Changing Climate, Changing Food Consumption? Impact of Weather Shocks on Nutrition in Malawi

Abstract In this study, we capture the impact of changing temperature and rainfall patterns on the estimated consumption of macro-and micronutrients among households in Malawi. We apply a fixed-effects model to household panel data collected between 2010 and 2017, which contains detailed information on food consumption, combined with rainfall and temperature data over the past 30 years. In turn, we aim to identify the impact of weather shocks on household food and nutrition security. We find decreases in rainfall from the long-term average results in declining daily consumption of macronutrients (carbohydrates, protein, fat), and micronutrients (iron, zinc, vitamin C, and B2). Increases in temperature from the long-term average are associated with reduced daily consumption of carbohydrates, protein, iron, zinc, vitamin A, B2, folate, and B12. These results suggest that reduced rainfall and higher temperatures will exacerbate food insecurity in Malawi. As the effects of climate change are becoming more apparent, enhancing our understanding of the effect of weather shocks on nutrition in developing countries will be vital to aid policymakers to implement targeted interventions to advance food and nutrition security in Sub-Saharan Africa. This will be essential to achieve the sustainable development goal of ‘Zero Hunger’ by 2030.


Introduction
As the effects of climate change are becoming more apparent, food insecurity has increased in tandem in several regions of the world.About 750 million people are affected by severe food insecurity with 2 billion people not having access to adequate amounts of safe and nutritious food in 2019.This is an increase of 60 million people globally between 2014 and 2019 who are impacted by hunger.Rising global temperatures will influence the intensity and frequency of weather events, such as droughts and flooding, with the risk of extreme weather events increasing alongside rising temperatures (IPCC, 2018).Such events will cause disruptions across entire food systems, affecting the quantity of food produced and the nutritional quality of diets, as well as reducing food distribution and access (Fanzo, Davis, McLaren, & Choufani, 2018;Myers et al., 2017).
The impact of these adverse events will be felt more intensely in countries in Sub-Saharan Africa (SSA) and other countries located at lower latitudes with warmer climates, and whose economy is heavily reliant on agriculture (Fanzo et al., 2018;Mendelsohn, Dinar, & Williams, 2006;Serdeczny et al., 2017).Alongside this 250 million people in Africa are currently suffering from undernourishment and by 2030 the continent is set to overtake Asia in having the highest number of undernourished people (FAO et al., 2020).Consequently, as a global community, we are not on track to meet the Sustainable Development Goal (SDG) of Zero Hunger (SDG 2) by 2030, which aims to reduce hunger and all forms of malnutrition (UN General Assembly, 2015).
Located in SSA, Malawi is vulnerable to climate change due to its reliance on rain-fed agriculture and heat-sensitive crops.Twenty-eight percent of the total Gross Domestic Product (GDP), and 61 percent of the land area is used for crop production and livestock, with farmers representing 80 percent of the total population (FAO, 2018).Maize is a staple crop, grown by 97 percent of farmers and contributes about 54 percent of total calorie intake in Malawi (Minot, 2010;Warnatzsch & Reay, 2020).Maize is sensitive to the increases in temperatures, which are projected to rise in Malawi by 1.1-3.0C by 2060 (GFDRR, 2011).
Furthermore, as the effects of climate change are expected to become more extreme food insecurity is also expected to increase.Consequently, there is a pressing need to quantify the impacts of climate change on food and nutrition security to help policy makers design and implement interventions that will promote resiliency among vulnerable communities in SSA.
Therefore, to address this research gap the objectives of this paper are to assess the impact of weather shocks on 11 food and nutrition security outcomes reflecting both macro-and micronutrient consumption.Estimated consumption of calories, carbohydrates, protein, and fat are used to quantify macronutrient consumption.Our approach also uses estimated micronutrient consumption as a proxy to capture the effects of a changing climate on food utilisation, which is the least studied pillar of food security in terms of climate change research (Zewdie, 2014).Intakes of micronutrients are especially indicative of dietary quality, with higher intakes being associated with better dietary diversity and consumption of more nutritious foods.Iron, zinc, vitamin A, C, B2, folate, and B12 have been selected as the micronutrients of study as they are vital for health and represent a significant public health burden in Malawi.
Malawi is used as the case study for this research due to the country's vulnerability to both food and nutrition security and weather shocks.Furthermore, a representative sample of Malawian household-level panel data, containing detailed information on food consumption covering the years 2010 to 2017 exists to allow the econometric assessment of the impact of shocks on macro-and micronutrient consumption.Household-level data have also been combined with 30 years of weather data, disaggregated at the district level, to quantify the household's exposure to weather shocks.
To the best of the author's knowledge, this is the first study to assess the impact of weather shocks on both macronutrient and micronutrient consumption in Malawi or SSA.Such an assessment can aid policymakers to identify those most vulnerable to climate change and implement targeted interventions to advance food and nutrition security in SSA.
The remainder of this article is organised as follows.Section 2 provides an overview of nutrition and policy in Malawi, while Section 3 presents our empirical framework.In Section 4 we present our empirical results and discussion of our main findings.Section 5 outlines the policy implications of our findings.We discuss the possible limitations of our approach and concluding remarks in Section 6.

Overview of nutrition and policy in Malawi
Micronutrient deficiencies (MND) are a serious global public health concern, contributing to disease burden, as they lead to more illnesses, physiological, physical, and cognitive impairments, metabolic conditions, as well as decreased resistance to infectious diseases (WHO, 2006).Micronutrient surveys in Malawi conducted between 2000 and 2010 reveal deficiencies in iron, zinc, vitamin A and B vitamins, such as folate, B2, and B12, making them a significant public health concern (National Statistical Office Malawi et al., 2017).
In the country, several nutrition and health policies have been implemented to promote adequate nutrition and health outcomes, with a special focus on vulnerable groups: children under the age of five years, adolescents and children, pregnant and lactating mothers, elderly populations, and individuals living with HIV and AIDS (Department of Nutrition, HIV and AIDS, 2018;Machira & Chirwa, 2020).Policies include the National Plan of Action for Nutrition (2000), Infant andYoung Children Nutrition Policy andGuidelines (2003-2020), Food and Nutrition Security Policy (2005), Food Security Policy (2006), National Nutrition Policy and Strategic Plan (2007-2011), Food Security Action Plan (2008-2013) Supported by these policies, interventions have been implemented across the country.These have focused on micronutrient supplementation (fortification) of staple foods, e.g.sugar, oil, maize, and wheat flour fortified with vitamin A, targeted supplementation for specific age groups e.g.young children with vitamin A and pregnant women with iron and folate, as well as education regarding nutrition (National Statistical Office Malawi et al., 2017;Williams et al., 2021).
Interventions have been successful in reducing the prevalence of stunting among infants and children under the age of five from 54.7 percent in 2000 to 37.4 percent in 2015 and the prevalence of undernourishment in Malawi has decreased from 24.7 percent to 17 percent between 2004and 2016(FAO, 2019;;UNICEF et al., 2020).Prevalence of vitamin A deficiency has also declined after the rollout of the vitamin A fortification programs, as revealed by the most recent Micronutrient Survey.Conducted in 2015-16, this survey did not detect vitamin A deficiency Changing Climate, Changing Food Consumption?1829 among adult women or men, but it revealed that 5 percent of pre-school and school-aged children are still suffering from this deficiency.These statistics suggest fortification of staple foods with vitamin A is an effective means to reduce vitamin and mineral deficiencies among the adult population.
Other food aid and social protection schemes include food assistance for refugees, the Malawi Social Action Fund (MASAF), which provides food or cash for work programs (Devereux, 2016;Sitko, Scognamillo, & Malevolti, 2021), the National Agriculture Policy (NAP) and the Agriculture Sector Wide Approach (ASWAp), which also focuses on food and nutrition security (CIAT and World Bank, 2018).The ASWAp identifies the importance of farm input subsidy programs to encourage fertilizer use to increase maize production, as well as promotion of droughtresistant crops, such as sweet potato and cassava (CIAT and World Bank, 2018).
Furthermore, in Malawi, property inheritance follows a matrilineal or patrilineal system, and whether the household head traces their lineage through their mother or father has important ramifications for land tenure (Asfaw & Maggio, 2018).Women from matrilineal communities are more likely to inherit land and a have significant role in agriculture and decision-making.Consequently, they are more likely to invest in technologies to increase their resilience to extreme weather events (Asfaw & Maggio, 2018).Thus, Malawi's matrilineal system may also have consequences for food and nutrition security.
In addition, a range of SSNs operate in Malawi, and they include free food and maize, food/cash/inputs for work, school feeding program, and targeted nutrition programs for children and mothers, supplementary feeding for malnourished children, scholarships/bursaries for education, direct cash transfers from government, development partners or NGOs.As well as public works programs operating under the Malawi Social Action Fund (MASAF).These SSNs are vital to building resilience and reducing a household's vulnerability to shocks (Devereux, 2016;d'Errico & Di Giuseppe, 2018).
However, despite these programs and some evidence of successful interventions, child stunting has increased to 39 percent between 2016 and 2018 and undernourishment has risen to 18.8 percent between 2017 and 2018 (FAO, 2019;UNICEF et al., 2020).The mentioned micronutrient survey conducted in 2015/16 also found zinc deficiency was the most prevalent micronutrient deficiency in Malawi, as 60 percent of preschool children and school-aged children, 63 percent of women, and 66 percent of men do not consume adequate amounts of zinc (National Statistical Office Malawi et al., 2017).Prevalence of iron deficiency was higher among certain population groups, as 22 percent of school-aged children and 15 percent of women are impacted (National Statistical Office Malawi et al., 2017).
Therefore, despite declining rates of undernourishment and reduced prevalence of specific MNDs, increases in undernourishment and child stunting indicate Malawi is not on track to meet the SDG2 of 'Zero Hunger' by 2030.

Empirical framework
In this section, we discuss the empirical framework used to capture the impact of rainfall and temperature shocks on macro-and micronutrient consumption.We do this by first outlining the estimation strategy used, followed by identification strategies.Regressions use data from 2628 households, which forms an unbalanced panel of 5986 observations.

Estimation strategy
The impact of weather shocks on macro-and micronutrient intakes has been estimated using the fixed effect estimator and is represented using the following equation: where subscripts h and r represent household and region level data, while t indicates that this variable varies through time.The dependent variable captures, in 11 distinct regressions indexed by superscript m, estimates of consumption of calories, carbohydrates, protein, and fat (i.e.macronutrients) as well as iron, zinc, vitamin A, vitamin C, vitamin B2, folate, and vitamin B12 per AME per day (i.e.micronutrients).A description of how nutrition variables were constructed is provided in the supplementary file.The inverse hyperbolic sine transformation of all nutritional variables is included as the dependent variables in each regression. 1 b, x, g and u represent the vector coefficients for each of the corresponding control variables.Control variables used in the econometric models of this study are described in appendix Table A1.Descriptive statistics for each variable are included in Table A2 while appendix Tables A3 and A4 show the mean apparent consumption among males and females present in the study sample.Vector x includes socio-economic variables such as age, gender, religion, and education of the household head, whether matrilineal regimes exist, household size, annual income, crop income as a ratio of total income, land holdings, access to microfinance, as well as household distance to the nearest market.This vector also includes variables determined through factor analysis.These denote indices for household adaptive capacity, assets, and access to basic services, and are included as a measure of a household's ability to withstand shocks.
Models incorporate the education of the household head, captured by a dummy variable indicating whether the head has attended school or not, as a measure of human capital.Increasing education is also linked with decreasing food insecurity (De Muro & Burchi, 2007).A dummy variable identifying whether the head of the household is female is included to account for the differences in food security between male and female-headed households (Tibesigwa et al., 2016).Household size, as captured by the number of people living in the household, is incorporated within analyses, due to the inverse relationship between household size and food security (Ahmed, Ying, Bashir, Abid, & Zulfiqar, 2017;Mango, Zamasiya, Makate, Nyikahadzoi, & Siziba, 2014).
Adjusted annual total income and crop income are important considerations for food security, as increasing incomes are linked with better-quality diets, in terms of calorie and micronutrient consumption (Van den Broeck, Mardulier, & Maertens, 2021).Monetary values are expressed in Malawian Kwacha (MK) and have been adjusted to account for inflation using the Consumer Price Index (CPI) for Malawi.Land holdings, while also indicating the wealth of the household, are also included due to the links with food security.For instance, households owning land can use their landholding as collateral to obtain credit, which in turn can be used to promote productivity and food security (Holden & Ghebru, 2016).In addition, increasing farm size is linked with higher yields, lower poverty, and reduced food insecurity (Nkomoki, Bavorov a, & Banout, 2019).
The variable access to finance is included to control for the negative association between a household's access to financial institutions and food security outcomes (Namayengo, Antonides, & Cecchi, 2018).The distance from the households to the nearest weekly market is included as a measure of market accessibility and the association between improved market access and increased food security (Ahmed et al., 2017;Kihiu & Amuakwa-Mensah, 2021).Variables are included to denote whether the household is part of a matrilineal or patrilineal community as an indicator of a household's resilience to extreme weather events (Asfaw & Maggio, 2018).
Adaptive capacity considers the number of social safety nets (SSN), ownership of property, and crop and livestock diversity to capture the household's ability to adapt to shocks (Murendo et al., 2020).
The asset index assesses the number of productive and non-productive assets, including mobile phone ownership, owned by the household.Binary variables differentiating between Changing Climate, Changing Food Consumption?1831 households whose roof, outer walls, and floors are constructed using durable materials, and healthy cooking fuels, as well as the number of household members sleeping per room, are also included in this index.Access to basic services considers whether households have access to electricity, safe drinking water, and improved toilet amenities (Murendo et al., 2020).
s captures agro-ecological zones (AEZs) where each household is located.These are not time invariant for all households, due to the relocation of some families between survey waves.Year fixed effects, as represented by p, have been included to account for aggregate trends in the dependent variables.e h is the household-specific error term.d is a vector of weather shocks: negative rainfall, positive rainfall, negative temperature, and positive temperature shocks.negrain rt ¼ 1 if monthly precipitation for the region ! 1 standard deviations SD ð Þ below the 30 year average prior to the date of data collection posrain rt ¼ 1 if monthly precipitation for the region ! 1 SD above the 30 year average prior to the date of data collection negtemp rt ¼ 1 if average monthly temperatures for the region ! 1 SD below the 30 year average prior to the date of data collection postemp rt ¼ 1 if average monthly temperatures for the region ! 1 SD above the 30 year average prior to the date of data collection See supplementary file for an additional description on how shock variables were constructed.
The panel structure of our dataset allows the use of the fixed effect estimator to estimate the effects of shocks on food and nutrition security.Fixed effects models capture within variation, and robust standard errors clustered by household control for cluster correlation and heteroskedasticity (Cameron & Trivedi, 2010;Cameron and Miller, 2015).Failure to control for error correlation leads to underestimation of standard errors and an overestimation of statistical significance (Cameron & Trivedi, 2010).Sampling weights have not been used and thus, empirical findings are not representative of the entire population in Malawi.

Identification strategy
Our identifying assumption is that d, a vector of exogenous rainfall and temperature variables, impacts the amount of macro-and micronutrients, as well as calories consumed.Estimated intakes of macro-and micronutrients per adult equivalent per day in our empirical framework are used as a proxy to measure food utilisation, in terms of calorific sufficiency and nutritional value of foods consumed (FAO et al., 2018).To make any statements regarding the causality of our variables of interest and food security, our identification strategy must adequately address the three sources of statistical endogeneity: unobserved heterogeneity, measurement error, and reverse causality.
Using fixed-effects controls for time-invariant unobserved heterogeneity present in the data and limits the source of bias to the time-varying component of the model (Collischon & Eberl, 2020).That is, fixed effect models permit time-invariant variables to be correlated with time-invariant components of the error term while remaining uncorrelated with the idiosyncratic error term (Cameron & Trivedi, 2010).
Our estimates of the amounts of macro-and micronutrients consumed per adult equivalent per day are prone to measurement error, as they are based on self-reported measures of intake and individual intakes are constructed from household-level data.However, this form of measurement error is present within individuals, and given that fixed effect estimations use withinindividual variation for estimations, individual fixed effects will remove this form of error (Collischon & Eberl, 2020).
Use of the fixed effect estimator does not remove bias due to reverse causality.However, as our variables of interest (rainfall, and temperature shocks) are exogenous, i.e. determined outside of the model, consumption of macro-and micronutrients does not influence monthly precipitation or temperatures.

Robustness checks
To investigate the robustness of our results we: 1. adjust the reported p-values from the main regressions using the Holm-Bonferroni (HB) correction procedure to control for multiple hypothesis testing (MHT), 2. apply fixed effect estimations with our four key shock variables of interest without additional controls, 3. use a general-to-specific (GETS) approach to remove the explanatory variables from the fully specified model based upon their relevance and statistical power in explaining the main dependent variables, i.e. the shock variables (Clarke, 2013).The GETS approach is used to understand which explanatory variables are driving the empirical results of this study, 4. test whether attrition is missing at random (MAR) or not missing at random (NMAR) using attrition probits (Fitzgerald, Gottschalk, & Moffitt, 1998) and, 5. control for possible attrition bias by comparing the estimation results of balanced and unbalanced panels (Nijman & Verbeek, 1992).
Many of the regression coefficients reflecting the impact of shocks on our nutrition outcomes are statistically significant.This is especially true for negative rainfall and positive temperature shocks.However, the statistical significance observed may be a result of MHT.To control for MHT, we adjust the reported p-values from the main regressions using the HB correction procedure.The HB procedure is selected as it is a more powerful and less conservative method of correcting for multiplicity than other methods such as the Bonferroni procedure (Holm, 1979).
Next, we run our main shock variables of interest with and without control variables to determine omission of controls could change the outcomes noted in Tables 1 and 2. Lastly, to determine whether attrition of households results in data that is not missing at random and is related to observable characteristics of the households we conduct an attrition probit model.Attrition probit models use baseline values (2010) and a dummy-dependent variable of attrition (1 ¼ households dropped out after the first wave, 0 ¼ households remaining across all three waves) is used in this model to determine whether attrition on observables could lead to biased estimates (Baulch and Quisumbing, 2011).Finally, to determine whether possible attrition biases does not alter the conclusions from our fixed effect estimations we run our main regressions with and without attritors.i.e. we run our estimations using balanced panels (households that remained) with our full unbalanced panel dataset to determine whether the unbalanced nature of our dataset is responsible for results.Results from our robustness checks are included in the supplementary file Tables S.3-S.7.
Changing Climate, Changing Food Consumption?1833

Results and discussion
This section presents our empirical findings.Appendix Table A2 outlines descriptive statistics for the dependent and explanatory variables used in regressions.The mean apparent consumption of each macro-and micronutrient among males and females present in the study sample is also presented in appendix Tables A3 and A4.Table 1 outlines the impact of rainfall and temperature shocks on macronutrient consumption, while Table 2 presents the impact on micronutrient consumption.Regression coefficients are reported as percentages and reflect a one-unit change in the amount of macro-and micronutrients consumed.Coefficients have also been converted to interpretable values by estimating the percentage change using the mean value of each nutrition outcome, as stated in Table A2.The calorie (kcal), gram (g) or microgram (lg) equivalent of each percentage change is reported in brackets.Households experiencing monthly rainfall falling one or more standard deviations below the long-term average at least once in the previous year consume 7 percent less calories per adult Changing Climate, Changing Food Consumption?1835 equivalent per day (i.e.minus 225 kcal).These negative rainfall shocks result in intakes of carbohydrate, protein, and fat to decline by 7 percent (35.0 g), 9 percent (9.3 g) and 8 percent (5.6 g), respectively.Amounts of iron, zinc, vitamin A, C, B2, folate, and B12 decreased by 9 percent (9.4 mg), 9 percent (1.9 mg), 8 percent (42.5 lg), 13 percent (20.8 mg), 4 percent (0.05 mg), 5 percent (23.81 lg), and 10 percent (0.7 lg), respectively.Results of the robustness checks (supplementary file Tables S3-S7) confirm negative rainfall shocks cause a reduction in the estimated consumption of all nutrition variables mentioned above, except for vitamin A, folate and B12.Findings are supported by Carpena, 2019, who also found droughts negatively impact household nutrition in India, with median intakes of calories, protein and fat falling by 1.4 percent when droughts occur.Albeit we find the decline in estimated consumption of macronutrients to be much greater than Carpena, 2019, which may suggest households in Malawi have greater vulnerability to drought than other regions of the world.Asfaw et al., 2017 andTibesigwa et al., 2016, further demonstrate how food expenditure and total calories consumption decline in response to negative rainfall shocks.Ajefu & Abiona, 2020, using the FCS, as an indicator of nutritional quality of the household diet, also find drought results in the FCS declining by 35.8 percent.
Inversely, households exposed to monthly rainfall one or more standard deviations above the 30-year average causes consumption of calories to increase by 4 percent (129 kcal).Positive rainfall shocks also result in the amounts of carbohydrate and protein consumed per adult equivalent per day to increase by 6 percent (30.0 g) and 4 percent (4.1 g).This contrasts with other studies, such as Abegaz, 2017, Akampumuza & Matsuda, 2017, Mekonnen et al., 2021, Akampumuza et al., 2020, which find rainfall shocks negatively impact food security.Furthermore, our robustness checks (supplementary file Tables S3-S7) do not support the assertion that periods of positive rainfall cause an increase in estimated consumption of macronutrients.
Monthly temperatures rising one or more standard deviations above the 30-year average at least once in the previous year caused a reduction in calorie consumption by 14 percent (450 kcal).In addition, positive (warmer) temperature shocks resulted in a decline in the consumption of carbohydrate by 13 percent (65.1 g), protein by 15 percent (15.6 g) and fat by 9 percent (6.3 g) per adult male equivalent per day.Amounts of iron, zinc, vitamin A, vitamin B2, folate, and vitamin B12 also declined by 26 percent (27.1 mg), 15 percent (3.2 mg), 29 percent (154.2 lg), 13 percent (0.1 mg), 21 percent (100.0 lg), and 41 percent (2.7 lg), respectively.All results, except estimated consumption of fat, are robust to both MHT and attrition bias (supplementary file Tables S3-S7).This is consistent with the literature, in which increasing temperatures have been linked with reduced crop yields and food security (Asfaw & Maggio, 2018;Gao & Mills, 2018;Mekonnen et al., 2021;Niles et al., 2021;Niles & Salerno, 2018).
Negative (colder) temperature shocks, i.e. monthly temperature falling one or more standard deviations below the long-run average, were not statistically associated with macronutrient consumption but increased iron consumption per adult equivalent per day by 12 percent (12.5 mg).This result is not robust to the effects of MHT and attrition bias (supplementary file Tables S3-S7).

Policy implications
Findings of regressions estimated using the fixed effect estimator and findings from our robustness checks outline the detrimental impact of negative rainfall and positive temperature shocks on both macro-and micronutrient consumption.Our results show rainfall falling at least 1 SD below the long-term average and temperatures rising at least 1 SD above the long-term average result in a decline in the number of calories, and amounts of carbohydrates, protein, iron, zinc, and vitamin B2 consumed.In addition, declining rainfall results in a reduction in fat and vitamin C consumption, whereas, rising temperatures causes a decline in the consumption of A, folate, and B12.These results are robust to both MHT and attrition bias.The results of this study highlight Malawi's vulnerability to weather events and the detrimental effects of changing rainfall patterns and increased temperatures on nutrition and food security.As the average annual temperature is set to increase by 1.9-2.5 C by 2050, and rainfall to decline by 0-4 percent, food insecurity is likely to intensify (Warnatzsch, Reay, Camardo Leggieri, & Battilani, 2020).
Efforts to mitigate and adapt agriculture to climate change will therefore be vital to improve food and nutrition security.The country has already undertaken some steps in this direction.For example, Climate smart agriculture (CSA) practices, such as conservation agriculture, using improved maize varieties, as well as soil and water conservation (SWC) have already proven to increase maize yields by 53 percent in Malawi despite the occurrence of drought (Amadu, McNamara, & Miller, 2020).However, limited resources and financial barriers constrain smallholder adoption of CSA (Amadu et al., 2020).Malawi's National Climate Change Management Policy (NCCMP) and the National Climate Change Investment Plan (NCCIP) constitute important instruments to coordinate and finance climate change adaptation (CIAT and World Bank, 2018).
Distribution of food aid and social protection programs, such as those discussed in Section 2, are also of vital significance.Previous research concluded that smallholders receiving food aid are more likely to adopt CSA techniques, such as SWC structures, legume intercropping, using organic fertilizer and accumulating livestock (Sitko et al., 2021).Therefore, as our empirical findings highlight Malawi's vulnerability to adverse weather events, programs and policies aiming to enhance food security will be vital to promote resilience in a changing climate.
Reduced rainfall and increased temperatures may contribute to exacerbated food insecurity in Malawi by worsening micronutrient deficiencies, such as zinc, which was the most prevalent deficiency in Malawi 2015/16.Vitamin A deficiency was not detected in 2015/16, however, as temperatures continue to increase intakes will decline, consequently undoing the gains made towards reducing deficiencies of vitamin A. Weather and climatic extremes will also aggravate deficiencies of iron, vitamin B2 and folate, which are also of concern in Malawi, as evidenced from micronutrient surveys conducted between 2000 and 2010.
Deficiencies of vitamin B2, folate, zinc, and vitamin A are also a concern as estimated individual intakes across age and sex groups, as shown in appendix Tables A3 and A4, indicate not all population groups were consuming adequate amounts.Consumption of zinc is of particular concern, as diet staples in Malawi are cereal based and compounds, such as phytates, present in these foods will limit absorption and thus bioavailability of zinc (NIH ODS, 2021).
Other chief dietary sources of zinc, consumed in Malawi as in several other African countries, include milk and dairy products, red meats, shellfish, legumes, and nuts (Lokuruka, 2012).However, staple foods in countries, such as Malawi, are calorie rich but micronutrient poor, and production of nutrient dense foods, such as vegetables, pulses, and animal products are not produced in the same quantity as cereal-based staples (Bouis & Saltzman, 2017).Therefore, consumption of zinc from non-cereal sources is likely to be limited.The high cost of non-staple foods further prevents poorer households from accessing higher-quality diets (Bouis & Saltzman, 2017), and thus restricting the nutritional quality of diets.
To the author's knowledge, in Malawi there are no food fortification programs focusing on the addition of zinc or B vitamins to staple foods.Therefore, as micronutrient supplementation of foods with vitamin A have proven an effective strategy in reducing vitamin A deficiency in Malawi, programs supporting the fortification of foods with vitamin B2, folate, and zinc should also be considered.We focus on vitamin B2, folate, and zinc, as apparent consumption of vitamin B2 was inadequate for all individuals within this sample, while only 34 percent and 53 percent, were estimated to be consuming the recommended nutrient intake (RNI) for folate and Changing Climate, Changing Food Consumption?1837 zinc, respectively (For additional commentary see appendices).Changes in rainfall temperatures were also associated with reduced consumption of these micronutrients.
Biofortification is also an effective means of increasing the nutritional content of crops, using plant breeding, transgenic, and agronomic practices (Bouis & Saltzman, 2017).This method has fewer costs than food fortification or supplementation schemes, has minimal marginal costs to farmers growing biofortified crops, and can effectively target rural populations (Bouis & Saltzman, 2017).

Conclusions
Using estimated intakes of macro-and micronutrients to measure the nutritional impacts of weather shocks, we have identified food and nutrition security in Malawi to be vulnerable to fluctuations in rainfall and temperatures.With climate change expecting to cause more frequent and large fluctuations in rainfall and temperature, micronutrient deficiencies in Malawi are also expected to become more prevalent.
Our findings should be interpreted with caution, bearing in mind the limitations of our data and approach, which we will now describe.Food consumption data is based on self-reported accounts of the amount of food consumed over 7 days within the household.Collection of data via recall, is subject to response and recall biases, and may lead to an overestimation of food consumption and recall periods cannot capture seasonality of diets or food consumed infrequently (Rosenman, Tennekoon, & Hill, 2011;Ecker & Qaim, 2011).However, a short recall period of 7-days reduces the probability of recall bias, as respondents can more accurately remember foods consumed over one week (Ecker & Qaim, 2011).
Direct dietary assessment methods, in comparison to food consumption data, are more accurate indicators of actual food consumption, as they consider the impact of intra-household food allocation (Zezza, Carletto, Fiedler, Gennari, & Jolliffe, 2017).Estimated individual consumption of nutrients should also be compared to RNIs and EARs with caution, as recommended intakes of nutrients will differ by country and across population groups (Bermudez et al., 2012).Bioavailability of nutrients is also country specific, for example, countries obtaining a large proportion of iron from plant-based sources, such as Malawi, have lower bioavailability (Bermudez et al., 2012).Thus, recommended intakes need to be adjusted accordingly.Nevertheless, in the context of developing countries, where more precise methods to collect nutrition data cannot be easily facilitated, HCES are a valuable source of information, in which to assess food and nutrition security.
Being the first of its kind in SSA, we hope our study encourages more research in this area and assists policymakers provide more targeted nutritional support to vulnerable communities, as well as achieve Zero Hunger by 2030.

Note
1. Due to the presence of variables with observations containing zero values, the inverse hyperbolic sine (arcsinh) transformation is taken to approximate the natural logarithm, while ensuring observations with zero values are retained (Bellemare & Wichman, 2020).The arcsinh of a variable (x) is calculated using the following procedure: arcsinh Total annual income from crops sales, wages, allowances, gratuities, gifts, savings, pension, rental sales, inheritance, and gambling.Ratio of crop income to total income Share of annual total income from crop sales.

Landholdings
Land area owned by the household (ha).

Access to microfinance
Dummy variable indicating whether households within a community have access to savings or credit cooperatives (1 ¼ yes, 0¼ otherwise).Adaptive Capacity Index Index generated using factor analysis and is based on the number of social safety nets, variety of crops and livestock produced and property ownership.Asset Index Index constructed using factor analysis and is based on the number of productive and non-productive assets and mobile phones owned, people sleeping per room and whether the roof, outer walls and floors of the household are made using durable materials and uses healthy cooking fuels.Access to Basic Services Index Index created using factor analysis and is based on whether the household has access to electricity, safe drinking water and improved toilet facilities.
Changing Climate, Changing Food Consumption?1843 temperatures and negative rainfall, respectively, while 22 percent were exposed to lower-thanaverage temperatures.Aggregate estimates of macro-and micronutrient consumption at the household level, presented in Table 2, do not account for differing age and sex characteristics of individuals present within the household.Therefore, to better disentangle consumption among each age and sex group, appendix tables A.3 and A.4 show the mean apparent consumption among males and females present in the study sample.Due to the limitations of available data, estimates of food consumption are indicative of apparent consumption for each age and sex group and not actual consumption.
A comparison to recommended nutrient intakes (RNI) suggests daily apparent intakes of carbohydrates, protein, vitamin C, vitamin B12 and iron are sufficient for all population groups.However, crude estimates of vitamin B2 consumption, suggest, that no age and sex group met the RNI.Fifty-one percent percent of apparent intakes met recommended intakes of calories and 43 percent of groups consumed enough fat.Estimated daily intakes of zinc and folate are adequate in 53 percent and 34 percent, respectively, of the study sample.About 74 percent of estimated individual intakes for vitamin A are above the estimated mean requirement (EAR).

Table 1 .
Estimation results determining the impact of rainfall and temperature shocks on household calorie and macronutrient consumption per adult equivalent per day.

Table 2 .
Estimation results determining the impact of rainfall and temperature shocks on household micronutrient consumption per adult equivalent per day.

Table A2 .
Summary descriptive statistics for macro-and micronutrient consumption per adult equivalent per day, shock variables, and household socioeconomic characteristics.
Notes: Observations: 5,986.Monetary values are adjusted to account for inflation using the Consumer Price Index for Malawi.