WPS8442 Policy Research Working Paper 8442 Food Insecurity and Rising Food Prices What Do We Learn from Experiential Measures? Dean Jolliffe Ilana Seff Alejandro de la Fuente Poverty and Equity Global Practice & Development Data Group May 2018 Policy Research Working Paper 8442 Abstract Throughout many countries in the world, the measure- insecurity. This finding controls for individual-level fixed ment of food security currently includes accounting for effects and changes in the economic well-being of the indi- the importance of perception and anxiety about meeting vidual. A particularly revealing finding of the importance basic food needs. Using panel data from Malawi, this of accounting for anxiety in assessing food insecurity is that paper shows that worrying about food security is linked individuals report a significant increase in experiences of to self-reports of having experienced food insecurity, and food insecurity in the presence of rapidly rising food prices the analysis provides evidence that rapidly rising food even when dietary diversity and caloric intake is stable. prices are a source of the anxiety and experiences of food This paper is a product of the Poverty and Equity Global Practice and the Development Data Goup. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/research. The authors may be contacted at at djolliffe@worldbank.org and adelafuente@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Food Insecurity and Rising Food Prices: What Do We Learn from Experiential Measures? Dean Jolliffe, Ilana Seff and Alejandro de la Fuente* Keywords: Food security, food prices, anxiety, stability, LSMS, Malawi. JEL Codes: Q18, I30, I32, E31 * Jolliffe (djolliffe@worldbank.org) is a Lead Economist in the World Bank’s Development Data Group, Seff is a doctoral candidate at Columbia University in the Population and Family Health Department, de la Fuente is a Senior Economist in the World Bank’s Poverty and Equity Global Practice. Jolliffe is also a Research Fellow with the Institute for the Study of Labor (IZA) in Bonn and a Fellow of the Global Labor Organization (GLO). Acknowledgements: This paper builds on analysis done for the Malawi Poverty Assessment report (World Bank, 2016). The authors wish to thank Jose Cuesta, Sailesh Tiwari, and Pablo Fajnzylber for comments and guidance. The analysis in this paper is based on the publicly available data from two rounds of Malawi’s Integrated Household Survey. These have received technical and financial assistance from the Living Standards Measurement Study- Integrated Surveys on Agriculture (LSMS-ISA) project. See http://go.worldbank.org/ZIWEL8UHQ0 for details on LSMS-ISA and links for downloading the files. The findings, interpretations, and conclusions of this paper are those of the authors and should not be attributed to the World Bank Group or its member countries. 1. Introduction One of the initial definitions of food security from the 1970s was based on the narrow concept of whether there was sufficient production of food in a nation to support the population.1 Influenced in part by Sen (1981), and in recognition of the obvious inadequacies of this view, the concept was expanded over the next two decades to include considering the distribution of the aggregate food supply, in particular whether individuals have resources to gain adequate access to food (FAO, 1983). The concept of food security was further broadened to account for whether individuals could properly utilize the nutrients in food (e.g. due to factors such as proper hygiene)2 and, also whether the availability, access and utilization of food was stable over time (FAO, 1996). As the conceptualization of food security broadened, so too did the measures. Much of the empirical analysis of food security has focused on measures of food supply (e.g. food balance sheets)3 and food access (e.g. calories and dietary diversity) both at a given point in time and over time.4 Analysis based on change over time in the measures of food availability and access have been interpreted as measures of stability of food security.5 Stability though, 1 First expressed at the 1974 World Food Summit as summarized by Ram et al. (1975). 2The concept of food utilization includes concerns of intra-household distribution of food, nutritional value of food, and nutrient absorption (Jones et al. 2013; Barrett 2010). Expanding the definition to include utilization acknowledges that two households, consuming similar calories per capita, can look substantially different in terms of food security status. 3 For an overview, including discussion of food availability measures, see Jones et al. (2013). 4 World Bank (1986) was one of the earlier reports to measure the stability of availability and access to food (as measured by food supply and caloric intake) over time. 5 First differences in these measures have also been used to separate chronic from transitory food insecurity. 2 has largely been interpreted as stability in realized outcomes such as food production, caloric intake, and dietary diversity.6 In this paper, we examine a broader conceptualization of stability that treats the absence of anxiety about insecurity, and perceptions of security, as attributes of the stability of food security. This notion of insecurity as being linked to anxiety and perceptions is neither new nor innovative. In the U.S., the distinction between someone who is identified as having “high food security” status and “marginal food security” status is most typically identified by whether the person reported experiencing anxiety over food sufficiency (or expected food sufficiency).7 The Food Insecurity and Experience Scale (FIES) is now the endorsed SDG indicator 2.1.2 and is used by the Food and Agriculture Organization (FAO) in nearly 150 countries covering 90 percent of the world’s population, and the first question in this instrument is whether the respondent worried whether they would have sufficient food to eat. What we aim to illustrate in this paper is that stated anxiety about food security is closely linked to perceptions of food insecurity status (suggesting that worrying about food security can itself be viewed as insecurity or as reflecting a lack of certainty or stability in food security status). This supplements the evidence that anxiety is a useful partial indicator of food insecurity. The analysis in this paper also reveals that stated experiences of food insecurity need not be associated with realized deprivations in food access as measured by 6 See for example, D’Souza and Jolliffe (2014), Jolliffe, Sharif, Gimenez (2013), Schmidhueber and Tubiello (2007), and Wheeler and von Braun (2013) as examples of examining the time stability of food availability and access as measured by food production, calories and dietary diversity. 7 In 2006, The U.S. Department of Agriculture revised their labels for different states of food security (e.g. high, marginal, low and very low food security) based on recommendation of the Committee on National Statistics (National Research Council, 2006) to distinguish the physiological state of hunger from other indicators of food security. 3 calories and dietary diversity.8 Our inference from this finding is that experiential measures contain an important signal about food insecurity status that is not conveyed in measures of access and availability (such as calories and dietary diversity). Another inference from our finding is that rising food prices, regardless of whether they have direct adverse effects on caloric intake or diet quality, lead to increased anxiety about meeting food needs. The next section of this paper proceeds by discussing the panel data used in this analysis and a descriptive profile of food security status in Malawi. Section 3 discusses the regression model used to examine the connection between rising food prices and the experiential measures of food security status, and section 4 provides some concluding discussion. 2. Data and Country Context 2.1. Data This paper makes use of two waves of panel survey data, representative at national and subnational (rural/urban) levels. The first wave was carried out from March to November of 2010. This wave is a sub-sample of the Third Integrated Household Survey (IHS3), which is part of the Living Standards Measurement Study - Integrated Surveys on Agriculture (LSMS- ISA) project in Malawi. The second wave was fielded from March to November of 2013 as part of the Integrated Household Panel Survey (IHPS). The panel element of the survey was designed such that individuals (not households) were followed between waves. Therefore, the IHPS sample includes 4,000 households that can be traced back to 3,104 baseline households 8 This finding aligns with the literature on armed conflict, in which it is well recognized that various measures of safety and security do not correlate well with perceptions of safety and security, and are sensitive to the finding that statistical improvements in safety and security may not directly improve perceptions of safety and security (Rigterink, 2015). 4 from IHS3. After restricting to individuals for which all variables are non-missing, our analysis is carried out on a balanced sample of 11,410 individuals. We construct three measures of food insecurity that are typically interpreted as being measures of food access: daily per capita caloric intake, undernourishment, and the food consumption score (FCS),9 which captures dietary diversity.10 The data on caloric intake come from a 7-day recall module on household-level food consumption; undernourishment is a binary outcome indicating whether an individual consumes less than 2,100 calories/day. In addition to standard survey modules on household characteristics, assets, income, and consumption, the survey contains a module on experiential assessment of food security. This module includes several questions which ask whether households have worried about having sufficient food (e.g. In the past 7 days, did you worry that your household would not have enough food?), and whether they have experienced food insecurity (e.g. In the last 12 months, have you been faced with a situation when you did not have enough food to feed the household?). To create our core panel data, we merged in an external food price data set that forms part of the Agricultural Market Information System (AMIS) collected by the Planning Department of Malawi’s Ministry of Agriculture and Food Security (MoAFS). Price data were collected weekly from 72 markets located in Malawi’s 26 districts and aggregated to the 9 The FCS considers dietary diversity, frequency of food groups consumed, and relative nutritional value of each food group consumed in the last 7 days (FAO 2008). The potential score range is 140. The higher the FCS, the better the diversity of the household’s food intake and, subsequently, the better the quality of the members’ diets. In categorizing households, those with scores less than 21 are determined to have “poor” dietary diversity; those falling between 21 and 35 are said to be in a borderline range; and those with scores above 35 are considered to have acceptable levels of dietary diversity (Weismann and others 2009). 10Dietary diversity is a concept that captures the degree to which households or individuals consume a variety of foods. Along with caloric intake and undernourishment, dietary diversity can help generate a greater understanding of a population’s food insecurity (Ruel 2003). Research also suggests that dietary diversity can be used as a proxy for household income, household-level access to food, macronutrient, and micronutrient intake (see Hatloy et al. 2000; Anzid et al. 2009; Hoddinott & Yohannes, 2002; Rah et al. 2010). 5 monthly level. We focus on maize prices, because maize is the predominant crop (grown by approximately 96 percent of rural households) in Malawi. The IHS3 panel and IHPS households are geo-referenced allowing us to map households to food prices found at the closest markets. In this way, we can measure changes in maize prices that households faced in the 12 months leading up to their survey date. 2.2 Context Malawi is among the poorest countries in the world, with limited resources and an economy relying heavily on single-crop maize agriculture. Maize is by far the most important staple crop, both in terms of the number of farm households cultivating this crop and in terms of total area harvested in the country (Dabalen and others, 2017). Maize also accounts for more than 50 percent of the daily calorie intake in Malawi (Minot, 2010). The country’s population is estimated to be 17.6 million in 2015, and about 80 percent of the people live in rural areas. GNI per capita is US$ 350 as of 201511 and 71 percent of the people are estimated to live on less than US$ 1.90 per day (2011 PPP prices) as of 2010.12 Malawi is relatively small in land size and densely populated. This, combined with a high population growth rate, has created significant pressure on available land for smallholder farming and on the environment and natural resource base. Such precarious conditions are compounded by the country’s high and recurrent exposure to drought, floods and food price shocks. Compared to other countries in Sub- Saharan Africa, food prices in Malawi are highly variable and subject to spikes, with maize 11 This estimate is from the World Bank’s World Development Indicators online database (databank.worldbank.org/WDI), and is expressed in nominal 2015 US$, converted from local currency using the Atlas method (based on market exchange rates). 12 This estimate is from PovcalNet, an online tool provided by the World Bank for global poverty measurement, see http://iresearch.worldbank.org/PovcalNet. 6 prices being the most variable (see Kaminski and others, 2014). In addition to greater volatility, the annual rate of inflation between 2000 and 2014 in Malawi (15.4 percent) was 2.5 times higher than the Sub-Saharan Africa regional average (6.1 percent) (Dabalen and others, 2017). Figure 1 presents monthly price data from 2009 to 2013 for staple foods from 72 markets across Malawi. Food prices in Malawi increased significantly between late 2011 and early 2013, which falls within the two waves of survey fieldwork. The price of maize increased steadily over this period, but also experienced a particularly sharp spike during the end of 2012. From 2009 through most of 2011, maize prices were stable, hovering around 40 to 50 kwacha per kilogram. In 2012, the price of maize began to climb slowly, but then in the short time between December and March of 2013, the price more than doubled, jumping from 65 kwacha per kilogram to 136 kwacha per kilogram. Other food staples experienced similarly large increases of prices between the waves of data collection. For example, from July 2011 to March 2013, the price of beans more than doubled, from 165 kwacha per kilogram to a peak near 380 kwacha per kilogram. 7 Figure 1. Rise in Median Food Prices in Malawi, 2009–13 (Kwacha) 400 350 300 <‐‐ IHS3 fieldwork ‐‐> Malawian Kwacha 250 200 <‐‐IHPS 150 fieldwork ‐‐> 100 50 0 Jul‐09 Nov‐09 Jan‐10 Jul‐10 Nov‐10 Jan‐11 Jul‐11 Nov‐11 Jan‐12 Jul‐12 Nov‐12 Jan‐13 Jul‐13 Mar‐09 May‐09 Mar‐10 May‐10 Mar‐11 May‐11 Mar‐12 May‐12 Mar‐13 Sep‐09 Sep‐10 Sep‐11 Sep‐12 May‐13 Sep‐13 Cassave prices Maize prices Rice prices Bean prices Soya prices Source: Authors based on the Agricultural Market Information System (AMIS) from the Ministry of Agriculture and Food Security. Nationally, average per capita daily caloric intake increased slightly from 2,256 to 2,371 between 2010 and 2013 (Table 1). The prevalence of undernourishment (as measured by calorie deficiency) also dropped slightly from 53 to 50 percent during this period. We find similar improvements in individuals’ dietary diversity between 2010 and 2013. Table 2 shows there was a slight increase in the value of the FCS from 48.5 in 2010 to 50.5 in 2013. The data also reveal a substantial improvement in the distribution of FCS between waves. At the national level, the prevalence of ‘poor’ or ‘borderline’ FCS decreased from 27 to 19 percent; this same statistically significant trend is found separately in urban and rural areas. The observational measures of caloric intake and diet quality suggest that food security status improved overall (at the national level) between 2010 and 2013 with no evidence of deterioration in either urban or rural areas. 8 Table 1. Trends in caloric intake and undernourishment Round 1 – 2010 Round 2- 2013 Panel A: Daily per capita caloric intake National 2,256 2,371** (39) (45) Urban 2,527 2,570 (104) (127) Rural 2,205 2,334** (42) (48) Panel B: Undernourished prevalence National 0.532 0.497* (0.017) (0.018) Urban 0.431 0.442 (0.029) (0.047) Rural 0.551 0.507* (0.019) (0.020) Change between two waves is statistically significant at *p<0.10, ** p<0.05, ***p<0.01. Note: Standard errors in parentheses correct for clustering and stratification. The descriptive statistics have been estimated from the cross-sectional samples corresponding to our final balanced sample. The national sample size includes 11,410 individuals in each round; 2,967 and 2,882 individuals are included in the urban samples in rounds 1 and 2, respectively, and 8,443 and 8,528 individuals are included in the rural samples in rounds 1 and 2, respectively. Table 2. Trends in dietary diversity Round 1 – 2010 Round 2- 2013 (SE) (SE) Panel A: Raw food security score (FCS) National 48.46 50.45*** (0.90) (0.83) Urban 62.48 63.51 (2.17) (1.31) Rural 45.84 47.97*** (0.85) (0.80) Panel B: Poor or borderline FCS National 0.268 0.192*** (0.017) (0.018) Urban 0.076 0.035*** (0.014) (0.009) Rural 0.304 0.222*** (0.018) (0.020) Change between two waves is statistically significant at *p<0.10, ** p<0.05, ***p<0.01. Note: Standard errors in parentheses correct for clustering and stratification. The descriptive statistics have been estimated from the cross-sectional samples corresponding to our final balanced sample. The national sample size includes 11,410 individuals in each round; 2,967 and 2,882 individuals are included in the urban samples in rounds 1 and 2, respectively, and 8,443 and 8,528 individuals are included in the rural samples in rounds 1 and 2, respectively. 9 The absence of an impact from high food prices on standard observational measures of food security does not necessarily mean the absence of a problem. While most of the population are net-buyers of food, there is a significant mass of people who are net-sellers (Darko et al., 2018); the finding that, on average, diet quantity and quality are improving is feasible. Moreover, even for those people consuming more food than they produce or sell, the extent to which food price increases may affect people’s access to food depends on a larger number of factors beyond the distribution of net sellers and net buyers of food staples. These include the rate at which national prices are passed through to local economies, the specific commodities for which prices increase, the ability of consumers to substitute into other less expensive food items, the coping strategies available to households, and the policy responses by governments. For many reasons, increasing food prices need not harm access to food. In contrast to the modest improvements in calories and dietary diversity between 2010 and 2013, trends in experiential measures portray a picture of significant decline in food security. The proportion of the population that indicated they had worried in the past seven days about having adequate food increased from 28 percent to 38 percent between the waves (Appendix Table 1). When asked whether they experienced food insecurity at least once in the previous 12 months (Table 3, Panel A), the estimated proportion of the population answering affirmatively increased by 30 percent, going from 51 percent in 2010 to 66 percent in 2013. This increase is found in both rural and urban areas, and is particularly pronounced in urban areas, where the proportion of households reporting food insecurity increased more than 60 percent. Similarly, the reported average count of months that people experienced food insecurity increased from 1.5 to 2.2 months at the national level, while the duration more than doubled in urban areas. The increase in both prevalence and duration of experiential food insecurity from 2010 to 2013 suggests that, despite modest improvement in measures of access 10 to food, significantly more people state that they are experiencing bouts of food insecurity. Table 3. Trends in experiential measures of food security Round 1 – 2010 Round 2 – 2013 Panel A: Self-reported food insecurity, last 12 months National 0.510 0.658*** (0.018) (0.017) Urban 0.326 0.534*** (0.031) (0.035) Rural 0.545 0.682*** (0.020) (0.018) Panel B: Number of months food insecure, last 12 months National 1.50 2.15*** (0.08) (0.09) Urban 0.71 1.45*** (0.07) (0.11) Rural 1.65 2.29*** (0.08) (0.10) Change between two waves is statistically significant at *p<0.10, ** p<0.05, ***p<0.01. Note: Standard errors in parentheses correct for clustering and stratification. The descriptive statistics have been estimated from the cross-sectional samples corresponding to our final balanced sample. The national sample size includes 11,410 individuals in each round; 2,967 and 2,882 individuals are included in the urban samples in rounds 1 and 2, respectively, and 8,443 and 8,528 individuals are included in the rural samples in rounds 1 and 2, respectively. 3. Methodology and Results 3.1. Model – explaining the puzzle The descriptive statistics indicate that despite escalating food prices in late 2012-2013, Malawians appear to have improving levels of food security as measured by availability, access, and utilization over the study period. Yet, despite this, they report both worrying more about having sufficient food and state that they are experiencing food insecurity more frequently. The next section explores in more detail the association between food price surges and the increased worries and anxiety of households to meet their food needs. The hypothesis examined is that the dramatic increases in food prices led to increased anxiety about the stability of food security, despite no change in the actual levels of consumption. 11 This surge in experiential food insecurity was accompanied by an increase in self- reported exposure to food price shocks. Interviewed households were asked to indicate the presence of food price shocks in the 12 months leading up to the survey. While only 25 percent of households in 2010 reported facing a significant increase in food prices during the previous 12 months, a staggering 83 percent of households reported facing this shock in 2013. In the urban South, approximately 93 percent of households reported experienced a food price shock, more than 6 times the percentage of households in 2010. The descriptive statistics create a compelling puzzle, but they are an incomplete evidence base. One concern is that changes in means of calories and dietary diversity might be driven by skewness in these distributions and may not be reflective of what the majority of the people are experiencing. For example, the mean of per capita calories is equal to the caloric intake at the 68th percentile of the calorie distribution in wave 1; FCS mean is at the 56th percentile in both waves. Another concern is simply that these are bivariate comparisons and it is possible that there are important confounding factors outside of food prices that are resulting in more people experiencing food insecurity. To address these concerns, we leverage the panel aspect of the IHS to examine change over time at the individual level (rather than changes in the overall distribution). This allows us to understand whether the inverse relationship between observational measures of food insecurity (i.e. calories and diversity) and the reported experiences of food insecurity at the national level holds at the level of the individual. To address the concern about potential confounding factors, we use multivariate regression to control for observable factors, and a fixed-effects estimator to control for all time-invariant unobservable individual-level factors with the fixed-effects estimator.13 As dependent variables in our model specifications, we 13 We considered a random-effects estimator, but results from a Hausman test led us to reject an 12 estimate change in the binary indicators for worrying about food in the last 7 days and experiencing food insecurity in the last 12 months, and also for the count of these months. There are many idiosyncratic reasons for why two individuals who look the same with respect to objective measures of food security (that is, they have similar levels of dietary diversity and caloric intake), may provide different responses to an experiential question on food security. For example, some individuals are inherently more nervous or risk-averse and are therefore more inclined to self-identify as food insecure. The fixed-effects estimator controls for these types of unobserved factors that are unique to individuals (thus reducing concerns of omitted variable bias). The fixed-effects regression can be expressed as follows: where is the set of experiential food security measures for individual i in time period t described above, is a vector of independent variables containing time-varying controls and measures of price variability and/or price shifts occurring in the last 12 months, represents the effect of each independent variable on food insecurity, and is the error term clustered (indicated by c ) at the level of the enumeration area (EA). We are interested in examining the hypothesis that increasing food prices alone increase anxiety about food security and reported exposure to food insecurity. To assess this, we include in our set of controls in observational (often referred to as objective) measures of food security status (i.e. FCS and daily per capita caloric intake). If worrying assumption of independence between the omitted variables and those in the model, suggesting that for this model specification, a random-effects estimator is biased. 13 about, and experiencing, food insecurity is due to these factors alone, then food prices should have no additional effect. Throughout the analysis, we use maize prices as our proxy for food prices. Maize accounts for more than 50 percent of households’ caloric consumption and so fluctuations in maize prices directly affect most of the population (Minot 2010). Panel weights are used to make the results representative of the national population of Malawi. 3.2 Results Table 4 shows the estimated average marginal effects (and conditional average marginal effects)14 from predicting our three outcomes of experiential food insecurity: a dichotomous measure of worrying about food in the last 7 days, a dichotomous indicator of experiencing food insecurity in the last 12 months, and a count of the number of months experiencing food insecurity in the last 12 months. For all models, we present the fixed-effects OLS estimates. For the binary outcome models, we also present fixed-effects logit estimates and for the count model, we report estimates from a fixed-effects Poisson estimator. The Logit estimator has the advantageous attribute that it bounds the fitted values between zero and one, and the Poisson estimator similarly is meant to better account for the clumping associated with count variables. Consistent with expectations, FCS is negatively associated with experiential food insecurity. A reduction in dietary diversity is associated with increased anxiety about, and experiences of, food insecurity at the national level (Table 4) and in rural areas of Malawi (Table 5). Given that FCS is intended to measure food security, it is reasonable to anticipate 14 For OLS, estimates are average marginal effects, and for the Logit and Poisson estimators the estimates are marginal effects averaged over the values of the observed controls but at a fixed point (zero) for the unobserved controls (i.e. individual-level fixed effects). We refer to the Logit and Poisson effects as conditional average marginal effects. 14 that they are negatively correlated even after controlling for fixed effects. In the bivariate case, we also observe that calories and the experiential measures are negatively correlated. The Pearson correlation coefficient between per capita calories and whether the household reported experiencing food insecurity is -0.16 with a p-value < 0.00005. The correlation with number of months experiencing food insecurity and whether the respondent worried about having sufficient food are both also negatively correlated with coefficients of -0.14 and -0.13, respectively (both with p-values < 0.00005). But, this correlation disappears when conditioning on total expenditure, food prices, and dietary diversity. Similarly, in the bivariate case, we observe that prices and experiencing food insecurity status are negatively correlated (i.e. increases in price levels, and increases in the rate of inflation, are associated with increased likelihood of reporting having experienced food insecurity). A plausible explanation for this bivariate association is that an increase in the price of maize could lead to a reduction in the household’s purchasing power, which could lead to a reduction in either the quantity or quality (as measured by FCS) of food consumed. We find though that even after controlling for total real expenditure, FCS, and caloric intake, changes in price levels and inflation are significantly correlated with the experiential measures of food insecurity nationally (Table 4) and in rural areas (Table 5). In all cases, the p-values for these estimates are less than 0.01 indicating that the partial correlations are all statistically significant. In addition to the power of the estimated partial correlations, the magnitudes of the effects are qualitatively important. The data indicate that a one-unit increase in the inflation rate of maize prices increases the expected number of months experiencing food insecurity by one-third (or approximately 10 days). While a one-unit increase in the inflation rate of maize between waves may seem large, it is smaller than the change experienced between the 15 two waves (see Table A1 in the Appendix). At the national level, the inflation rate of maize prices increased from -9.5 percent in 2010 to 117.6 percent in 2013 – this is a first difference increase of 1.27. Table 5 presents results by rural and urban status. Narrowing in on our key finding about the association between inflation, anxiety, and experiencing food insecurity, we continue to find that increases in inflation rates are associated with increased anxiety about food sufficiency and increased reporting of experiencing food insecurity. This association is statistically significant across all models, separately for urban and rural areas, with p-values of less than 0.01. The association also holds when conditioning out annual, real expenditure, calories consumed, and dietary diversity. An additional pattern that is observed in Table 5 is that the magnitude of this negative association is approximately twice as large in urban areas as in rural areas. A change in the rate at which maize prices are changing has a significantly more detrimental effect in urban areas relative to rural areas. 16 Table 4. Food-price inflation and experiential food insecurity, controlling for individual-level fixed effects Worried about food, Food insecure in the Months food insecure, last 7 days (se) last 12 months (se) last 12 months (se) OLS Logit OLS Logit OLS Poisson Maize price 12 months ago (month 1) 0.004*** 0.005*** 0.004*** 0.005*** 0.019*** 0.014*** (0.001) (0.001) (0.001) (0.001) (0.003) (0.005) Change in inflation rate of maize (past 12 months) 0.038*** 0.250** 0.074*** 0.092*** 0.339*** 0.220*** (0.011) (0.025) (0.011) (0.021) (0.046) (0.054) Dietary diversity (Food Consumption Score) -0.003*** -0.005*** -0.002*** -0.003** -0.017*** -0.013*** (0.001) (0.001) (0.001) (0.001) (0.003) (0.003) Daily per capita calorie consumption (thousands) -0.013 -0.010 -0.022* -0.016 0.004 0.050 (0.011) (0.018) (0.011) (0.021) (0.043) (0.041) Annual per capita expenditure (Kwacha) -0.000 -0.000 -0.000 -0.000 -0.000 -0.000* (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of observations 22,820 8,596 22,820 8,800 22,820 16,402 Adjusted R2 0.045 0.082 0.072 0.139 0.079 --- *** p<0.01, ** p<0.05, * p<0.1. Notes: All models control for fixed effects and observations are weighted to be representative of the population. For OLS, estimates are average marginal effects, and for the Logit and Poisson estimators the estimates are marginal effects averaged over the values of the observed controls but at a fixed point (zero) for the unobserved controls (i.e. individual-level fixed effects). We refer to the Logit and Poisson effects as conditional average marginal effects. The models in columns (ii) and (vi) are estimated in Stata using xtlogit and the model in column (iv) is estimated using xtpoisson; models in columns (i), (iii), and (v), are estimated using xtreg. Unlike the OLS fixed effects estimator, which uses all observations, the Logit and Poisson estimators only use observations whose outcome variable changes across time and whose outcome variables are not ‘0’ in both rounds, respectively. Therefore, number of observations presented in columns (ii) and (iv) reflect the number of observations tied to individuals whose status changes between waves. Panel sampling weights were rescaled to sum to the number of observations in the data and are treated as importance weights in Stata. Standard errors, in parentheses, from columns (ii), (iv), and (vi) are calculated using the jackknife method, clustered at the EA (enumeration area) level. 17 Table 5. Food-price inflation and experiential food insecurity by area, controlling for fixed effects Worried about food, Food insecure in the Months food insecure, last 7 days (se) last 12 months (se) last 12 months (se) Urban Rural Urban Rural Urban Rural Maize price 12 months ago (month 1) 0.004* 0.004*** 0.003 0.004*** 0.000 0.021*** (0.002) (0.001) (0.002) (0.001) (0.007) (0.004) Change in inflation rate of maize (past 12 months) 0.115*** 0.032*** 0.139*** 0.068*** 0.617*** 0.320*** (0.035) (0.011) (0.039) (0.011) (0.130) (0.049) FCS -0.002 -0.004*** -0.000 -0.003*** -0.008 -0.019*** (0.002) (0.001) (0.002) (0.001) (0.006) (0.004) Daily per capita calorie consumption (thousands) 0.007 -0.017 -0.043* -0.012 -0.000 0.032 (0.022) (0.013) (0.024) (0.014) (0.066) (0.057) Annual per capita expenditure (Kwacha) -0.000 -0.000 0.000 -0.000* 0.000 -0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Number of observations 5,358 16,480 5,358 16,480 5,358 16,480 Adjusted R2 0.098 0.041 0.103 0.067 0.133 0.076 *** p<0.01, ** p<0.05, * p<0.1. Notes: All models are estimated in Stata using xtreg and control for fixed effects. Observations are weighted to be representative of individuals who lived in urban areas in both waves, and those who lived in rural areas in both waves. 18 4. Discussion Some research shows that experiential measures of food security are consistent with objective, quantitative indicators. Frongillo (1999) and Webb et al. (2002) both show that the United States’ Household Food Security Survey Module (HFSSM), which includes a combination of quantitative and qualitative questions on households’, adults’ and children’s access to and utilization of food, is a strong measure of food security across multiple dimensions and even in multiple developing country contexts. In Malawi and Cambodia, analysis of a household living conditions survey found experiential food security to be correlated with household expenditure and dietary diversity (Headey and Ecker 2012). However, self-reported experiences of hunger and food insecurity are not necessarily indicative of problems with access to food. Deaton and Dreze (2009) find that reported experiences of hunger in India are not correlated with caloric intake. Headey and Ecker (2012) find that experiential food insecurity exhibits no relationship with caloric intake in Ethiopia. Similarly, using Living Standards Measurement Study (LSMS) data from Albania, Indonesia, Madagascar, and Nepal, Migotto et al. (2006) find that experiential food security indicators are weakly correlated with measures of food utilization, as measured through dietary diversity and anthropometry. The lack of correlation observed between objective and experiential measures of food security has led to the supposition that experiential indicators are perhaps capturing households’ perceptions of food stability and vulnerability. Because vulnerability refers to a point in the future, and is therefore difficult to measure, it is often assessed using experiential indicators of food security (Dercon 2001). Households may appear to be food secure in all objective measures of availability, access, and utilization at a given point in time and yet, they may perceive pending vulnerability or instability. An extension of this view is that the 19 experiential questions can help to measure psychological stress over the future of food security. Qualitative research on the impact of the food price crisis (along with the fuel and financial crises) between 2008 and 2011 found high levels of stress and anxiety over the ability to provide adequate food for the family was reported in nearly all interviewed communities from a study of 17 countries (Heltberg and others, 2012). Within the medical health literature, it is reasonably well accepted that seemingly subjective attributes such as stress and anxiety can have harmful physical effects. Rozanski and others (1999, p. 2192) note that “an extensive recent literature now establishes that psychosocial factors contribute significantly to the pathogenesis of CAD” [coronary artery disease]. Examples from this literature document that psychological stress is associated with increased cardiovascular risk factors, such as hypertension and insulin resistance, and with outcomes such as ischaemia, arrhythmia, and pump failure (see Brotman and others 2007, and references therein). It may be reasonable to assert, as Headey and Ecker (2012) do, that objective measures of nutritional outcomes should be of paramount importance in the context of poor countries. This though presumes an ability to precisely measure adverse effects of food insecurity in a timely manner. We believe there are a couple of reasons to be cautious in this view that objective measures of food security are of greater importance in poor countries. The first is simply about timing. Worrying about the future ability to meet food needs may be a useful early indicator of food insecurity leading to adverse nutritional outcomes at a later point in time. Objective measures of adverse outcomes from food insecurity may only be measurable after the onset of food insecurity. Attention to the need for early warning signs can aid rapid policy responses and potentially blunt the adverse consequences of food insecurity. A second reason is that the adverse physical effects of anxiety and stress are likely 20 to be challenging (and expensive) to measure, particularly in the context of poorer countries. Using subjective measures of whether someone has worried about food insecurity is a significantly less costly measure of stress. And the fact that some people will worry, while others may not, in response to the same shock is not an inherently negative attribute of asking whether someone has worried. This simply means that the adverse effects linked to anxiety will differ across these individuals, and whether they worried may be a direct measure of anxiety. The findings in this paper suggest that increasing food prices were found to heighten people’s worries about meeting their food needs, and similarly their assessments of experiencing food insecurity. This finding controls for several observable candidates to confounding factors. For example, anxiety occurs regardless of whether the price changes adversely affect diet quantity or quality. Furthermore, the findings in this paper demonstrate that people report experiencing food insecurity, regardless of whether the price changes affect their monetary wellbeing. In other words, large changes in food prices need not work through a chain of reduced purchasing power, and thereby reduced food consumption, to lead to food insecurity. This finding also controls for unobservable, time-invariant confounding factors. The use of individual-level fixed effects allows us to also rule out as an explanation that the correlation is simply due to some people being more nervous and for some unobservable reason, these people were hit by greater changes in prices. Our inference from this is that being exposed to changes in food prices introduces uncertainty about the ability to meet food needs and this is directly welfare-damaging. Incorporating experience-based measures into the conceptualization and measurement of food security is therefore important for practical and conceptual reasons. One practical reason is that experiential measures are significantly easier, and less costly, to 21 collect than caloric data and allow for analysis of food security at the level of the household or individual (Maitra and Rao 2014). Ease and cost of fieldwork are critical elements of successful monitoring of wellbeing, particularly in the poor country context. A conceptual basis for including experiential questions as elements of the measure of food insecurity is that the experiential measures may well be an early warning system of pending instability in access to food and future food insecurity. 22 References Anzid, K., F. Elhamdani, A. Baali, G. Boetsch, S. Levy-Desroches, P. Lopez, and M. Cherkaoui. 2009. “The Effect of Socio-Economic Status and Area of Residence on Household Food Variety in Morocco.” Annals of Human Biology 36:727–49. Barrett, C. 2002. “Food security and food assistance programs” in Handbook of Agricultural Economics. Elsevier. 2(2): 2103-2190. Barret, C. 2010. “Measuring Food Insecurity”. Science. 327. Behrman, J. & A. Deolaliker. 1990. “The Intra-Household Demand for Nutrients in Rural South India”. Journal of Human Resources. 25(4): 665-696. Block, S., L. Kiess, P. Webb, S. Kosen, R. Moench-Pfanner, M. Bloem, and C. Timmer. 2004. “Macro Shocks and Micro Outcomes: Child Nutrition during Indonesia’s Crisis.” Economics and Human Biology. 2(1): 21-44. Brotman, Daniel J, Sherita H Golden, and Ilan S Wittstein. 2007. “The Cardiovascular Toll of Stress.” The Lancet 370 (9592): 1089–1100. https://doi.org/10.1016/S0140- 6736(07)61305-1. Committee on World Food Security. 2013. Global Strategic Framework for Food Security and Nutrition. FAO. http://www.fao.org/fileadmin/templates/cfs/Docs1213/gsf/GSF_Version_2_EN.pdf Dabalen, A., A. de la Fuente, A. Goyal, W. Karamba, N. Viet Nguyen and T. Tanaka. 2017. Pathways to Prosperity in Rural Malawi Directions in Development. Washington, DC: The World Bank. https://openknowledge.worldbank.org/handle/10986/25743 Darko, Francis Addeah, Amparo Palacios-Lopez, Talip Kilic, and Jacob Ricker-Gilbert. “Micro-Level Welfare Impacts of Agricultural Productivity: Evidence from Rural Malawi.” The Journal of Development Studies 54, no. 5 (May 4, 2018): 915–32. https://doi.org/10.1080/00220388.2018.1430771. Deaton, A. and J. Dreze. 2009. “Food and Nutrition in India: Facts and Interpretations.” Economic and Political Weekly. 44(7):42-64. Dercon, S. 2001. “Assessing vulnerability”, Accessed from: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.196.3266&rep=rep1&type =pdf D’Souza, A., and D. Jolliffe. “Food Insecurity in Vulnerable Populations: Coping with Food Price Shocks in Afghanistan.” American Journal of Agricultural Economics 96, no. 3 (April 1, 2014): 790–812. doi:10.1093/ajae/aat089. FAO. 1983. Director-General’s Report on World Food Security: A Reappraisal of the Concepts and Approaches : Item IV of the Provisional Agenda. https://books.google.com/books?id=cSwKMwEACAAJ. 23 FAO. 1996. Rome declaration on world food security and world food summit plan of action. Rome: FAO. FAO. 2008. “Food Consumption Score: Consumption of the FCS.” PowerPoint slides. http://home.wfp.org/stellent/groups/public/documents/ena/wfp196627.pdf. FAO 2008. State of Food Insecurity in the World 2008. World Food Programme and IFAD. Rome: FAO. Frankenberg, E., D. Thomas, and K. Beegle. 1999. The Real Costs of Indonesia’s Economic Crisis: Preliminary Findings from the Indonesia Family Life Survey. Santa Monica, CA: RAND. Frongillo, E. 1999. “Validation of Measures of Food Insecurity and Hunger.” Journal of Nutrition. 129: S506-9. Gilbert, C. & C. Morgan. 2010. “Food Price Volatility.” Philosophical Transactions of the Royal Society. B(265): 3023-3034. Hatloy, A., J. Hallund, M. Diarra, and A. Oshaug. 2000. “Food Variety, Socioeconomic Status and Nutritional Status in Urban and Rural Areas in Koutiala (Mali).” Public Health Nutrition 3:57–65. Headey, D. & O. Ecker. 2012. Improving the Measurement of Food Security. IFPRI Discussion Paper 01225. Helen Keller International. 2011. The Food Security and Nutrition Surveillance Project, Round 4: February-May 2011. Bulletin No. 2. Helen Keller International: Bangladesh. Heltberg, Rasmus, Naomi Hossain, and An Reva, eds. 2012. Living through Crises: How the Food, Fuel, and Financial Shocks Affect the Poor. The World Bank. https://doi.org/10.1596/978-0-8213-8940-9. Hoddinott, J., and Y. Yohannes. 2002. “Dietary Diversity as a Food Security Indicator.” Food Consumption and Nutrition Division Discussion Paper 136. IFPRI (International Food Policy Research Institute), Washington, DC. Jolliffe, Dean, Iffath Sharif, Lea Gimenez, and Faizuddin Ahmed. “Bangladesh - Poverty Assessment : Assessing a Decade of Progress in Reducing Poverty, 2000-2010.” World Bank Poverty Assessment. World Bank Poverty Assessment. The World Bank, June 1, 2013. http://documents.worldbank.org/curated/en/2013/06/17886000/bangladesh- poverty-assessment-assessing-decade-progress-reducing-poverty-2000-2010. Kaminski, J., Christiaensen, L., Gilbert, C.L., 2014. The End of Seasonality? New Insights from Sub-Saharan Africa (No. 6907), Policy Research Working Paper Series. The World Bank. 24 Jones, A., F. Ngure, G. Pelto, and S. Young. 2013. “What Are We Assessing When We Measure Food Security? A Compendium and Review of Current Metrics.” American Society for Nutrition. 4: 481-505. Maitra, C. and D. Rao. 2014. An Empirical Investigation into Measurement and Determinants of Food Security in Slums of Kolkata. University of Queensland Discuss Papers Series No. 531. http://www.uq.edu.au/economics/abstract/531.pdf Maxwell, D. and R. Caldwell 2008. The Coping Strategies Index: Field Methods Manual. 2nd edition. CARE and World Food Programme. Maxwell, S. and M. Smith. 1992. “Part I. Household Food Security: A Conceptual Review”. Indicators and Data Collection Methods for Assessing Household Food Security. Ed. Maxwell, S. & Frankenberger, T. New York and Rome: UNICEF and IFAD. pp. 2-72. Migotto, M., B. Davis, G. Carletto, and K. Beegle. 2006. “Measuring food security using respondents’ perception of food consumption adequacy”. UNU-WIDER, United Nations University (UNU), Research Paper No. 2006/88. Narayan-Parker, Deepa, ed. 2000. Crying Out for Change. Voices of the Poor. New York: Published by Oxford University Press for the World Bank. National Research Council. 2006. Food Insecurity and Hunger in the United States: An Assessment of the Measure. Committee on National Statistics, Panel to Review the U.S. Department of Agriculture’s Measurement of Food Insecurity and Hunger, Gooloo S. Wunderlich and Janet L. Norwood (eds.). Washington, DC: The National Academies Press. Oddo, Vanessa M., and James Mabli. 2015. “Association of Participation in the Supplemental Nutrition Assistance Program and Psychological Distress.” American Journal of Public Health 105 (6): e30–35. https://doi.org/10.2105/AJPH.2014.302480. Oshaug, A. 1985. “The Composite Concept of Food Security,” in Introducing Nutritional Considerations into Rural Development Programs with Focus on Agriculture: A Theoretical Contribution. Ed. Eide, W. et al. Report No. 1. Oslo: Institute for Nutrition Research, University of Oslo. Pinstrup-Andersen, P. 2009. “Food security: definition and measurement.” Food Security. 1:2. Ram, Jagjivan, John Crawford, and M. S. Swaminathan. “World Food Conference Rome November 1974.” India International Centre Quarterly 2, no. 1 (1975): 13–20. Rah. J., N. Akhter, R. Semba, S. de Pee, M. Bloem, A. Campbell, R. Moench-Pfanner, K. Sun, J. Badham, and K. Kraemer. 2010. “Low Dietary Diversity Is a Predictor of Child Stunting in Rural Bangladesh.” European Journal of Clinical Nutrition 64:1393–98. 25 Rigterink, Anouk S., (2015). “Does Security Imply Safety? On the (Lack of) Correlation Between Different Aspects of Security.” Stability: International Journal of Security and Development, 4(1), p.Art. 39. DOI: http://doi.org/10.5334/sta.fw Rozanski, A., J. A. Blumenthal, and J. Kaplan. 1999. “Impact of Psychological Factors on the Pathogenesis of Cardiovascular Disease and Implications for Therapy.” Circulation 99 (16): 2192–2217. https://doi.org/10.1161/01.CIR.99.16.2192. Ruel, M. 2003. “Operationalizing Dietary Diversity: A Review of Measurement Issues and Research Priorities.” Journal of Nutrition 133 (11): 3911S–26S. Schmidhuber, J., and F. N. Tubiello. “Global Food Security under Climate Change.” Proceedings of the National Academy of Sciences 104, no. 50 (December 11, 2007): 19703–8. doi:10.1073/pnas.0701976104. Skoufias, E. 2003. “Is the Calorie-Income Elasticity Sensitive to Price Changes? Evidence from Indonesia.” World Development. 31(7): 1291-1307. Swift, J. 1989. “Why are Rural People Vulnerable to Famine.” UDS Bulletin. 20(2): 8-15. Timmer, C. 2000. “The macro dimensions of food security: economic growth, equitable distribution, and food price stability.” Food Policy. 25: 283-295. United Nations. 1975. Report of the World Food Conference. Rome. Webb, P., J. Coates, E. Frongillo, B. Rogers, A. Swindale, and P. Bilinsky. 2006. “Measuring Household Food Insecurity: Why It’s So Important and Yet So Difficult to Do.” Journal of Nutrition. 136(5):1404S-1408S. Weismann, D., L. Bassett, T. Benson and J. Hoddinott.. 2009. “Validation of the World Food Programme’s Food Consumption Score and Alternative Indicators of Household Food Security.” IFPRI, Washington, DC. Wheeler, T., and J. von Braun. “Climate Change Impacts on Global Food Security.” Science 341, no. 6145 (August 2, 2013): 508–13. doi:10.1126/science.1239402. World Bank. 1986. Poverty and Hunger: Issues and Options for Food Security in Developing Countries. World Bank Policy Study. Washington D.C. World Bank. 2016. Malawi Poverty Assessment, Poverty and Equity Global Practice, Washington, DC: The World Bank. 26 APPENDIX Table A1. Descriptive stats for panel sample Round 1 – 2010 Round 2- 2013 (SE) (SE) Food insecure in last 12 months 0.510 0.658 (0.018) (0.017) Months of food insecurity in last 12 months 1.503 2.154 (0.078) (0.090) Worried about food in last 7 days 0.282 0.373 (0.018) (0.017) Maize price twelve months ago 41.538 53.693 (1.049) (0.958) Inflation rate of maize prices over last 12 -0.095 1.176 months (0.025) (0.073) Food consumption score 48.464 50.448 (0.901) (0.835) Daily per capita calorie consumption 2.256 2.371 (thousands) (0.039) (0.045) Annual per capita expenditure (Kwacha) 138,558 142,292 (6,976) (5,956) Observations 11,410 11,410 Notes: Descriptive statistics provided for panel sample included in fixed effects OLS regressions presented in Table 4. This includes individuals interviewed in both waves for which none of the variables included in the regression are missing. Observations are weighted to be representative of individuals at the national level. Standard errors are adjusted for clustering and stratification. 27 Table A2. Descriptive stats for panel sample, rural and urban Urban in both rounds Rural in both rounds Round 1 Round 2 Round 1 Round 2 (SE) (SE) (SE) (SE) Food insecure in last 12 months 0.329 0.522 0.547 0.682 (0.034) (0.038) (0.020) (0.019) Months of food insecurity in last 12 months 0.723 1.410 1.660 2.287 (0.075) (0.118) (0.085) (0.101) Worried about food in last 7 days 0.192 0.369 0.297 0.373 (0.028) (0.025) (0.020) (0.020) Maize price twelve months ago 44.180 60.130 41.046 52.290 (1.601) (1.449) (1.215) (1.074) Inflation rate of maize prices over last 12 -0.119 0.965 -0.092 1.227 months (0.042) (0.082) (0.029) (0.085) Food consumption score 61.291 63.582 45.795 47.687 (2.289) (1.460) (0.844) (0.814) Daily per capita calorie consumption 2.499 2.522 2.199 2.317 (thousands) (0.102) (0.133) (0.042) (0.049) Annual per capita expenditure (Kwacha) 266,374 216,793 114,502 125,540 (38,925) (33,222) (4,004) (3,935) Observations 2,679 2,679 8,240 8,240 Notes: Descriptive statistics provided for panel sample included in fixed effects OLS regressions presented in Table 4. This includes individuals interviewed in both waves for which none of the variables included in the regression are missing. Additionally, ‘urban’ and ‘rural’ subsamples are restricted to individuals who are either urban or rural in both rounds; individuals who change urban/rural status between rounds are excluded. Observations are weighted to be representative of individuals at the urban and rural levels. Standard errors are adjusted for clustering and stratification. 28 Figure A1. Duration of subjective food insecurity 60 IHS3 (2010) 50 IHPS (2013) % of households 40 30 20 10 0 0 1 2 3 4 5 6 7 8 9 10 11 12 No. of months in last 12 months 29