serious issues, especially in developing countries, that harm many public health outcomes through increased under nutrition, mental health problem, and premature mortality. It is widespread socio-economic problem of Ethiopia, with unequal distribution among its regions, during COVID-19 and other shock event manifestations for the last three years. This study aimed to analyse country-wise and region-specific food insecurity prevalence; assess its variation among regions; and identify predictors that influenced households’ food insecurity in Ethiopia during COVID-19.

Methods

This study used longitudinal data from the World Bank's Ethiopia-High Frequency Phone Survey, which looked at 3,300 households' experiences of food insecurity over five rounds, yielding 13,517 observations throughout time. The non-parametric model, Kruskal–Wallis Test, was used to asses food insecurity differences across regions; while the parametric, Generalized Multilevel Binomial Regression Model, was used to identify significant predictors of households’ food insecurity experience.

Results

There are significant variations in food insecurity among regions of Ethiopia during COVID-19. Sumali was the region with highest food insecurity prevalence followed by Tigray, SNNP, Oromia, and Amhara where these regions were also facing another shocks, in addition to COVID-19, such a displacement and drought. Female-headed household and income loss are directly associated with likelihood of being food insecure. Dwelling in urban (coefficient = -0.3707, p = 0.0003), being employed (coefficient = -0.1869, p = 0.0161), benefiting assistance (coefficient = -0.3504, p = 0.0029), and operating non-farm business during COVID-19 (coefficient = -0.4074, p = 0.0000) were significant and negatively associated predictors of households’ food insecurity. Besides, household’s worry and financial threat due to the outbreak of pandemic were the two COVID-19 related predictors that had significant effect on household’s food insecurity. Income loss was the most determinant variable (coefficient = 0.8562, p = 0.0000) that had largest influence on household’s likelihood of being food insecure. As time went, the decline in food insecurity was attributed to either decreased outbreak of the pandemic and/or improved households’ resilience to shocks.

Conclusions

Even while food insecurity is a major issue in Ethiopia, not all its regions are at equal status. Household’s food insecurity is determined by his ability to handle the problem economically, and withstand shock events like COVID-19 that subtly disrupts social and economic networks. Intervention measures taken to insure food insecurity in the country should take in to account regions’ food insecurity inequalities and their vulnerability to shock event manifestations. During shocks, boosting households’ ability to cope up with unexpected risk event can save the exacerbation of food insecurity problem.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-22234-0.

Background

Food insecurity problem has continued as a widespread social challenge that affects more than 2.4 billion people in the world. The updated and improved estimates show that more than one-third of people in the world – about 2.8 billion – could not afford a healthy diet in 2022. Inequalities are evident with low-income countries having the largest percentage of population that are unable to afford a healthy diet (71.5%) as compared with lower-middle-income countries (52.6%), upper-middle-income countries (21.5%), and high-income countries (6.3%) [1]. In spite of the fact that significant efforts have been made for the last two decades, both food insecurity and undernourishment have been increasing in different part of the world; particularly almost all nations of Africa with 20.4% (298.4 million) people facing hunger in 2023. Food insecurity is more prevalent in developing countries due to rapid population growth that is not matched with sufficient food production and stagnating agricultural productivity [1, 2].

It is a crucial and, socio-economic and public health problem in Ethiopia that upsets its population with varying degree and impact among its nations. According to [3, 4], although Ethiopia has been making continuous efforts over the past two decades to reduce overall food insecurity and poverty, the problem of food insecurity scenario in Ethiopia worsened significantly. In addition to pre-existing food insecurity prevalence, the outbreak of Covid-19, external Russia-Ukraine war, internal conflicts with in the country, and factors related to unpredictable climate variability exacerbated the problem of food insecurity for the last three years. Further, Ethiopia is landlocked country that imports machinery, vehicles, cereals, fertilizers, refined petroleum, pharmaceuticals, and other goods and commodities from abroad to meet its domestic needs. This made the problem of food insecurity in the country more serious due to lockdowns during the outbreak of the pandemic. The country ranks 101st among 125 countries with a score of 26.2 in 2023 indicating the serious level of hunger [5]. According to the most recent assessment of food security needs in Ethiopia, 15.8 million people require food support in 2024 with 4 million internally displaced people due to 2020–2022 conflict in the north, and severe drought in the south and southeast; and 7.2 million having high levels of acute food insecurity that need emergency assistance [6].

However, during Covid-19 and other shock event manifestations with rising number of food insecure population in the country, different regions of the country experienced food insecurity differently. This difference among regions could be due to variations in exposures to chock events like Covid-19, displacements due to within conflicts, drought due to unpredictable weather condition, and other natural differences such as ethnicity and language, culture, customs, and other unknown factors. On the other hand, if there are disparities in food insecurity experience across regions, predictions made on prevalence and experience of food insecurity at country level does not give real picture for making valid inference at regional level. It needs to identify regions that are likely to have chronic food insecurity in order to take measures in those regions that require special interventions [7].

Many previous studies conducted on food insecurity situation in Ethiopia taking in to account it as a part of the major socio-economic challenge of the country found economic, social, health, and shock as the factors for the influence of households’ food insecurity. The study [8] showed that households’ food insecurity would be worse when families have less disposable income and Summer Nutrition programs are less widely available where this factors vary from state to state indirectly affecting the households dwelling in the state. The research on households’ food insecurity risk conducted in Europe evidenced that the socio-economic and demographic risk factors were more prevalent among economically disadvantaged groups [9].

Besides, the assessment made on interplay of demographic considerations and food security in Nigeria has revealed that population trends affects the demand for food; which in turn affects the supply-side solution reducing the food deficit at household’s level. This study recommended that demographic projections need to be incorporated into plans to improve agricultural production and achieve greater food security at household level [10].

The study [11] in Ethiopia revealed that urban dwellers are relatively better with regard to food security as compared to the rural dwellers. This study evidenced that educational attainment of the household head, proximity to service centres and wealth positively affect to household food security; whereas dependency ratio and shock increase the household food insecurity. Similar result was found [12] in southern Ethiopia with participating urban agriculture, health status of households, average income, socio cultural practices, and asset ownership significant predictors of urban household food insecurity. According to the finding [13] and [14] indicated, seasonal variations were observed to affect food insecurity of households where March (post-harvesting) to June were the times during which many households are more food insecure.

The study [15] has forwarded the importance of considering spatial variations during food security assessment. This study has proposed the comprehensive policy response to food security that fully addresses spatio-temporal variation components is likely to contribute to improving food security prediction. Similar to this study, the study carried out in East Gojjam zone, Ethiopia, “to really mitigate food insecurity, it needs to consider micro level spatial and temporal variation of household food insecurity during planning interventions “ [16]. A spatio-temporal analysis to identify zone specific covariates effect on under nutrition was investigated [17]. This study has found that zone level higher breastfeeding, lower percentage of comorbidity, higher percentage of literacy, and higher percentage of Body Mass Index (BMI) of women were positively associated with higher values of under nutrition. However, a higher percentage of water sanitation facilities, a low percentage of women’s autonomy, a higher percentage of the employment status of women, a higher percentage of wealth index, and higher values of a zone precipitation were positively associated with higher proportions of under nutrition. The recent study [18] on spatiotemporal modelling of household level food insecurity in Ethiopia indicated that the space–time assessment using the measurement corrected for different feeding cultures can make more informative & relevant inferences in designing society-based interventions for the reduction of households’ food insecurity.

The impact of shock event on food insecurity of households largely depends on household’s related factors i.e. their ability to cope with, adapt, and creating shocks resilience [19]. The finding of this study revealed the experience of shocks increase the odds of food insecurity by 1.73; and hence suggested that development strategies should build resilience to future risks in order to alleviate the prevalence of food insecurity. Other study [7] revealed that availability of credit services, proximity to service centres, average years of schooling of the household members, and household assets are negatively associated with household level food insecurity. This study evidenced that shocks, age, and dependency ratio increase the odds of a household to be food insecure. In view of the studies conducted during Covid-19 pandemic in United States of America, the study [20] investigated the effects of household level factors on food insecurity during Copvid-19. This finding asserted that women headed, black headed, Hispanic-headed households, and households that did not have a college education experienced significantly higher level of food insecurity than their counter parts. In Ethiopia, the study [21] inflicted that the pandemic has caused a disproportionate effect on food security across gender of the household heads. According to this study, female-headed households experienced food insecurity significantly higher than male-headed households mainly due to lower education level.

A similar finding from meta-analysis found out that female-headed households are about two times more prone to food insecurity than male-headed households in Ethiopia [22]. In contrast to this finding, the study [23] showed no systematic food insecurity differences between male-headed and female-headed households during the first year of the COVID-19. However, this study forwarded that cross-section (first round data) could be the factor for the insignificance of the association. However, the comparison of food insecurity between rural and urban residents depicted that there was increased food insecurity more in rural areas relative to urban.

In spite of several previous literatures on food insecurity situation in Ethiopia, none of those studies addressed the question of the current study. Some of the studies covered only particular area of the country (region, zone, or lower level); while others used cross-sectional data that were not appropriate to examine the long-time effects of shock events. Besides, some of them did not incorporate food insecurity variations among regions and regional level predictors in their analysis; while others had methodological gaps that could not incorporate the complexities of the food insecurity problem. Some important recent studies such as [18] conducted in Ethiopia assessed households’ food insecurity well using spatio-temporal analysis. However, these studies used longitudinal data, collected before the occurrence of Covid-19, which does not represent the recent food insecurity problem that occurred in Ethiopia during the outbreak. Moreover, these studies suggested future researchers to investigate the recent food insecurity status by considering the future panel study data set together with recent shocks such as the COVID-19 pandemic, internal conflicts, displacement, and other factors confounding effects to get updated understanding of the process.

So, the differences in occurrence of food insecurity among regions of Ethiopia during Covid-19 and their associated household and regional level predictors have not been well investigated by previous studies. Due to this, assessing food insecurity during Covid-19 in Ethiopia taking in to account the regional variations is the major concern that requires immediate attention from all stakeholders; particularly from academics to provide evidence-based information to policy makers to effectively curb the problem in the country. This study work is, therefore, aimed to assess the country-wise and region-specific food insecurity prevalence simultaneously; detect the variations of food insecurity experience among regions; and identify household and regional level potential predictors of food insecurity during shock event manifestations for the targeted interventions.

Methods

Data, sample, and sampling designs

The current study used HFPS-HH 2020–2023 dataset collected by World Bank (WB) as the main investigator and Ethiopian Central Statistics Agency (CSA), currently known as Ethiopian Statistical Services as collaborator; while United States Agency for International Development (USAID), World Bank Group (WBG), and Global Financing Facility (GFF) were involved as funding organizations. The dataset was gathered in 5 African countries—Nigeria, Ethiopia, Uganda, Tanzania, and Malawi to support government and development partners through providing realistic data that can be used to assess socioeconomic shocks such as COVID-19 implications on the households. It is a panel survey data collected monthly in 15 consecutive rounds from May, 2020 to October, 2023.

The study subjects, unit of analysis, of the current study are households that directly experience the actual burden of food insecurity problem of individuals based on Food and Agricultural Organization (FAO) initiative Voices of the Hungry Project [24, 25].

The sampling design and procedure of HFPS-HH 2020–2023 surveys was based on pre-pandemic Living Standards Measurement Study (LSMS) in 2018/2019 that was designed by Ethiopian Socio-economic Survey (ESS) which is built on nationally, regionally, residence, and gender representative sample of households in Ethiopia. A subsample of 3300 households accessible through phone was selected from the sampling frame of 5,374 households that were interviewed in 2018/2019 by ESS. The survey questionnaires, developed by World Bank, were administered to all the households in the sample. Following up 3300 households, data were collected in 15 rounds by using Computer Assisted Techniques (CATI) Software & Survey CTO built-in daily data monitoring techniques. The current study used the data collected in five rounds during which food insecurity experience of households, the response variable, was measured. This included 3,107 households in round 2; 3058 households in round 3; 2770 households in round 5; 2704 households in round 6; and 1982 households in round 11 that completed the required interviews. This yielded the total observation with completed interviews over time to be 13,621 from which 13,517 were measured longitudinally (at least in two rounds) and used directly for the analysis.

Study variables

Food security and its measurements

The problem of food security is multifaceted in that it can occur at global, national, community, and household or individual level. Food security can be defined in the contexts of availability, accessibility to obtain appropriate foods, nutritious, and stability. It would be difficult to assess all the dimensions of it under a specific research work. This study focused on food insecurity at household level during Covid-19 Pandemic and other shock event manifestations in Ethiopia. Thus, the response variable of the current study is food insecurity experience of a household, where the burden of social and socio-economic implications of shocks lies. So, in this study, food security is viewed as the ability of households to physically and economically obtain safe and nutritious food (accessibility domain of food security).

Despite the needs to have single measure of food insecurity that incorporates all dimension of food security and efforts made so far, no single measure emerged as comprehensive enough to meet the criteria [26, 27]. Thus, there is no single and unique comprehensive measure of food security while assessing it in a way that it acknowledges multifaceted dimensions of food security concept. Due to this, the measurement of food security used in different literatures vary; and it is more of the compromise of researchers based on contexts of the study rather than absolutely perfect. Taking the context of this study in to consideration, the current study used Food Insecurity Experience Scale (FIES). FIES measures food insecurity in various domains of food insecurity at the household level as the condition of household not being able to freely access the food one needs to conduct a healthy, active and dignified life. From available food insecurity measurements such as Food Consumption Scale (FCS), Food Insecurity Experience Scale (FIES), Household Income and Expenditure Surveys (HIES), the current study preferred FIES due to the following reasons.

• It is the best method of analysing food insecurity at household level as it captures both physical and psychosocial dimensions.

• The study research questions: region-specific food insecurity, households’ food insecurity experience and the associated predictors can be well investigated with FIES.

• The study data, (HFPS-HH) 2020–2023, collected by World Bank was surveyed in FIES; to enable evidence based decision on socio-economic implication of Covid-19.

• FIES directly measures food insecurity in line with measuring food insecurity by Food and Agriculture Organization (FAO), Global Report on Food Crisis (GRFC), and Sustainable Development Goal [24, 28–30].

A summary of eight standard FIES questions, each with yes or no possible response, include the following.

• FI1: Household members have been worried that they will not have enough to eat because of a lack of money or other resources?

• FI2: Household members have been worried that they cannot eat nutritious foods because of lack of money or other resources?

• FI3: Household members had to eat always the same thing because of lack of money or other resources?

• FI4: Household members had to skip a meal because of lack of money or other resources?

• FI5: Household members had to eat less than they should because of lack of money or other resources?

• FI6: Household members found nothing to eat at home because of lack of money or other resources?

• FI7: Household members have been hungry but did not eat because of lack of money or other resources?

• FI8: Household members have not eaten all day because of lack of money or other resources?

Food insecurity severity classification

Approximate comparability of prevalence rates across regions with in the same country can be achieved by assigning food security status of households discretely based on raw score [31]. This method is used in all countries with established periodic assessment of food security using experience-based measurement scales. The scale used for households’ food insecurity severity is based on classifications developed by [24, 30] and later used by [23]. According to them, the severity of a household’s food insecurity is categorized with cut off points as: food secure if his FIES raw score is 0; mild food insecure if his FIES raw score is 1 to 3; moderate food insecure if his FIES raw score is 4 to 7; and severe food insecure if his FIES raw score is 8.

The predictors of household food insecurity experience included in the current study are from household and regional levels. Household level predictors considered include Sex of household head (1 = Male; 2 = Female); Age of household head (1 = Below 40; 2 = 40 and above); Residence of household (1 = Rural; 2 = Urban); Current employment status of household head (0 = No, 1 = Yes); Assistance benefit of household (0 = No, 1 = Yes); Household’s income change (1 = Not reduced, 2 = Reduced); Household Non-farm business (0 = No, 1 = Yes); Household farm since last call (0 = No, 1 = Yes); Worry of Covid-19 (0 = Worried, 1 = Not worried); Financial threat due to covid-19 (0 = Financial threat, 1 = No financial threat); Hand washing with soup after being in public (0 = No, 1 = Yes); Mask wearing in public (0 = No, 1 = Yes); and Time since first follow up period measured in months. Regional level predictors were derived by aggregating household characteristics.

Method of data analysis

The current study analysed the data by using R software and employed Kruskal Wallis Test (non-parametric test) and Generalized Multilevel Binomial regression model (parametric inference) to produce valid result targeted to addressing the intended objectives. Since food insecurity experience is measured as a count variable and compared among 11 independent regions, Kruskal Wallis Test is the appropriate statistical test for examining the variation across regions. In the overall test, the null hypothesis is tested against alternative using H- test statistic given by:

$$H=\left(\frac{12}{n\left(n,+,1\right)},\sum _{j=11}^{11},\frac{T_j^2}{n_j}\right)-3\left(n,+,1\right)$$

where: n: is the total number of households of all regions.

T_j: is the rank total for each j^th region.

n_j: is the number of households in j^th region.

Statistical model

Let $X_{\mathit{hj}}$ be food insecurity status of h^th household for $j^{\mathit{th}}$ FIES question. Then, the responses of each household to the eight FIES questions are measured as:

$X_{\mathit{hj}}=\left\{\begin{array}{c}1:ifahouseholdhexpriencesfoodinsecurityfor{j}^{\mathit{th}}question\\ 0:ifthehouseholddoesnotexpriencefoodinsecurityfor{j}^{\mathit{th}}question\end{array}\right)$ j = 1,2,…,8.

$$X_{\mathit{hj}}\sim Bernoulli(P)\Rightarrow P(X_{\mathit{hj}}=x)=p^x{(1-p)}^{1-x},x=0,1$$ 1

Given that $Y_{\mathit{rht}}$ is food insecurity experience of household h, the outcome variable, in r^th region at time t, then it is binomially distributed random variable obtained by counting food insecure responses to eight FIES questions for each h^th household given as:

$$Y_{\mathit{rht}}=\sum _{j=1}^8{X}_{\mathit{hj}}\sim Bin(8,p)\Rightarrow P(Y_{\mathit{rht}}=y)=\left(\begin{array}{c}8\\ y\\ \end{array}\right)p^y{(1-p)}^{8-y},y=0,1,2,3,4,5,6,7,8$$ 2

where: y is the number of successes (FIES questions to which a household get food insecure).

Generalized multilevel binomial regression model for longitudinal data

The primary assumption of this study is that there is variation in food insecurity among regions. In such case, the independence assumption of observations by classical regression model is not satisfied and makes ordinary least squares (OLS) to yield both inefficient parameter estimates and biased standard errors; thus it calls for mixed effect modelling where both random and fixed effect portions have to be included. Besides, the response variable (food insecurity experience of household), is binomially distributed (non-normal) as it is derived from summing Bernoulli distributed responses as described in Eq. (2) leading to Generalized Mixed Effect Modelling (GMEM). Moreover, the current study believed that food insecurity data set has nested structure (households are nested in regions) with both household level and regional level predictors that might affect food insecurity experience of the households, calling for two level analyses under the umbrella of generalized mixed modelling. Further, when longitudinal data are analysed under the generalized multilevel regression framework, the first nesting structure reflects repeated measurement of observations making it Level I nested within the households (Level II); and households nested within the regions (Level III). Thus, current study used Generalized Multilevel Binomial Regression model which is appropriate for analysing the effects of higher and lower level predictors for data having nested structure in cases where dependent variable is binomially distributed at the lowest level and explanatory variables can be defined at any level including aggregates of level-1 variables [8, 32, 33]. Thus, for the present study, the final model estimates the common sharing intercepts, two random intercepts, model residuals, and fixed effect coefficients corresponding to household and regional level predictors [32, 34, 35].

To account for the variations at each level securing the model’s parsimony and predictive power, this study employed random intercept model because the large number of predictors would make the model more complex when random slope and cross-over interaction are considered.

The study used log-odds,$ln\left[\frac{p(y_{rht)}}{1-p(y_{\mathit{rht}})}\right]$, to measure household’s food insecurity where.

$p(y_{\mathit{rht}})$: is the probability of food insecurity experience of household h in r^th region at time t with $y_{\mathit{rht}}$ measured as binomial random variable in Eq. (2).

The first multilevel binomial regression model is the intercept (null) model without predictors given as:

$$ln\left[\frac{p(y_{rht)}}{1-p(y_{\mathit{rht}})}\right]={\beta }_{0hr}+e_{\mathit{rht}}$$ 3

where ${\beta }_{0hr}$: is the random intercept that varies among regions and across households and.

$e_{\mathit{hrt}}$: is error term.

The multilevel binomial regression model with time and household level predictors is given by:

$$ln\left[\frac{p(y_{\mathit{rht}})}{1-p(y_{\mathit{rht}})}\right]={\beta }_{0hr}+{\pi }_{h}T_{_{\mathit{ht}}^{}}^{}+{\beta }_{10}{X}_{1hr}+{\beta }_{20}{X}_{2hr}+......{\beta }_{p0}{X}_{\mathit{phr}}+e_{\mathit{rht}}$$ 4

where $X_{1hr}...X_{\mathit{phr}}$: refer to p household level predictors of food insecurity with respective fixed effect regression coefficients (${\beta }_{10}......{\beta }_{p0}$); ${\pi }_{h}T_{_{\mathit{ht}}^{}}^{}$ is fixed time effect in months since the initial follow up with coefficient (${\pi }_h$); ${\beta }_{0hr}$ is the random intercept that varies among regions and across households and $e_{\mathit{hrt}}$ is error term. Since, ${\beta }_{0hr}$ (in Eq. (4) is the random variable, allowing the household’s food insecurity to vary among regions and across households, and adding region specific predictors’ yields:

$${\beta }_{0hr}={\gamma }_{00}+{\gamma }_{01}{W}_{1h}+......+{\gamma }_{0k}{W}_{\mathit{kh}}+V_{0h}+U_{0r}$$ 5

where $W_{1r}.....W_{\mathit{kr}}$ are k region-specific household’s food insecurity predictors with corresponding fixed effect regression coefficients (${\gamma }_{01}.....,{\gamma }_{0k}$); ${\gamma }_{00}$ is the common intercept shared by all households; $U_{0r}$ and $V_{0h}$ are random variations among regions and across households respectively. Substituting Eq. (5) with none region specific predictors in Eq. (3) yields intercept model without predictors as:

$$ln\left[\frac{p(y_{rht)}}{1-p(y_{\mathit{rht}})}\right]={\gamma }_{00}+U_{\mathit{or}}+V_{\mathit{oh}}+e_{\mathit{rht}}$$ 6

where: ${\gamma }_{00}$ is the common intercept and $U_{\mathit{or}},V_{\mathit{oh}},e_{\mathit{ij}}$ are the variances at the third, second and first level respectively. When Eq. (5) is substituted into Eq. (4), a single food insecurity prediction equation can be given by:

$$ln\left[\frac{p(y_{\mathit{rht}})}{1-p(y_{\mathit{rht}})}\right]={\gamma }_{00}+{\pi }_{h}T_{_{\mathit{ht}}^{}}^{}+{\beta }_{10}{X}_{1hr}+...{\beta }_{p0}{X}_{\mathit{phr}}+{\gamma }_{01}{W}_{1h}+...+{\gamma }_{0k}{W}_{\mathit{kh}}+U_{\mathit{or}}+V_{\mathit{oh}}+e_{\mathit{rht}}$$ 7

where ${\gamma }_{00}$ is the common intercept; ${\pi }_{h}T_{_{\mathit{ht}}^{}}^{}$ is fixed effect of time measured in months since the initial follow up; ${\beta }_{20}{X}_{2hr}+......+{\beta }_{n0}{X}_{\mathit{phr}}$ are fixed effect of household level predictors; ${\gamma }_{01}{W}_{1h}+......+{\gamma }_{0k}{W}_{\mathit{kh}}$ are fixed effect part of regional level predictors; $U_{\mathit{or}}$; $V_{0h}$ and $e_{\mathit{rht}}$ are as previously defined. In Eq. (7), the parameters under interest are ${\gamma }_{00},V_{0h}$ and $U_{0r}$.

based on [35].

The three models namely, Null model (M0), Model with household level predictors (M1), and Model with household and region-specific predictors were fitted and compared using null model as a baseline. Then, the model with the smallest AIC was used as a part of the study predictions.

Measure of dependence at each level

The intra-class correlation (ICC) measures the degree of clustering within the given group membership. At each intercept variance level, it ranges from 0 to 1 and indicates the proportion of the total variance explained by the i^th grouping. The total variance at higher levels is assessed through ICC_s at each random intercepts [33, 35].

The variance explained by level III is computed as: ${\rho }_R=IC{C}_1=\frac{{\widehat{U}}_{0r}}{({\widehat{U}}_{0r}+{\widehat{V}}_{0h}+{\widehat{e}}_{\mathit{thr}})}$.

The variance explained by level II is calculated as: ${\rho }_H=IC{C}_2=\frac{{\widehat{V}}_{0h}}{({\widehat{U}}_{0r}+{\widehat{V}}_{0h}+{\widehat{e}}_{\mathit{thr}})}$.

Variance explained by higher levels (level II and level III) is obtained by:

$${\rho }_{\mathit{HR}}=IC{C}_1+IC{C}_2+=\frac{{\widehat{V}}_{0h}+{\widehat{U}}_{0r}}{({\widehat{U}}_{0r}+{\widehat{V}}_{0h}+{\widehat{e}}_{\mathit{thr}})}$$

Proportion of household food insecurity variation explained by the predictors

In multilevel regression model, R-square can be quantified as total or level-specific [22]. As this study aimed to estimate the amount of variation and the predictors contribution to each level, the study computed the proportion of reduction in residual variance, level-specific rather than total variance measures, at each level due to inclusion of covariates based on [36].

$$Level(i)-R^2=\frac{Level(i)Varianc{e}_{\mathit{Null}}-Level(i)Varianc{e}_{\mathit{Full}}}{Level(i)Varianc{e}_{\mathit{Null}}}$$ 8

Although total variance of the multilevel can be suggestible when both random interest and random slope variations are considered, constraining the random slope variances to 0, the current study examined marginal total R² explained by predictors via fixed above and.

beyond the null model based on [33].

$$R_{S\&B}^2=1-\frac{{\tau }_{(full}^2)+{\gamma }_{(full)}^2+{\sigma }_{(full)}^2}{{\tau }_{(null}^2)+{\gamma }_{(null)}^2+{\sigma }_{(null)}^2}$$ 9

Results

The current study used longitudinal data collected from 3300 households in five follow ups. The analysis was made on total observations over time of n = 13,517 gathered from 11 regions of the country with a different, but representative, number of households of each region.

Region-specific and countrywide food insecurity prevalence

The comprehensive food insecurity prevalence of the country and each region with categories of mild, moderate, and severe food insecure statuses is presented in Table 1.

Table 1.

Results on the prevalence of household food insecurity at country and regional level

Regions in the country	Food insecurity prevalence (%)	Food insecurity prevalence(%) in severity
		Mild food insecure (%)	Moderate food insecure (%)	Severe food insecure (%)
Tigray	68.30	38.2	26.3	3.8
Afar	45.60	29.7	14.5	1.4
Amhara	59.10	28.1	23.4	7.5
Oromia	65.80	23.9	36.9	5.0
Somali	88.90	28.4	58.7	1.8
Benishangul-Gumuz	53.10	26.1	25.6	1.4
SNNPR	66.10	23.0	34.7	8.4
Gambella	55.7	30.0	22.5	3.3
Harar	42.2	21.7	18.7	1.8
Addis Ababa	52.7	35.0	16.3	1.4
Dire Dawa	48.2	26	18.4	3.8
Country level	57.3	28.7	25.0	3.6

Food insecurity prevalence (%) in severity is derived based on Cafiero et.al [24]; Adjognon et.al. [30]; and Rudin-Rush et.al. [23]. A household is mild food insecure if 0 < raw sum of FIES $\le 3;$ moderate food insecure if 3 < raw sum of FIES $<8$; and sever food insecure if $\text{raw sum of FIES}=8$

Result in Table 1 indicated that Sumali region experienced the highest (88.9%) food insecurity prevalence followed by Tigray (68.30%), SNNP (66.10%), Oromia (65.80%), and Amhara (59.10%); while, Addis Ababa(52.7%), Diredawa (48.2%), Afar (45.60%), and Harar (42.2%) had lower food insecurity prevalence as compared to other regions where Harar is with the least food insecure during the pandemic. Considering the severity of food insecurity prevalence, SNNPR was the leading (8.0%) followed by Amhara (7.5%) and Oromia (5.0%).

Food insecurity prevalence of regions by rounds

Figure 1 showed that all regions, except Afar and Amhara, experienced smaller food insecurity during round 6; while most regions experienced larger food insecurity during round 3 except SNNP, Gambella, and Dire Dawa. Besides, variation in food insecurity experience of a particular region among rounds is obtained at points representing it on different lines. Considering the food insecurity variation of each region among five rounds, Afar was the region with the most variable food insecurity experience followed by Sumali; while Oromia was the most consistent experiencing almost similar food insecurity prevalence during five follow ups followed by Gambella region.

Fig. 1.

Food insecurity prevalence of each region by rounds

Variation of Food Insecurity Experience of Households across Regions

Figure 2 showed that food insecurity experience of households is the highest (median = 4) in Sumali region where about 50% of the households get food insecure in 4 of 8 FIES questions during Covid-19 followed by SNNPR, Oromia, Tigray, and Amhara. On the other hand, Afar, Harar, and Dire Dawa were regions with the minimum food insecurity distribution. Considering the spread of food insecurity with in the region, Sumali region is the most consistent (shortest box) with households experiencing food insecurity in 3 to 5 of the FIES questions lie within the interquartile range. To ascertain the existence of variation in food insecurity experience among regions, the results from Kruskal Wallis test (overall test) and Dunn’s test (Post-hoc) are depicted in Table 2 for only significant pairs.

Fig. 2.

A boxplot summarizing the spread of food insecurity experience among regions

Table 2.

Results on test of variation in food insecurity experience among regions

Kruskal-Wallis rank sum Test (Over all test)		Chi-squared	Df	P-value	Sig.
		856.74	10	< 2.2e-16	***
Post-hoc (Dunn’s) test	Group 1	Group 2	Statistic	P.adj	Sig.
	Tigray	Afar	−10.8386685	9.029299e-26	***
		Sumalia	10.4657742	4.590155e-24	***
		Benshangul-Gumuz	−5.4608670	1.137156e-06	***
		Gambella	−4.5884668	8.037238e-05	***
		Harar	−11.4698056	7.670239e-29	***
		Addis Ababa	−10.6148742	9.657191e-25	***
		Dire Dawa	−8.6602352	1.553613e-16	***
	Afar	Amhara	8.8791378	2.358322e-17	***
		Oromia	14.1105710	1.601909e-43	***
		Sumalia	18.4272969	4.205551e-74	***
		Benshangul-Gumuz	4.8932176	1.984012e-05	***
		SNNP	12.6700636	3.898988e-35	***
		Gambella	5.3903543	1.617335e-06	***
	Amhara	Oromia	5.6054266	5.193567e-07	***
		Sumalia	12.4020687	1.095050e-33	***
		Benshangul-Gumuz	−3.4222275	9.937651e-03	**
		SNNP	5.2641682	3.098160e-06	***
		Harar	−9.2791645	6.149259e-19	***
		Addis Ababa	−8.1155156	1.499553e-14	***
		Dire Dawa	−6.4357295	3.194924e-09	***
	Oromia	Sumalia	8.8203627	3.879046e-17	***
		Benshangul-Gumuz	−8.3763340	1.747258e-15	***
		Gambella	−7.3272041	6.815110e-12	***
		Harar	−15.4209014	6.040415e-52	***
		Addis Ababa	−15.3403444	2.055044e-51	***
		Dire Dawa	−12.3210693	2.930050e-33	***
	Sumalia	Benshangul-Gumuz	−14.1072798	1.644183e-43	***
		SNNP	−7.2399857	1.256450e-11	***
		Gambella	−13.2372178	2.460726e-38	***
		Harar	−19.2816577	4.212033e-81	***
		Addis Ababa	−18.9626181	1.876025e-78	***
		Dire Dawa	−17.0964053	8.207930e-64	***
	Benshangul-Gumuz	SNNP	7.7793002	2.187804e-13	***
		Harar	−4.7048474	4.827063e-05	***
	SNNP	Gambella	−6.9318292	1.121670e-10	***
		Harar	−13.2968242	1.135098e-38	***
		Addis Ababa	−12.5817944	1.170049e-34	***
		Dire Dawa	−10.7928118	1.451672e-25	***
	Gambella	Harar	−5.2419659	3.336369e-06	***
		Addis Ababa	−3.7161972	3.438144e-03	**

P.adj Adjusted p-value, Sig significant, *** &** represent pair of regions that experienced significantly different food insecurity at 0.1 and 1% level respectively

From the overall Kruskal–Wallis test in Table 2, p-value < 2.2e-16 < 0.05=$\alpha$ provided a significant statistical evidence to reject the null hypothesis revealing that at least one of the regions experienced the problem differently.

Then, the question of which pair/s of regions was/were experiencing difference in food insecurity was analysed by using Post-hoc (Dunn’s) test in the Table 2 evidenced that there were significant differences in food insecurity experience among the majority of the regions. For instance, Sumali region experienced significantly higher food insecurity as compared to all regions; while SNNPR and Oromia had statistically the same food insecurity distribution; but they experienced significantly larger food insecurity as compared to all other regions except with Tigray. Other pairs of regions that experienced significance difference in food insecurity are presented in Table 2.

Identifying predictors of households’ food insecurity experience

Model fitting and selection

To assess reduction in random variation due to the predictors included in the models, the random effect parts of the three fitted models are given in Table 3.

Table 3.

Results on random effects of the three fitted models

Model 1: Random intercept/null model with no predictors (M0)
Random effects:
Groups	Name	Variance	Standard deviation
Region	Intercept	0.4945	0.7032
Household id: Region	Intercept	3.1821	1.7838
Residual	Error	2.3069	1.5189
Model 2: When only household level predictors are included (M1)
Region	Intercept	0.280	0.5291
Household id: Region	Intercept	2.807	1.6753
Residual	Error	2.047	1.4308
Model 3: When household and regional level predictors are included (M2)
Region	(Intercept)	0.1641	0.4051
Household id: Region	(Intercept)	2.8087	1.6759
Residual	Error	2.0461	1.4304

Random intercept model with no predictors (M0)

Dependence at level III (variation across regions) is: 

$IC{C}_R=\frac{{\tau }_{00}}{({\tau }_{00}+{\gamma }_{00}+{\sigma }^2)}=\frac{0\text{.4945}}{(0\text{.4945}+3\text{.1821}+2\text{.3069})}=0\text{.082644}=8.26\%$

This shows the correlation coefficient of food insecurity experience among households within the same regions is 0.0826 explaining that approximately, 8.26% of the total variation in food insecurity of a household is explained by variability at regional level which could be due to region specific predictors or region specific exposures to different factors. The value of ICC_R = 0.0826 > 0.05%, gives an evidence to include regional variation as random intercept.

Dependence at level II (variation across households) is:

$$IC{C}_H=\frac{{\gamma }_{00}}{({\tau }_{00}+{\gamma }_{00}+{\sigma }^2)}=\frac{3.1821}{(0.4945+3.1821+2.3069)}=0.531812=53.18\%$$

This value implied that 53.18% of the total variability in household food insecurity is explained by household characteristics. This larger variation gives an evidence to include households as random effect and assess the effect of household level predictors on their being of food insecure.

The total variation accounted for households food insecurity at regional and household level variability (dependence higher levels) is given by:

$${\rho }_{\mathit{RH}}=\frac{{\tau }_{00}+{\gamma }_{00}}{({\tau }_{00}+{\pi }_{00}+{\sigma }^2)}=\frac{(0.4945+3.1821)}{0.4945+3.1821+2.3069)}=0.614456=61.45\%$$

The remaining 38.55% of the variability is error at lowest level which might be due to unknown reasons (not due to group membership of the food insecurity data). Before predictors inclusion, variations at all (three) levels are noted as unexplained variances where some of these portions are explained as predictors are included to the model.

Selection of the fitted model

After inclusion of the household and, both household and regional level predictors in to the null model (M0), variations at all levels reduced with different degrees (see Table 3). This shows the included predictors explained some of the variations of food insecurity experience at all levels. Then, model comparison was made by using AICs of three models in Table 4. Based on this result, the most appropriate model is the third model (M2) with the smallest AIC. This model is more predictive and used as a part of the inference for the study.

Table 4.

Model comparison

Models	Npar	AIC	BIC	Log likelihood	Deviance	Pr(> Chisq)
M0	5	55,721	55,759	−27,856	55,711
M1	15	46,730	46,826	−23,352	46,704	< 2.2e-16 ***
M2	16	46,724	46,827	−23,348	46,696	0.003133 **

Npar number of parameters, AIC Akaike Information Criteria, BIC Bayesian Information Criteria, *** and ** represent model predicting significantly better than M0 at 0.1% & 1% levels respectively

Percentage explained at each level by the included predictors

This study considered the proportion of reduction in level-specific residual variance. The variance in the intercept due to variation among regions accounted for by the included predictors is: $R^2(\text{Re}gion)=\frac{0.4945-0.1677}{0.4945}=0.4935=49.35\%$

The variance in the intercept due to variation across households accounted for by the included predictors is: $R^2(Household)=\frac{3.1821-2.7582}{3.1821}=0.1332=13.32\%$

Similarly, the proportional reduction in lowest term (residuals) variance is:

$$R^2(observation)=\frac{2.3069-2.0545}{2.3069}=0.1094=10.94\%$$

Total variance explained by the included predictors could not be computed. However, to see the contributions of included predictors via only fixed effect, the marginal R² is computed as:

$$R_{S\&B}^2=1\frac{T_{(full)}^2+y_{(full)}^2+{\sigma }_{(full)}^2}{T_{(null)}^2+y_{(null)}^2+{\sigma }_{(null)}^2}=1-\frac{0.1677+2.0545+2.7582}{0.4945+2.05453.1821}=16.76\%$$

This means that portion of the total variance explained via only fixed effect over and beyond null model, constraining the random slope variances to 0, is approximately 17%.

Effect of predictors on household’ food insecurity experience

The result from the random part in Table 5 indicated that when household and regional level predictors are included in to the model, the variations at all levels in random model (M0) reduced i.e. variance across regions decreased from 0.4945 to 0.1641; variance across household decreased from 3.1821 to 2.8087; and variance of lowest level decreased from 2.3069 to 2.0461. This ascertained the included predictors explained some of the variations of the two intercepts and residuals.

Table 5.

Results on estimates of random and fixed effect parameters

Random effects:
Fixed effects:
Predictors of household food insecurity	Estimate	St. Error	t value	Pr( >|t|)
Groups	Name	Variance		Standard deviation
Region	Intercept	0.1641		0.4051
Household id: Region	Intercept	2.8087		1.6759
Residual	error	2.0461		1.4304
Intercept	4.919	0.5268	9.337	6.27e-07 ***
Sex of household head (reference = male)	0.4200	0.0793	5.297	1.25e-07 ***
Age of household head (reference age < 40)	−0.04176	0.0725	−0.576	0.5646
Residence of household (reference = rural)	−0.3707	0.1033	−3.589	0.0003 ***
Household employment (ref = not employed)	−0.1869	0.0776	−2.407	0.0161 *
Household assistance (ref = no assistance)	−0.3504	0.0045	−2.980	0.0029 **
Income loss after Covi-19 (ref = not reduced)	0.8562	0.0601	14.258	< 2e-16 ***
Household non-farm business (ref = no)	−0.4074	0.0729	−5.582	2.48e-08 ***
Household farm business (ref = no)	0.06896	0.0933	0.7390	0.4596
Households worry of Covid-19 (ref = yes)	−0.1586	0.0594	−2.670	0.0076 **
Financial threat due to Covid-19 (ref = yes)	−0.6189	0.0659	−9.398	< 2e-16 ***
Time in month since first follow-up	−0.02318	0.0114	−2.040	0.0415*
Urban household proportion in the region	−0.01878	0.0069	−2.701	0.0240*

^***, **, * represent a predictor influencing significantly at 0.1%, 1% & 5% levels respectively

Except age of household head and household farm business, all the included predictors had significant fixed effects in influencing households’ food insecurity during Covid-19. Gender of household head and income loss of the household were the positive predictors of household food insecurity experience. These mean that being a member of female headed household (${\beta }_1=$ 0.4200, t = 5.297, p < 1.25e-07) and income loss since Covid-19 (${\beta }_6=$ 0.8562, t = 14.258, p = < 2e-16) were significantly associated with increased household’s food insecurity. Other predictors: residence of household (${\beta }_3=$−0.3707, t = −3.589 p = 0.000336); current employment status of household (${\beta }_4=$−0.1869, t = −2.407, p = 0.016105), assistances provided to household (${\beta }_5=$−0.3504, t = −2.980, p = 0.002894), operating household non-farm business at least as before Covid-19 (${\beta }_7=$−0.4074, t = −5.582, p = 2.48e-08), household worry of Covid-19 (${\beta }_9=$−0.1586, t = −2.670, p = 0.0076), financial threat due to Covid-19 (${\beta }_{10}=$−0.6189, t = −9.398, p < 2e-16), and time duration since first follow (${\beta }_{10}=$−0.02318, t = −2.040, p = 0.0414) had significant indirect effect on household’s food insecurity.

These indicated that households that were living in urban; got employed; benefitted assistance; carried out non-farm business; not worried of Covid-19; and did not face financial threat due to Covid-19 experienced less food insecurity and vice versa.

The relative strength of predictor’s effect is associated with the magnitude of the corresponding log-odds. Income loss of households due to Covid-19 was the most determinant factor of food insecurity with the log-odds ${exp}^{0\text{.8562}}=2.4$ revealing that a household incurred income loss due to Covid-19 was $2.4$ times more likely to get food insecure as compared to the household whose income was not decreased due to COvid-19. Region’s proportion of urban population is the only regional level predictor that significantly contributed indirectly to food insecurity.

Discussion

Several studies conducted on food insecurity in the country supported the finding of the current study. However, the finding of this study is unique in that the pre-existing studies neither analysed food insecurity prevalence of the country and its all regions simultaneously nor assessed food insecurity variation amid regions with their associated lower and higher level predictors using longitudinal data. So, this study inroads future intervention to account for regional difference, higher level covariates and time varying effects.

The finding of this study indicated that there were statistically significant variations in food insecurity experience among regions during Covid-19; where Sumali region experienced highest food insecurity followed by SNNP, Oromia, Tigray and Amhara. This might be due to, in addition to Covid-19 pandemic, Sumali and Oromia regions were facing the problem of drought and displacement; SNNP region was incurring the same problem due to conflicts with in the region and partly with Oromia region; while households in Tigray and Amhara regions were also exposed to displacement due to political unrests in northern part of the country. This finding is similar with the fingings [7, 11, 19] in that the manifestation of shock increase the likelihood of a household to be food insecure. On the other hand, variation in food insecurity among regions could also be attributed to differences in ethnicity and language, feeding culture, agro-ecological, and spatial dependence among regions which confirms with the findings [18]. As time increased, food insecurity experience declined. This might be due to the fact that, as time goes, Covid-19 outbreak decreases and/or households might adapt to coping up with the problem of Covid-19. This is consistent with the findings [19] that revealed the impact of shock events on food insecurity experience largely depends on their ability to cope up with, adapt, and creating shocks resilience.

The finding from the current study indicated female-headed households experienced significantly higher food insecurity than male-headed ones which is similar to many findings of previous studies [7, 11, 20–22, 37]. This might be due to many females especially in rural areas of developing countries like Ethiopia are housewives; but become household heads when they get widowed or divorced, being they are less resilient to challenging environment like shocks.

Similar to studies [11, 12, 23]; but in contrast [30], households in urban are less likely to experience food insecurity as compared to their counter parts. According to this study, living urban increases food security experience of household not only due to his living in urban but also proportion of urban population in the region (regional level predictor) where the household resides. This might be in line with the finding [8, 11, 17, 18] in that zone and/or region-specific, or society-based analysis is necessary for the targeted interventions of households’ food insecurity as there could be higher level or contextual predictors.

Considering the employment status and assistance benefits of households during Covid-19, the current study result is consistent with the finding [7, 8, 38] in that increase in employment status and assistance benefits of households decrease their food insecurity experiences. Besides, similar to the findings [7, 12, 21, 37], increase in households income loss due to different factors is associated with increased likelihood of their being food insecure during Covid-19.

As the evidence of this study, operating household’s non-farm business during Covid-19 had significantly decreased his food insecurity experience; whereas, household’s farm business was not found to be significant. Thus, the problem of Covid-19 Pandemic might not be directly associated with farming; but associated more of with non-farm business. This might be due to measures taken to curb the spread of Covid-19 pandemic such as lockdowns and social distancing restrictions influenced household’s food security through affecting his non-farm business. Households’ fear/worry and financial threat due to Covid-19 were the two Covid-19 related factors that directly affected food insecurity. This finding is consistent with the findings [7, 19] that found shocks coming from various causes are expected to worsen food insecurity. However, other Covid-19 pandemic related factors namely, hand washing after being in public and mask wearing in public were not found to be significant predictors of households’ food insecurity. This can be consistent with many studies such as [23] and [39] in that Covid-19 influenced food insecurity indirectly through policy measures such as lockdowns, social distancing, and restriction requirements that might disrupt food supply, price, availability, accessibility, and transport cost.

The largest effect of income loss (${\beta }_6=$ 0.8562) may be due to the fact that, during Covid-19, several variables affect household’s food security indirectly through disrupting their income. This implied that income loss of household was one of the socio-economic implications of Covid-19 or other shock events manifested in Ethiopia for the last three years. The loss in households’ income during Covid-19 may be associated with many unknown factors disrupting the pre-existing income generating networks, indirectly affecting food security.

In spite of the important findings, this study is not free of limitation due to lack of the recorded data on some important variables such as family size, education level, household land holding and shock evens such as displacement, drought, and varying climatic conditions that occurred during Covid-19 in Ethiopia. For more understanding of food insecurity interplay with various disrupting situations in the country, it needs a further study by including variables not covered in this study and incorporating spatial information in to the effect.

Conclusions and recommendations

This study found the significant disparity in regions’ food insecurity experience in Ethiopia during COVID-19. Regions that experienced larger problem were also facing additional shock events such as displacements due to within conflicts and political unrests, and drought due to unpredictable weather variability. The reasons for the differing regions’ food insecurity experience could be the occurrence of shocks and other natural differences such as ethnicity, culture, agro-ecology and other factors that might be related to the regional administration. So, policy measures designed to insure food insecurity in the country should consider regions’ food insecurity inequalities and possible region-wise challenges. Specifically, while Productive Safety Net Program (PSNP), World Food Program (WFP), and Household Asset Building Programs (HABP) are implemented, government should consider the regions’ food insecurity inequalities and their vulnerability to shock event manifestations. Among other significant predictors, Covid-19 influenced households’ food security mainly through affecting their non-farm businesses and disrupting pre-existing socio-economic activities which led them to income loss. The decline in food insecurity experience of households, over time during COVID-19, was either due to the decrease in the outbreak of the pandemic or increase in adaptations with the problem. Enabling households to withstand sudden food insecurity occurrence reduces the magnitude of food insecurity problem that might occur due to shocks. To boost the households’ resilience to food insecurity manifested due to unexpected risk, government should conduct continuous awareness creation campaigns and provide advisory supports to households. Region-specific risk assessment before the occurrences of risk events and their possible remedial actions outsourcing region-wise options is important to ease the socio-economic impact of shocks.

Abbreviations

AIC

Akaike’s Information Criteria;

COVID-19

Coronavirus Disease of 2019

ESS

Ethiopian Socio-economic Survey

FAO

Food and Agricultural Organization

FIES

Food Insecurity Experience Scale

HFPS

High Frequency Phone Survey

SDG

Sustainable Development Goal

UNICEF

United Nations Children’s Fund

WB

World Bank

WHO

World Health Organization

Authors’ contributions

Henok Wariso Waqo: Designed the study conceptualization and made substantial contribution during acquisition or analysing of data, writing the draft and approved version for submitted article. Gezahegn Mekonnen Woldemedihn: Supervised, reviewed & edited the manuscript throughout all progress of the study (starting from conceptualization to final article submitted). Zeytu Gashaw Asfaw: Advised the whole work of the study (conceptualization, methodology, data analysis and interpretation) and made final edition to bring it to the finally submitted manuscript. Finally, all authors read and approved the final manuscript submitted.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The data sets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethical approval and consent to participate

This study used publicly available and de identified data collected by the World Bank (HFPS-HH 2020–2023), available via the World Bank Microdata catalogue: https://microdata.worldbank.org/index.php/catalog/3716/get-microdata. Besides, the original data collection obtained ethical clearance and all participants completed appropriate informed consent procedures prior to participating in the survey. Then, only consented surveys were kept in the dataset with all personal identifying information dropped from the clean dataset. As such, this analysis is exempted from ethical approval. However, authors also obtained a letter of consent for secondary analysis from College of Natural and Computational Science Research Ethics Review committee of Hawassa University, on May 22, 2024 with RERC reference number: CNCS-REC 038/24.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.