Shocks, socio-economic status, and food security across Kenya: policy implications for achieving the Zero Hunger goal

This study assessed the association between shocks, socio-economic factors, and household food security across Kenya, and provided policy implications for achieving the Zero Hunger goal at national and local levels in Kenya. We analysed the Kenya Integrated Household Budget Survey 2015–16 data for 24 000 households by employing regression models. Our multiple findings show that: (a) half of the surveyed population across Kenya were food insecure; (b) large disparities in food security status exist across the country; (c) demographics (e.g. gender, urban areas), and other socio-economic aspects (e.g. education, income, remittances), positively influence food security; and (d) social and economic shocks negatively influence food security. In summary, the food security status in Kenya is not satisfactory. Our findings suggest that, in general, achieving the sustainable development goals (SDGs) Zero Hunger goal by 2030 will likely remain challenging for Kenya. Ultimately, a redoubling of efforts is required to achieve SDG 10 (reducing inequality) to ensure no one is left behind. Further, the findings could be useful in the formulation and implementation of national and regional policies for achieving the Zero Hunger goal by 2030 in Kenya.


Introduction
For decades, one of the most popular global goals of human society has been to reduce persistent food insecurity. Actions included the declaration of food security as a basic human right in 1948, the World Food summit of 1996, the Millennium Development Goals of 2001, and the 2015 sustainable development goals (SDGs). Despite these remarkable initiatives, the status of food security in various world regions is far from satisfactory. For example, the second SDG on Zero Hunger is behind track and will only be achieved with substantial additional efforts (United Nations Department For Economic And Social Affairs 2019). By definition, food insecurity is limited physical, economic, or social access to food, while hunger is the uneasy or painful sensation caused by insufficient consumption of food (Jones et al 2013, FAO 2019. The Food and Agriculture Organisation (FAO) of the United Nations (2019) frames hunger as chronic undernourishment.
According to the most recent report by the FAO, one in ten people are food insecure. More than two billion people globally are experiencing moderate or severe food insecurity, and at least 690 million people are still hungry (Davis et al 2020, FAO, IFAD, UNICEF, WFP & WHO 2020. While food security across the world is slowly improving, sub-Saharan Africa is the only region in the world where food insecurity has risen since 2014. More than onequarter of the population in Eastern and Middle Africa is food insecure (Coughlan de Perez et al 2019, FAO 2020. Among these Eastern African countries, Kenya is one of the most food insecure; it has made slow progress in achieving its millennium development goal targets, and its progress in achieving the SDGs (in particular the Zero Hunger goal) lags behind expected achievements (FAO, IFAD, UNICEF, WFP & WHO 2020, Musyoka et al 2020. Food insecurity in Kenya affects 2.6 million people, with significant differences between counties and regions (KNBS 2018a). In general, more than half of the population in Kenya is suffering from moderate to severe food insecurity. Kenyan arid and semi-arid lands, urban slums, and rural households have high food and nutrition insecurity compared to the national averages (FAO, IFAD, UNICEF, WFP & WHO 2020). Kenya was ranked 86 out of 113 countries for food insecurity by the global food security index in 2017 (Government of Kenya-GoK 2018). Despite several national and international initiatives, Kenya still is in the level of serious hunger with a rank 84th out of the 107 countries globally in 2020 (GHI 2020).
Achieving the Zero Hunger goal by 2030 will be highly challenging due to the future impacts of climate change (Stevens andMadani 2016, Niles andBrown 2017), spatial distribution of the food insecure population (Hossain et al 2016), and social and economic shocks at household, local, and national levels (FAO, IFAD, UNICEF, WFP & WHO 2020, Ingram 2020. Understanding the association between food security and socio-economic characteristics is necessary to understand the way multiple factors influence food security across different scales (FAO 2013, Ingram 2020.
Shocks are additional threats to achieving household food security (DFID 2003, Ifejika Speranza et al 2008, Alinovi et al 2010. In general, shocks are events that can cause significant reduction of wellbeing such as income loss and food insecurity (Marques 2003), and typically sudden disturbing events (e.g. floods, epidemics or rapid rise in food prices), with often unpredictable and traumatic impacts such as collapse of livelihoods and economies. Further, shocks can be sudden social changes (e.g. the death of a household member) (Berend 2007, Kozel et al 2008 which also increase vulnerability and threats to food security (DFID 2003). Socio-economic factors, conflicts or climate trigger shocks such as a food crisis due to sudden rise in food prices and increased income inequality (FAO 2019). Economic, social, and environmental shocks prolong and exacerbate the severity of acute food insecurity (Conklin et al 2018, Cottrell et al 2019. This is because they reduce households' ability to maintain food security. If ignored, these shocks may have unpleasant effects on food insecurity in all its forms. The FAO (2019) notes that shocks disproportionately challenge food security in places where inequalities in the distribution of socio-economic factors and other resources are profound. One way to overcome this problem is to understand better the impacts of such disparities in order to prioritise actions and implement tailored strategies depending on available resources (Hong et al 2019). There is thus a need to monitor all SDGs at regional and sub-regional levels to identify ways to reduce inequalities, an aspect addressed in SDG 10. In particular, reducing inequality within countries helps to ensure the progress of SDGs, leaving no one behind. Ultimately, it is important to understand the spatial pattern of food insecurity and recognise the drivers associated with the food insecure population using reliable data sets. This will help to monitor variability in food insecurity and its drivers and thus provide scientific knowledge for long-term planning to achieve Zero Hunger through geographically and socially targeted interventions.
Previous studies on food security in Kenya mostly focused on demand and access to food (Koir et al 2020), household vulnerability to food security shocks (Musyoka et al 2020), impacts of drought on food security and gender perspective (Huho andMugalavai 2010, Kassie et al 2014) and basics of food consumption and poverty status (KNBS 2018b). However, it has not yet been explored how household socio-economic characteristics in the context of combined social, environmental, and economic shocks influences household food security across Kenya. Most studies are based on case studies (e.g. Ulrich et al 2012, Mutea et al 2019 of food security making it difficult to gain an overview of food security at the county and national levels. Yet, data collected for national overviews such as the Kenya Integrated Household Budget Survey 2015-16 (KIHBS), can fill this gap of gaining a national and county level overview of food security and complement insights gained from case studies. Thus, we analyse the spatial heterogeneity of food security and the associated drivers (socio-economic factors and shocks) using the 2015-16 KIHBS collected across Kenya in order to provide policy insights for achieving the Zero Hunger goal in the methods section, we explain the 2015-16 KIHBS datasets and data analysis (logistic regression) including how we define food security. Next, we explain the results focusing on food security across Kenya, and the association with shocks and socioeconomic characteristics, before discussing the progress of food security and policy implications for achieving the Zero Hunger goal in Kenya. This novel study highlights the usefulness of national-level datasets for understanding food security in Kenya and could be useful in the formulation and implementation of national and regional policies for achieving the Zero Hunger goal by 2030 in Kenya and other similar East African countries.

Data and variables
The KIHBS 2015-16 data is a household survey that collects information from the Kenyan population in order to guide national development policy decisions (KNBS 2018a). The KIHBS questionnaire, designed by experts, is a set of modules that are administered to collect information on household characteristics, housing conditions, education, general health characteristics, nutrition, household income and credits, household transfers, information and communication technology, domestic tourism, shocks to household welfare, and access to justice. From these key variables, we chose our outcome and predictor variables for food insecurity.
The survey was conducted by the Kenya National Bureau of Statistics from September 2015 to August 2016. Three-stage sampling was followed in order to determine sample size independently for each of the 47 Counties of Kenya, resulting in a planned national sample of 24 000 households. However, due to missing values, the total sample consists of 21 773 households. The samples are representative at the national level, the county level (n = 47), and the local level (urban or rural place of residence). We limited our analysis to KIHBS 2015-16, as the previous dataset KIHBS 2005KIHBS -2006 is not consistent with the current dataset of KIHBS 2015-16, which has been improved in terms of indicators and data collection. For example, the number of indicators for shocks and food items are higher in KIHBS 2015-16 due to the inclusion of new variables. Some other variables such as remittances have been recently included in KIHBS 2015-16. In addition, some of the variables such as dead and stolen livestock are divided into two shocks in KIHBS 2015-16. Therefore, considering these points, we limited our approach to the crosssectional analysis of the KIHBS 2015-16.

Data analysis
The outcome variable was household food security. We measured this variable using indicators proposed by the International Food Policy Research Institute Subandoro 2007, Szabo et al 2015). The approach considered two key indicators of food security: the percentage of total household expenditure on food and the total daily calorie availability at the household level. The survey did not explicitly assess food security using these indicators, therefore, we combined variables in the dataset to compute the aforementioned food security indicators.
The share of total household expenditure (as a proxy of income) spent on food is an indicator of household food security because it is widely documented that the poorer and more vulnerable a household, the larger the share of household income spent on food. A rise in food prices results in a higher share of total household expenditure spent on food, which constrains poorer households' resources. These force poor households to spend more on basic staples, reduce the quality of their diets, or even reduce the quantity consumed of the least expensive foods, while also reducing non-food expenditures that may be equally needed such as on health and education (Lele et at 2016). This indicator uses the monetary value of household consumption disaggregated into food and non-food items. Thus, the share of household food expenditure is equal to the percentage of expenditure on food divided by total expenditure (Smith et al 2014). A household was categorised as food insecure if more than 75% of its total expenditure went on food items (Smith and Subandoro 2007).
In the calorie-based food security analysis, a household was classified as food secure if daily calorie requirements were higher than total reported energy intake per capita. We made a final categorisation based on the combination of the above two variables; a household was categorised as food secure if at least one of the above conditions was met. This study used two key categories of predictor variables: household socio-economic characteristics and shocks to household welfare that comprised 19 and 22 independent variables, respectively (table 1). On one hand, socioeconomic characteristics comprise factors such as education, income and social support that influence households' wellbeing. On the other hand, shocks are sudden events such as death of household head that make households vulnerable.
We performed logistic regression modelling in order to test the main predictor variables influencing household food security. Before running the regression modelling, polychoric correlation was used as a test for independence and multicollinearity. In polychoric correlation, variables are redundant if the correlation is higher than 0.70 (Aletras et al 2010), As a result, we dropped the following variables: marital status, source of domestic water, electricity connection, source of cooking and lighting energy, number of livestock, and large rise in food and farm input prices.
Given that the outcome variable was dichotomous, we applied a series of logistic regression models with food security as the outcome variable in all the models to check the robustness of the final regression model. Model 1 examined the relationship between the outcome variable (food security) and 18 predictor (independent) variables defining the shocks to household welfare. The second model included the socio-economic characteristics in addition to the model 1 predictor variables (shocks). The third model included the outcome variable while the predictor variables were shocks and socioeconomic characteristics excluding household remittances. The fourth model represented (Model 2 and assests) the relationship between the outcome variable and shocks, socio-economic characteristics, and assets as the predictor variables. The adjusted regression model with predicted variables was specified as follows: where Y i is household food security status with binary values (0 = food secure, 1 = food insecure), β 0 is a constant, X 1i denotes shocks to household welfare, β 1 is the coefficient that shows the magnitude and direction of the relationship, and X 2i , X 3i , X 4i are socio-economic characteristics. The results of the logistic regression were interpreted using odds ratios (OR) and associated confidence intervals (CI). Standard post-estimation tests were applied to evaluate model fit and facilitate model selection (Szabo et al 2015). These included the Bayesian information criterion (BIC) and Akaike information criterion (AIC)

Food security across Kenya
The description of the studied households' characteristics is presented in the supplementary file. Overall, 52% of households were food secure. This classification was based on a combination of calorie deficiency and food expenditure indicators as explained in the data analysis section. Of the 52% food secure households, 70% and 30% were male and female-headed households, respectively and, 51%, 12% and, 37% of households were food secure respectively in rural, peri-urban and urban areas. The prevalence of food security was similar between households that did not practice agriculture and those involved in agriculture (50% and 50% respectively). Regarding our food security indicators, calorie deficiency was the major cause of food insecurity, affecting 84% of households. The mean calorie intake per adult equivalent was 2828 ± 12 calories. Moreover, 58% of the households spent over 75% of their income on food, making them food insecure. Surprisingly, rural households spent on average 79% of their income on food, whereas for urban households this figure was 62%.
However, as shown in figure 1, there were variations across the country, with less than 50% of households found to be food insecure in almost half of the counties. Based on our analysis, households across Kenya were divided into four clusters according to food security status (10%-30%, 31%-50%, 51%-70% and 71%-90%) across the 47 counties. This aimed to simplify the food security status across counties by allowing a quick glance on counties that have similarities in terms of food security status across the country, useable for future interventions. Cluster one contained four counties (7% of the total-Nairobi, Mombasa, Machakos, and Kiambu), with over 70% of households being food secure. Most of these counties are in the central region, with one in the coastal region. Cluster two comprised 21 counties (45% of the total), where over half of households were food secure. Cluster three comprised of 20 counties (42%), where more than 50% of households were food insecure. The fourth cluster contained two counties from the north-eastern arid region (Wajir and Mandera), with 85% and 75% of households living in food insecurity respectively. Surprisingly, in Garissa County, which is also in the northeastern arid region, 59% of households were food secure.

Association of food security with shocks and socioeconomic variables
The results of regression modelling between household food security and the predictor variables (see table 2) are shown in table 3. These results are based on Model 4. Regression analysis showed that household demographics (e.g. gender of household head, household size), socio-economic characteristics (e.g. remittances, household income, farming, cooking appliances, television), and four types of shock (death of livestock, death of household head, death of a working household member, jail term for household head) significantly influenced household food security (table 3).
Among the socio-economic variables, household income and remittances were the strongest predictors of household food security across the 47 counties. For instance, the OR (95% CI) of becoming food secure were 1.54 (1.42-1.68) from receiving remittances compared to those that did not receive remittances. The odds of becoming food secure from receiving remittances were higher in urban areas (1.86, p = 0.00) compared to rural (1.47, p = 0.00) and peri urban (1.52, p = 0.00) areas. The odds of food security for households increased along with household income.
Households headed by a woman were 21% (95% CI: 1.12-1.32) more likely to be food secure than male-headed households. Households with secondary education had 0.88 times the odds of households with no education for food security. The odds of food security were higher for households with primary and university education than for those with no education, but these results were not significant. In comparison to households living in rural areas, households in urban areas had higher odds (0.89, p = 0.04) of food security (95% CI: 0.78-0.99).
However, in terms of the age of the household head, an increase in age was only associated with a very slight increase in household food security: an OR of 1.01. The OR of food security for families that owned a television were 0.62 (95%: CI 0.55-0.69) and statistically significant (p = 0.00). The odds of food security for households not engaged in agriculture were 20% higher than for households in agriculture.
The regression model showed that only four shocks out of 19 (breadwinner jailed, death of household head, death of a working household member, death of livestock) were found to have a significant influence on food security. Death of livestock was found to have severely and significant influence on food security in rural areas (0.81, p = 0.00) compared to urban (0.79, p = 0.23) and peri-urban (1.01, p = 0.92) areas of Kenya. The majority (53%) of counties had encountered all four shocks; 34% had been hit by three (death of household head, death of a working household member, and death of livestock); and 13% had experienced two types of shocks (death of livestock and household head). All four significant shocks were social and economic in nature and had a negative impact on household food security. Interestingly, no environmental shock had a significant effect on household food security.
The odd ratios of food security after death of livestock were 21% less than cases where no livestock had died. This result was statistically significant at p-value 0.00. The most affected region was the Rift valley, followed by Eastern and Nyanza. Similarly, households that experienced death of household head (p = 0.00, 95% CI: 0.58-0.91) or working household member (p = 0.03, 95% CI: 0.45-0.96) were ∼40%   less food secure than households who had not experienced these shocks. Regression Model 1 shows that nine shocks were statistically significant (table 4); in contrast, in Model 4, only four shocks remain statistically significant considering socio-economic variables. Regression Models 2, 3, and 4 showed that socio-economic variables were a strong predictor of food security. Furthermore, social and economic shocks had a stronger influence on food security than environmental shocks. Considering the lowest values of AIC and BIC from regression results, we argue that Model 4 performed best among the four models.

Progress and drivers of food security
This study assessed food security status at the national level and across the 47 counties of Kenya. Additionally, we assessed the socio-economic aspects and shocks affecting household food security. Our findings show that half of the households across Kenya were food insecure. Out of the 47 counties, 25 counties were within national food security levels, while the rest were below the national average. However, our results also indicate differences in food security levels across the 47 counties in Kenya.
This study reveals a positive association between food security and socio-economic variables such as gender of household head, family size, remittance, and income. These results are in line with those of Our analysis also revealed a negative significant association between household food security and socio-economic characteristics (e.g. ownership of a television set) and shocks (e.g. death of livestock, death of a working household member, death of household head).
We found that all the shocks were spread more or less equally across the 47 counties, with the most common being death of livestock. This implies that for those households owning livestock, death of livestock and by extension ownership of livestock are significant drivers of food security. Livestock keeping (e.g. sheep, goat, dairy cows and poultry) in urban areas makes important contributions to the livelihoods of urban livestock keepers (Roessler et al 2016, Alarcon et al 2017, Pablo et al 2017, Crump et al 2019. Urban livestock keeping is a source of food security due to provision of essential micronutrients to avoid malnutrition and can release pressure on poor households (that spend 60%-80% of income in food) (Alarcon et al 2017). Rearing livestock enables smallholders to have improved livelihoods and to avoid food insecurity through income generation and can be used as a coping strategy during times of need (Nabarro and Wannous 2014).
Kenya has addressed the issue of food security in its Vision 2030 plan and the present government's 'big four' agenda. These initiatives emphasize investing in agriculture, with the aim of transforming agriculture Table 3. Regression results for the four models with household food security as the outcome variable ( * 0.05 * * 0.01 * * * 0.001  from subsistence to productive commercial farming as a pathway to food security (GoK 2007(GoK , 2018. However, our findings reveal that households not involved in agriculture are 20% more likely to be food secure. There are two possible explanations for this result.
First, most of Kenya is semi-arid and its agricultural production is challenged by climate variability and climate change, use of outdated technology, poor infrastructure (especially roads linking farmers to markets), soil degradation, regions with low cropping potential, diseases and pests, lack of fallows, and nutrient amendments (Foeken and Owuor 2008, Thornton and Herrero 2016, KARI 2019. These problems result in little or no harvest, leading to food shortage and hence food insecurity. Surprisingly, no significant impacts on food security were found from environmental shocks such as droughts, floods, pests, and diseases, which are usually related to climate variability. This could be due to the cross-sectional datasets of KIHBS, collected from September 2015 to August 2016. Longitudinal datasets are often a prerequisite for analysing the social impacts of climate change (Geffersa andvan den Berg 2015, Bahruid et al 2019). As droughts and floods are widespread in Kenya, they are systemic factors that can affect all inhabitants hence socio-economic characteristics becomes a differentiating and important factor in face of such system-wide exposures. This may be the reason for the non-significant association between food security and environmental shocks such as drought as the result show a non-significant possibility of 10% less food security for households experiencing drought and flood. In addition, households are also adapting to diversified livelihoods, resulting in less dependency on agriculture, where resources are becoming increasingly scarce (Babatunde andQaim 2010, Menike andArachchi 2016). In response to coping with drought, households mostly depend on livestock when adapting to climatic change (Ifejika Speranza 2010). Often environmental shocks (e.g. diseases, drought, floods etc) trigger livestock diseases, which may lead to livestock death, so environmental shocks can be the underlying drivers of livestock loss, which then directly affects food security.
In addition, we found that female-headed households were more likely to be food secure than maleheaded households. There were no major variations across the counties in terms of gender of household head, with over 60% of households being maleheaded in most counties. A possible explanation for this outcome is that women play a decisive role in dietary diversity at the household level. Other scholars have also found a significant association between the availability of a diverse diet at household level and women's participation in decision-making (Amugsi et al 2016). Women are also more involved in a variety of food system activities such as production and processing food, which are key in food availability and utilisation. However, such households are more often reported to be less endowed with necessary resources such as land and finances compared to male-headed households, which makes them vulnerable to food insecurity (Kassie et al 2014).

Policy implications for achieving the Zero Hunger goal in Kenya
The results suggest (figure 2) that given current conditions, achieving the Zero Hunger goal by 2030 is achievable for very few counties (e.g. those with 60%-70% population food secure) in Kenya; the rest (less than 40% of population food insecure) will likely continue to be food insecure for a long time if no additional efforts are put in place. These findings suggest that, in general, achieving Zero Hunger by 2030 will likely remain challenging for Kenya. This is because of the huge variations and disparities existing across the country. There are four counties that could certainly meet this goal, even before 2030. Twenty one further counties, with some effort, could feasibly be food secure by 2030. However, it is highly unlikely that the remaining 22 counties will be 100% food secure by that time. Considering the results, that social and -economic shocks had a stronger influence Figure 2. Progress of the Zero Hunger goal to achieve food security across counties (A > 50%, B < 50% food secure) in Kenya. See SI table 2 for county-wise food security data. on food security than environmental shocks, holds implications for achieving the zero-hunger goal.
First, there is a need for actions to improve system-wide resilience to environmental shocks. While these shocks seem not to have significant impacts at the inter-household level, they condition the agricultural production system for all households through influencing natural production conditions. Measures are thus needed to reduce the sensitivity of crop and livestock production systems to environmental shocks. Kenya is guided by several strategic documents towards the achievement of food security: nationally by Vision 2030 and the 'big four' agenda (GoK 2007); its national adaptation plan and drought management strategies to end drought emergencies (GoK 2016), regionally by the African Union (AU) Malabo Declaration (AU 2014); and globally by the United Nations (UN) post-2015 goals (UN 2019). The effectiveness of such initiatives thus needs to be monitored to ascertain to what extent they address the systemic vulnerability underpinning food insecurity in Kenya.
Second, our results show that attaining food security for all involves more than just producing more food, even though increasing agricultural production is a big part of the solution to eradicating hunger. The results highlight the need to also address disparities in socio-economic characteristics. It is important that governments comprehensively combine sustainable agricultural investments with crosssectoral developments (e.g. appropriate technology, market infrastructure) to improve agricultural production and to diversify and increase income levels. This approach has worked well in Ghana, leading to agricultural development (Adolph and Grieg-Gran 2013). Elsewhere, in Malawi and Bangladesh, subsidies have been effective in reducing food insecurity and contributing to environmental sustainability (Hossain et al 2016), hence such an option is worth exploring for Kenya.
Moreover, to ensure no one is left behind along the Zero Hunger goal pathway, it is essential to redouble efforts towards addressing the challenges that affect the most food insecure counties in terms of socio-economic characteristics. On the more challenging side, access to quality education is essential, as educated households are food secure. Our results suggest that households with secondary and other types (primary and university) of education are significant and non-significant respectively, but have a positive influence on food security in Kenya. Therefore, achieving other SDGs such as quality education (SDG 4) is necessary to end hunger across Kenya.
This study was limited to a cross-sectional (snapshot of a single moment in time) analysis, with the aim of ascertaining policy implications for achieving the Zero Hunger goal by assessing the status and drivers of food security at both national level and administrative unit (county) level. An analysis of qualitative data and a longitudinal study (repeated observations) considering seasonality of shocks may offer a deeper contextual understanding of the impacts of environmental shocks on food security, its complexities, and its subjectivity. Further studies that extend and collect social and ecological datasets may also offer an understanding of the interactive relationships between the Zero Hunger goal and other goals, which would help to set meaningful targets and achieve these targets comprehensively. However, the result of our study will be useful for assessing how Kenya has progressed in terms of the Zero Hunger goal and for guiding national and regional policies aimed at progressing towards achieving the SDGs in Kenya and other similar East African countries by 2030.

Conclusion
Food security analysis across Kenya can provide important information about achieving the Zero Hunger goal; it can also be useful for decisionmakers at global, national, and local levels. In this research, analysis of KIHBs datasets has shown that demographics (e.g. gender of household head, family size,) and other socio-economic characteristics (e.g. income, remittances, education) are positively associated with food security and that social and economic shocks (e.g. death of household head or livestock) are negatively associated with food security across Kenya. In general, food security status both at national and county levels is not satisfactory. It is unlikely that Kenya will be able to achieve the Zero Hunger goal by 2030, considering current food security levels, social (e.g. poverty, inequality) and environmental (e.g. climate, land degradation) challenges, and the ambitious targets set out by the SDG for Zero Hunger goal. These findings highlight the usefulness of regular (e.g. every 5 yr) collections of data sets at national-level for understanding food security, and can complement insights from household food security surveys, considering the larger efforts needed for case studies at household and local levels.

Data availability statement
The datasets used and/or analysed during the current study are available from the Kenya National Bureau of Statistics. The data that support the findings of this study are not openly available due to the copyright of Kenya National Bureau of Statistics.
No new data were created or analysed in this study.