“…it will not be possible to tackle the many crises affecting humanity if we do not work and walk together, leaving no one behind. This demands, first and foremost, that we see others as our brothers and sisters, as members of the same human family, whose sufferings and needs affect us all, for “if one part suffers, all the others suffer with it” (cf. 1 Cor 12:26) … the goals set are ambitious and may seem unattainable. How can we achieve them? First of all, by not losing sight of the fact that at the heart of any strategy are the people, with concrete stories and faces, who live in a given place; they are not numbers, data or endless statistics”.
Pope Francis, 2022
“As long as primitive counts and raw scores are routinely mistaken for measures by our colleagues in Social, Educational and Health research, there is no hope of their professional activities ever developing into a reliable or useful science. We owe it to them, and to ourselves, to teach them how to construct measures which work as well as the ubiquitous physical measures by which they manage their everyday living, so that they can do a better job in making sense out of the profusions of data which they collect so enthusiastically.”
Benjamin D. Wright, 1999
“Even in reasoning upon some subjects, it is a mistake to aim at an unattainable precision. It is better to be vaguely right than exactly wrong.
Carveth Read, 1914
Introduction
With this short paper, my aim is to present the analytic approach that underlies the Food Insecurity Experience Scale (FIES), a globally valid standard for measuring the severity of the food insecurity conditions experienced by individuals or households.
The FIES is the latest innovation in the long-established field of food security assessments. It has been created to address some of the continuing challenges analysts encounter when trying to respond to the growing demand for valid, reliable, timely and sufficiently granular food security data to inform action. My main objective is to motivate the choice of the approach that led to the establishment of the FIES, which relies on important innovations in the theory and practice of measurement in the human behavioral and social sciences, not only because it allowed addressing the fundamental questions of the validity and reliability of FIES-based measures, but also because a similar approach may prove useful in addressing some serious shortcomings of other tools currently used to generate widely-used data to assess food insecurity.
I start by reminding readers of the more traditional and perhaps more intuitively obvious approach in measuring food security, based on the analysis of food consumption data. My objective there is to explain why to be normatively correct, while being able to assess adequacy of food consumption at the individual or even at the household level based on the food data typically collected in practice, is an almost impossible task. A conclusion, reached for example by the round table of experts convened by FAO’s Committee on Food Security (CFS) in October 2011 to discuss FAO methods for food security assessment, is that, given the nature of the food data that is typically available from most commonly available sources, including via carefully conducted population surveys, one has to limit the ambition of being able to provide very precise, disaggregated head-counts of the actual people who do not eat enough, and admit that the best that can be made is to provide rather approximate estimates of the prevalence of inadequate food consumption at the population level. The discussion will also highlight why some of the simplified methods for food consumption data collection, such as the ones used to inform the Food Consumption Score (FCS), originally proposed as being operationally feasible and because of that become very popular among analysts who conduct emergency food security assessments, are problematic. Rather than addressing them, these solutions exacerbate the validity problem (“are these numbers measuring what they are meant to be measuring?”) and leave the reliability question (“how good are the measures obtained in different empirical contexts?”) unanswered. As food consumption data will unlikely provide a solution to the problem of how to obtain valid, reliable, timely and feasible food security assessments, part of the profession looked elsewhere.
In the following section of the paper, I describe how, supported by convincing evidence cumulated over more than thirty years of research and applications worldwide, food insecurity at the individual or household level is best seen as a measurable latent trait. To do so requires relying on the so-called Rasch model, from the name of the Danish statistician who first introduced it in 1960 in the field of pedagogy and educational testing. Though rapidly growing, applications of the Rasch model have not yet permeated all areas of social research and therefore can be still seen as an innovation, and that is why I devote a specific short section in this paper to present its fundamental elements, suggesting ways in which it may be invoked to improve on the analysis of existing data collected with food frequency questionnaires as the one currently used, for example, to compute the Food Consumption Score.
Applied to the severity of food insecurity, as it was done for the first time in 1995, when a team of researchers at the Economic Research Service of the United States Department of Agriculture used to analyze items collected with the food security supplement of the general population survey, the Rasch model has provided the conceptual and practical means to answer the validity and the reliability questions with reference to food security assessments. With that in mind, I will finally turn to the main objective of the paper, which is to illustrate the steps taken at FAO, starting in 2012, to develop the Food Insecurity Experience Scale (FIES), which can be described as a globally valid, complete, food insecurity severity measurement system that can be adapted for any practical situation in which household or individual food security assessments are required.
When is food consumption “poor”, “borderline” or “adequate”?
A common interpretation of the food security question, when seen from the perspective of an individual, is to ask whether the person is consuming amounts of food that meet the requirements for an active and healthy life. This has led to operationalize empirical food security assessments by defining food requirements in terms of dietary energy and relying on the analysis of data on food consumption. To many, that may appear as a rather simple task. As put for example by Bill Gates, commenting on the method used to compute the Prevalence of Undernourishment (PoU) – still the most popular indicator of the extent of hunger in the world – “when the United Nations Food and Agriculture Organization calculates national rates of undernourishment, they don’t do what you might expect: take a large sample of people and determine how many are not eating enough” (Gates 2013, 32). Unfortunately, things are not so simple, because of both a methodological and a data-related problem. From a methodological point of view, the problem resides in the fact that, as people can – and normally do – adapt their food consumption to their energy needs, observed food consumption levels reflect, in part, underlying requirements, which are determined by individual characteristics such as sex, age, body mass and physical activity levels. Even if one could associate the relevant individual characteristics to each person sampled in Mr Gates’ ideal survey, the problem would remain of how to establish the appropriate normative threshold for energy requirement. Nutritionists go as far as suggesting that the typical methods used to assess the adequacy of consumption in population groups can be used for all nutrients for which the correlation between consumption and requirements is negligible, but cannot be used for energy (see for example the discussion in Gibson 2005 section 8.3, pp. 214-220). Without entering into much technical detail, let it suffice to mention that, over the years, researchers working at the FAO statistics division have devised an appropriate methodology to control for the partial endogeneity of the cut-off threshold to be used to establish dietary energy inadequacy starting from food consumption data collected in population surveys.[1] The method recognizes that a certain degree of variation in the levels of dietary energy consumption is to be expected in any population group, including when no one is under- or over-consuming. The concept of the Minimum Dietary Energy Requirement (MDER), defined as the lower bound of the range of energy requirements determined by normal differences in body masses and physical activity levels of active and healthy people, is introduced as the appropriate normative threshold to use to assess the PoU (see Figure 1). This ought to be recognized as a quite sophisticated, important contribution to the analysis of social phenomena where it is of crucial importance not to confound diversity with inequalities.
The figure illustrates the problem arising when analyzing dietary energy consumption data, due to the fact that, even in a population group where everybody is adequately nourished, there would be levels of dietary energy consumption above and below the average dietary energy requirement. Using the average requirement, which corresponds to the recommended dietary energy intake level for that population group, would imply a systematic overestimation of undernourishment.
The other important aspect to be considered when analyzing food consumption data with the aim to determine adequacy, is that one should obviously refer to levels of habitual, or usual, rather than occasional food consumption. As food consumption can vary from day to day, but also occasionally or periodically, due to seasonal variations or to cultural reasons, the data collected should be “cleaned” from the effect of excess variability that could be present because the data were collected with short reference periods (to avoid recall bias), in addition to the many other sources of measurement error that must be recognized as always present in food consumption data collected in surveys.[2] The lesson learned from decades of experience by researchers at FAO and elsewhere when analyzing these types of data, is that collecting food consumption data of sufficient quality is difficult and that measurement errors – which can be significant – are virtually unavoidable. Neglecting them implies deriving systematic overestimation of the extent of food consumption inadequacy (See Figure 2).
The figure illustrates the consequence of estimating the PoU using a frequency distribution derived from data characterized by substantial measurement error. While the presence of measurement errors will not influence the estimated mean, it will inflate estimates of the variance of the distribution, leading to heavier tails.
But to treat excess variability in a meaningful way adds another layer of complication to the methodology used to obtain PoU estimates. As can be imagined, the resulting methodology may be difficult to grasp by people who are not very familiar with sophisticated principles of statistical inference, which may well explain – though not justify – positions such as the one expressed by Mr Gates’ quote above. However, Mr Gates was right on spot when, a few lines later in the same article he wrote: “This is no way to track an important piece of the first Millennium Development Goal (MDG). The countries of the world committed themselves to cutting hunger in half, but they don’t know who is, in fact, hungry” (Gates 2013, 32). He was right and, from the discussion thus far, it should be evident that substantial improvements in the reliability of estimates of the extent of food insecurity, interpreted as the percentage of people with inadequate food consumption, might have been obtained only through the collection of more granular and detailed food consumption data, which unfortunately would imply substantial increases of data collection costs. Something else was needed.
One of the proposed solutions to conduct less expensive and quicker assessments relied on simplified food consumption data collection and analysis, as it is done with the Food Consumption Score (FCS), a concept developed by the World Food Program in the early 2000s to inform estimates of the prevalence of food insecurity in a population.[3]
The FCS is a simple numerical score that can take values on a range from 0 to 112 as a result of computing as a simple weighted average the frequency of consumption of food items from different food groups, during a 7-day reference period, as reported in a survey. The weights used to compute it are meant to a capture the “nutritional importance” of the various food groups, though it is unclear according to which criteria such importance is established. The computed score and conventional threshold values of 28 and 42 are used to classify the reporting households as having “poor” (FCS < 28), “borderline” (42 < FCS ≤ 28) and “adequate” (42 ≤ FCS) food consumption.
The attitude taken by food security analysts who created the FCS might be described using the same words Joel Michell, a researcher who has devoted significant efforts to study measurement in psychology (see for example Michell 1999), used to describe the one taken by psychologists, and that is worth quoting in full:
“Guided by a mixture of Pythagoreanism and operationalism, psychologists have devised a wide range of procedures that generate numerical data, including mental tests, rating scales, attitude and personality questionnaires, and magnitude estimations. For many it seemed that no more was involved in psychological measurement than devising such procedures.
Even if psychologists did not know exactly what they measured, they could be confident that because the procedures resulted in numerical assignments they must be measuring something”.
Joel Michell, 1990
To underestimate the relevance of the FCS in contributing to the prevailing narrative on the frequency and severity of food crises would be a serious mistake. It is one of the metrics most frequently used today to assess the adequacy of food consumption during emergencies. As it has been included since the beginning in the reference table of indicators for acute food insecurity classifications, its values condition virtually all assessments conducted according to the IPC (Integrated Food Security Phase Classification), a set of protocols presented as a standard for food security assessments during emergencies. (IPC Global Partners 2021) Until very recently, FCS data collected by WFP or others were the main – if not the only – source of quantitative data used in IPC assessments. Moreover, very recently, WFP has launched the Hunger Map, an Internet-based data portal where the percentage of people with FCS-based “poor” or “borderline” food consumption is presented as the only indicator of current levels of food consumption.
Given its popularity, it is perhaps more important than ever to scrutinize the properties of the FCS as an indicator of food insecurity, starting with the validity question: Does it measure what it is supposed to be measuring? Obviously, to explore this question requires that the measurand, that is, the attribute being measured, is clearly identified. This is already a problem for the FCS, as which attribute of “food consumption” it captures exactly remains unclear. Paraphrasing Joel Michell, its proponents at WFP believed that, as the procedure described in the manual produces numbers, it must be measuring something.
Based on published attempts at validating it, where values of the FCS are contrasted with measures of the dietary energy contained in the food consumed by the households over the same reference period, one derives the impression that its proponents consider it to be a proxy for caloric consumption.[4] There is not much else to do than to report the charts presented by Wiesmann et al. (2009) and by Baumann, Webb and Zeller (2013) to explain why that attempt at validating FCS as a metric of dietary energy consumption failed.
Note: the three charts in panels (a), (b) and (c) correspond to Figure 4, 8, 13, respectively, in Wiesmann et al. (2009), where household level food consumption (kcal/capita/day) is plotted against versions of WFP’s FCS modified to achieve the largest possible correlation using data from Burundi, Haiti and Sri Lanka. The chart in panel (d) is taken from Figure 3 in Baumann, Webb and Zeller (2013).
The scatter plots reproduced in Figure 1 above show the extremely weak correlation that can be found between the two variables. Though, from a purely statistical point of view, the Pearson’s correlation coefficients computed on the two variables can be considered significantly different from zero for series of the lengths considered in all four examples, this can hardly be interpreted as evidence that the FCS can serve as a good “proxy” of dietary energy consumption. Such a conclusion might have been defended, eventually, if the correlation coefficient where positive, of an order of magnitude greater than, say, 0.9. So: the FCS is clearly not measuring dietary energy consumption.
I shall return to the validity question and how to possibly address it shortly, but there is another piece of evidence in the scatterplots reproduced in Figure 3 that is worth highlighting: whatever the FCS may be capturing, it clearly does it with a lot of noise, as shown, for example, by the wide range of FCS values reported by households that would be consuming the same amounts of foods per capita. This raises a question of reliability, especially if the intention of the analysts were to respond to the call to be able to identify “who the hungry are”. Unfortunately, the FCS user manuals and other supporting materials provide no guidance on how to determine the extent and relevance of measurement errors.
The final observation that I make on the FCS, has to do with comparability of the assessments, over time and across space. Even setting aside the validity question for a moment, and assuming that the FCS might be measuring something, one would hope that such “something” is the same, independently of when and where the FCS data are collected. This is particularly important, for example, when FCS data are used in the context of initiatives that propose assessments involving many countries, over time, and for which comparability of the classifications is of essence.
The charts above represent daily time series of the percentage of households with either “poor” or “borderline” FCS in the population of two different provinces of Nigeria (panel a) and Afghanistan (panel b) retrieved from the https://static.hungermapdata.org/api-catalog/
Consider now the series reported in the charts in Figure 4, which are derived from data currently released under WFP’s Hunger Map Live. In the charts, the red dots represent the percentages of households in the population that has “inadequate food consumption”, defined as either “poor” or “borderline”, based on the reported values of the FCS, while the blue dots represent the percentages of households in the same population reporting having relied to relevant strategies to cope with food shortages, as captured by the so-called reduced Coping Strategy Index (r-CSI), another indicator commonly used in emergency food security assessments. It is evident that two series cannot be representing the same phenomena in the two contexts. In the first case, while the estimated percentages are very similar, at levels of around 30% of the population, from a certain moment in time they clearly show a divergent path. In the second case, however, the two series depict very different levels, with the FCS pointing to percentages around 90%, while the rCSI pointing to levels of around 60%.
Admittedly, these are just two examples (though many more can be found by perusing the rich database made generously available by WFP), and a creative analyst may indeed be able to find explanations to justify these patterns. Moreover, I do not deny that the data behind these statistics may contain very relevant information. That is not my objective, as in a world where information is so strategic, every kind of information is precious.[5]
The points I am making in this paper are different and relate to how these statistics are presented to the general audience and how the information contained in the data collected is processed. The main problem I have is that the very fact that these are presented as “statistics” based on “key indicators” conveys an aura of objectivity and scientific neutrality that, unfortunately, is neither proven nor qualified. Perhaps the behavior of the series is dominated by the noise that inevitably distorts the signal contained in the data, perhaps the arbitrarily set thresholds used for classification are not truly equivalent in different contexts, and therefore reporting “poor” or “borderline” food consumption may have different meanings in terms of what the actual “food insecurity” status of these populations may be. Unfortunately, despite the efforts I have put over the last decade into exploring the literature on food security measurement, I have failed to identify research that addressed these fundamental questions of reliability and validity, which are the only criteria that should be used to define what is being done quantitatively as science.
Measuring the unobservable: the Rasch model
Given the discussion in the preceding section of this paper, I would not be surprised that readers may be skeptical regarding the possibility to find a solution to the problem of scientifically measuring food insecurity. Two things should be clear, that make it a special challenge. First, food insecurity, as a theoretical construct, the way it is probably interpreted by most of us, cannot be taken simply as a synonym of reduced quality and/or quantity of the food consumed. Further, though being one of its most frequent determinants, food insecurity is also not a synonym of malnutrition and therefore is not a physical or biological human attribute. Seen from the social science perspective, it is also not just yet another name given to extreme poverty, to limited resilience or coping ability, or similar constructs. Food insecurity has long been recognized as something complex, that involves all dimensions that, at any moment, contribute to people’s inability to access the food they need to conduct active, healthy, and dignified lives, irrespective of their cultural, demographic, or socio-economic status. Second, it should also be clear that simply devising a procedure to attach numbers to cases falls very short from quantifying the magnitude of the attribute of interest, in ways that – as Benjamin D. Wright reminds us in the quote reported in the opening of this paper – “work as well as the ubiquitous physical measures by which [we] manage [our] everyday living”.
In this section I argue that, though impossible to observe directly, as a physical or biological attribute, the severity of the food insecurity condition of people can be measured in a scientifically proper sense. This can be done by applying the Rasch model (Rasch 1960; Bond, Yan, and Heene 2020) to measure the severity of the food insecurity condition of individuals or households through data on the self-reported occurrence of behaviors and conditions that are typically associated with a situation of inability to freely access food.
The Rasch model is a general approach to infer the unobservable magnitude of a subject’s latent trait (such as, for example, the level of competency on a subject matter possessed by a student) from the observable responses that the subject gives to a number of different “items” (such as the questions included in an evaluation test), via rigorous application of sound statistical principles. This is not the place to enter into a detailed technical discussion of the way in which the Rasch model is applied in practice, referring the reader to the existing voluminous literature on the subject.[6] It may be important, however, to underline that the only requirement to consider a problem treatable through the lenses of the Rasch model, is that both the subject and the items used for assessment are located on the same unidimensional scale (say, the one that measures both the student’s ability and the test items’ difficulty in the example made above), and that the greater the distance between the subject and the item, the more likely it is to observe a certain, definite response. In formal terms, the model postulates that the probability of observing a certain response (say, an affirmative answer), by a respondent to a question, is a logistic function of the distance, on an underlying scale of severity, between the position of the respondent, and that of the item [see mathematical formula in original paper].
Despite its simplicity, this formulation of the relationship between the observable data (the set of responses) and the unobservable measures (the parameters and ) is the only one (among many possible alternative, more flexible formulations proposed in the broader field of item-response theory – IRT) that ensures the property that measures be invariant.[7] Loosely speaking, invariance of the measures means that the way a measurement tool works is not distorted by the specific object that happens to be measured.[8] I hope my readers will agree that this looks like a very important property of measurement tools to be used in science, and that therefore, unless proven impossible, any creative way of transforming the data collected with the best possible intention to capture something relevant to the policy question should be tested for adherence to the restrictions imposed by the Rasch model.
After all, applying the Rasch model in practice should not be difficult for experienced quantitative analysts versed in statistical inference [See mathematical formulas in original paper].
In this sense, the most important message from this section of my paper is that analysts interested in resolving the potential shortcomings of indicators such as the FCS (but a similar proposition can be made for the Household Dietary Diversity Score, the Coping Strategy Index, or some of the very recently proposed indicators of diet quality at https://www.dietquality.org/) that are being used for international assessments, should feel compelled to attempt to construct Rasch-based measures from the data they have collected. This will force them to settle on the premise to the fundamental validity question, by having to provide an operational definition of their measurand (it may be the “adequacy of food consumption levels”, for a measure derived from data collected with FCS, HDDS or DQQ questionnaires), and to explore, in rigorous quantitative analytic terms, issues of reliability. If successful, they may finally confidently claim to have established a measure.
Experience-based food security measurement and prospects for better food security assessment
My final section turns to the Food Insecurity Experience Scale, as the first example, within the broad field of food security assessment, of successful completion of the entire process I sketched in closing the previous section.
When FAO started exploring ways to respond to the solicitations which emerged from the 2011 meeting of the CFS, to find a way to substantially improve global food security monitoring, it did not have to start from scratch. Analysis of food insecurity experiences data, which has its roots in the very pioneering works conducted in the early 1990s at Cornell University (Radimer, Olson, and Campbell 1990; Radimer et al. 1992; see also Kendall, Olson, and Frongillo 1995) and in the context of the Community Childhood Hunger Identification project (Wehler, Scott, and Anderson 1992), had already successfully led to the development of a quantitative food insecurity and hunger measurement scale, thanks to the research conducted by sociologists at the Economic Research Service of the US Department of Agriculture (Hamilton et al. 1997; Nord, Jemison, and Bickel 1999) using the Rasch model. Since 1995, experience-based food security measurement had been successfully applied in many countries and in different contexts, and the approach had been discussed at an International Symposium in 2002 (E. Kennedy 2003). In addition to the US, use of experience-based measures as a way to collect official data to inform national food security assessment had been implemented in Canada, Brazil and Mexico, and successfully tested in several other countries.
By 2012, there was a sufficiently broad interest towards the method, and diffused applications, though questions were raised on the possibility to obtain universally comparable assessments across countries and cultures (Coates et al. 2006; Jensen 2003).
Indeed, one problem was that we were still lacking a proper method to calibrate the food insecurity measures obtained in different contexts, possibly using slightly different survey modules adapted to the local cultural and linguistic conditions, against a common reference standard. It was through the “Voices of the Hungry” project,[9] started in 2012 thanks to generous support to FAO from the Government of Belgium and the United Kingdom, and later by the Bill and Melinda Gates Foundation and the European Commission, that researchers at the Statistics Division of FAO where finally able to develop the methods and tools needed to produce such comparable measures (Cafiero et al. 2016; Cafiero, Viviani, and Nord 2018) and to introduce the Food Insecurity Experience Scale (FIES) to the international community, as a complete food security measurement system, composed of a data collection tool (the FIES survey module), protocols for its field application, procedures to conduct the statistical validation of the data collected and to assess their reliability, to compute food security measures, and, most importantly, to calibrate those measures against a global FIES reference scale, so that assessments conducted in different populations and over time could be considered properly comparable, in a deeply scientific way.
As a result, the FIES has been endorsed as the basis to compute SDG indicator 2.1.2. and since its introduction, the FIES has been extensively validated for use all over the world and in very different contexts. (Frongillo 2022) The relevance of indicator 2.1.2 in the context of the ambitious 2030 Agenda for Sustainable Development should not be underestimated, as it allowed to enrich the prevailing global narrative on food insecurity (which tended to be focusing on hunger) to include moderate food insecurity as an important driver of various forms of malnutrition. In this respect, experience-based food security measures will be a very useful tool to guide policies intended to fight not only hunger, but malnutrition in all its forms (Pérez-Escamilla 2012; Pérez-Escamilla et al. 2017).
More recently, a modified FIES survey module, adapted to report conditions and experiences occurring during the 30 days preceding the interviews, has been used to collect data in several food crisis countries in 2020 and 2021, during the height of the COVID-19 pandemic. Results have led to publishing two consecutive reports on access to food in these countries (Boero et al. 2021; Cafiero et al. 2022) and to inform discussions within IPC Technical Advisory Group, to include the 30-day referenced FIES-based indicators in the reference table for acute food insecurity assessments under the IPC. During the COVID-19 pandemic, FIES has been used to collect food security data in the LSMS-High Frequency Phone Surveys initiative of the World Bank. Since 2020, the FIES is the chosen questionnaire to be proposed to measure household food insecurity in the DHS Program administered by USAid. Since 2022, FIES is used to collect data in the agricultural community of food crisis countries, disseminated through the Data in Emergency (DIEM) Information System initiative of FAO.
Concluding remarks
A series of unprecedented global events such as the COVID-19 pandemic, and heightened attention devoted to traditional drivers of food insecurity and hunger, such as conflicts, extreme weather events linked to climate change, and the continuation of economic downturn and crises caused by distortions in the international commodity and financial markets has caused the attention given by experts, policy makers and the general public on food security measurement issues to reach levels never achieved before. At the same time, with the fundamental revolution that accompanied the emergence of the new Information and Communication Technology era we live in, we see a proliferation of channels through which a continuous flow of data and information reaches the general public, often with no time for the traditional mechanisms that human society has put in place to exercise the fundamental role of separating the information wheat from the chaff.
Science is perhaps the most important of such mechanisms, a reason why this workshop on “Food and Humanitarian Crises: Science and Policies for Their Prevention and Mitigation” convened by the Pontifical Academy of Sciences is of crucial importance at a crucial moment, when decisions must be made quickly to protect the fundamental human right to adequate food and nutrition.
In this context, contrasting the deep scientific theory and practice that underscored the development of the Food Insecurity Experience Scale with the more pragmatic and operationally oriented intent that motivates other popular food security and nutrition assessment tools, this paper has attempted to raise some caution against too simplistic interpretations of published data on “real-time” food security assessments, while offering suggestions on how current practices in this important area of social policy might be substantially improved.
References
Baumann, Soo Mee, Patrick Webb, and Manfred Zeller. 2013. ‘Validity of Food Consumption Indicators in the Lao Context: Moving toward Cross-Cultural Standardization’. Food and Nutrition Bulletin 34 (1): 105–19. https://doi.org/10.1177/156482651303400112
Boero, Veronica, Carlo Cafiero, Filippo Gheri, Anne W. Kepple, José Rosero Moncayo, and Sara Viviani. 2021. Access to Food in 2020. Results of Twenty National Surveys Using the Food Insecurity Experience Scale (FIES). FAO. https://doi.org/10.4060/cb5623en
Bond, Trevor G., Zi Yan, and Moritz Heene. 2020. Applying the Rasch Model: Fundamental Measurement in the Human Sciences. 4th ed. New York: Routledge. https://doi.org/10.4324/9780429030499
Cafiero, Carlo. 2020. ‘Measuring Food Insecurity’. In Food Security Policy, Evaluation and Impact Assessment, edited by Sheryl L. Hendriks, 169–205. London and New York: Routledge.
Cafiero, Carlo, Filippo Gheri, Anne W. Kepple, José Rosero Moncayo, and Sara Viviani. 2022. Access to Food in 2021: Filling Data Gaps: Results of Twenty National Surveys Using the Food Insecurity Experience Scale (FIES). Rome, Italy: FAO. https://doi.org/10.4060/cc0721en
Cafiero, Carlo, Hugo R. Melgar-Quiñonez, Terri J. Ballard, and Anne W. Kepple. 2014. ‘Validity and Reliability of Food Security Measures’. Annals of the New York Academy of Sciences 1331 (1): 230–48. https://doi.org/10.1111/nyas.12594
Cafiero, Carlo, Mark Nord, Sara Viviani, Mauro Eduardo Delgrossi, Terri J. Ballard, Anne W. Kepple, Meghan Miller, and Chiamaka Nwosu. 2016. Methods for Estimating Comparable Prevalence Rates of Food Insecurity Experienced by Adults throughout the World. Voices of the Hungry Technical Report 1. Rome: Food and Agriculture Organization of the United Nations.
Cafiero, Carlo, Sara Viviani, and Mark Nord. 2018. ‘Food Security Measurement in a Global Context: The Food Insecurity Experience Scale’. Measurement 116 (February): 146–52. https://doi.org/10.1016/j.measurement.2017.10.065
Coates, Jennifer, Edward A. Frongillo, Beatrice Lorge Rogers, Patrick Webb, Parke E. Wilde, and Robert Houser. 2006. ‘Commonalities in the Experience of Household Food Insecurity across Cultures: What Are Measures Missing?’ The Journal of Nutrition 136 (5): 1438S-1448S. https://doi.org/10.1093/jn/136.5.1438S
Engelhard, Jr., George. 2012. Invariant Measurement: Using Rasch Models in the Social, Behavioral, and Health Sciences. New York: Routledge. https://doi.org/10.4324/9780203073636
Engelhard, Jr., George, and Jue Wang. 2021. Rasch Models for Solving Measurement Problems: Invariant Measurement in the Social Sciences. SAGE Publications.
FAO and The World Bank. 2018. Food Data Collection in Household Consumption and Expenditure Surveys: Guidelines for Low- and Middle-Income Countries. The World Bank Group.
Fischer, Gerhard H., and Ivo W. Molenaar, eds. 1995. Rasch Models. New York, NY: Springer. https://doi.org/10.1007/978-1-4612-4230-7
Frongillo, Edward A. 2022. ‘Validity and Cross-Context Equivalence of Experience-Based Measures of Food Insecurity’. Global Food Security 32 (March): 100599. https://doi.org/10.1016/j.gfs.2021.100599
Gates, Bill. 2013. ‘Vital Statistics’. Development Asia. Beyond the MDGs: What Will the Global Development Agenda Look like after 2015?, November 2013. https://www.adb.org/publications/beyond-mdgs-what-will-global-development-agenda-look-after-2015
Gibson, Rosalind S. 2005. Principles of Nutritional Assessment. Oxford University Press.
Hamilton, William L, John T. Cook, William W. Thompson, Edward A. Frongillo, Christine M. Olson, and Cheryl A. Wehler. 1997. ‘Household Food Security in the United States in 1995 Technical Report of the Food Security Measurement Project: Home Economics and Household Collection’. U.S. Dept. of Agriculture, Food and Consumer Service, Office of Analysis and Evaluation. http://libcdm1.uncg.edu/cdm/ref/collection/HENP/id/159
IPC Global Partners. 2021. Integrated Food Security Phase Classification Technical Manual Version 3.1. Evidence and Standards for Better Food Security and Nutrition Decisions. Rome: The Integrated Food Security Phase Classification (IPC) Global Partners.
Jensen, Helen H. 2003. ‘Discussion Opener – Qualitative Measures of Food Insecurity and Hunger’. In Measurement and Assessment of Food Deprivation and Undernutrition, by FIVIMS. Rome, Italy: Food and Agriculture Organization of the United Nations. https://www.fao.org/3/y4249e/y4249e0c.htm#bm12.3
Kendall, A., C.M. Olson, and E.A. Frongillo. 1995. ‘Validation of the Radimer/Cornell Measures of Hunger and Food Insecurity’. The Journal of Nutrition 125 (11): 2793-2801. https://doi.org/10.1093/jn/125.11.2793
Kennedy, Eileen. 2003. ‘Keynote Paper: Qualitative Measures of Food Insecurity and Hunger’. In Measurement and Assessment of Food Deprivation and Undernutrition, by FIVIMS. Rome, Italy: Food and Agriculture Organization of the United Nations. https://www.fao.org/3/y4249e/y4249e06.htm#bm12
Kennedy, Gina, Andrea Berardo, Cinzia Papavero, Peter Horjus, Terri Ballard, MarieClaude Dop, Jan Delbaere, and Inge D. Brouwer. 2010. ‘Proxy Measures of Household Food Consumption for Food Security Assessment and Surveillance: Comparison of the Household Dietary Diversity and Food Consumption Scores’. Public Health Nutrition 13 (12): 2010-18. https://doi.org/10.1017/S136898001000145X
Michell, Joel. 1999. Measurement in Psychology: A Critical History of a Methodological Concept. Ideas in Context. Cambridge: Cambridge University Press. https://doi.org/10.1017/CBO9780511490040
Naiken, Loganaden. 1996. ‘Appendix 3’. In The Sixth World Food Survey, by FAO. Rome, Italy: Food and Agriculture Organization of the United Nations.
———. 2003. ‘Keynote Paper: FAO Methodology for Estimating the Prevalence of Undernourishment’. In Measurement and Assessment of Food Deprivation and Undernutrition, by FIVIMS. Rome, Italy: Food and Agriculture Organization of the United Nations. https://www.fao.org/3/y4249e/y4249e06.htm#bm06
———. 2014. Methodological Issues in the Estimation of the Prevalence of Undernourishment Based on Dietary Energy Consumption Data: A Review and Clarification. FAO Statistics Working Paper Series 14-03. Rome, Italy: FAO. https://www.fao.org/documents/card/en/c/9e1ec927-bffa-4ac6-bdfe-2d069fe57b71/
Nord, Mark, Kyle Jemison, and Gary Bickel. 1999. ‘Prevalence of Food Insecurity and Hunger, by State, 1996-98’. Research Report 2. Food Assistance and Nutrition Research Report. Washington, DC. https://www.ers.usda.gov/webdocs/publications/46583/51195_fanrr2.pdf?v=1899.4
Pérez-Escamilla, Rafael. 2012. ‘Can Experience-Based Household Food Security Scales Help Improve Food Security Governance?’ Global Food Security 1 (2): 120-25. https://doi.org/10.1016/j.gfs.2012.10.006
Pérez-Escamilla, Rafael, Muriel B. Gubert, Beatrice Rogers, and Amber Hromi-Fiedler. 2017. ‘Food Security Measurement and Governance: Assessment of the Usefulness of Diverse Food Insecurity Indicators for Policy Makers’. Global Food Security 14 (September): 96-104. https://doi.org/10.1016/j.gfs.2017.06.003
Perline, Richard, Benjamin D. Wright, and Howard Wainer. 1979. ‘The Rasch Model as Additive Conjoint Measurement’. Applied Psychological Measurement 3 (2): 237-55. https://doi.org/10.1177/014662167900300213
Radimer, Kathy L., Christine M. Olson, and Cathy C. Campbell. 1990. ‘Development of Indicators to Assess Hunger’. The Journal of Nutrition 120 (suppl_11): 1544-48. https://doi.org/10.1093/jn/120.suppl_11.1544
Radimer, Kathy L., Christine M. Olson, Jennifer C. Greene, Cathy C. Campbell, and Jean-Pierre Habicht. 1992. ‘Understanding Hunger and Developing Indicators to Assess It in Women and Children’. Journal of Nutrition Education 24 (1, Supplement 1): 36S-44S. https://doi.org/10.1016/S0022-3182(12)80137-3
Rasch, Georg. 1960. Probabilistic Models for Some Intelligence and Attainment Tests. Studies in Mathematical Psychology, I. Oxford, England: Nielsen & Lydiche.
Wehler, Cheryl A., Richard Ira Scott, and Jennifer J. Anderson. 1992. ‘The Community Childhood Hunger Identification Project: A Model of Domestic Hunger – Demonstration Project in Seattle, Washington’. Journal of Nutrition Education 24 (1, Supplement 1): 29S-35S. https://doi.org/10.1016/S0022-3182(12)80135-X
Wiesmann, Doris, Lucy Bassett, Todd Benson, and John Hoddinott. 2009. ‘Validation of the World Food Programme’s Food Consumption Score and Alternative Indicators of Household Food Security’, IFPRI Discussion Paper, June, 105.
[1] Details of the way in which FAO computes PoU estimates are given, every year, in an Annex of the State of Food Security of Nutrition report and can be found in the official metadata document of the SDG indicator 2.1.1 and the references therein. For an extended discussion on the PoU methodology, see in particular (Naiken 1996; 2003; 2014).
[2] See FAO and the World Bank (2018) for an extensive treatment of the issue that must be considered when collecting food consumption data in surveys, to reduce the extent of possible bias.
[3] The FCS was conceived as an attempt to simplify the collection of food consumption data in surveys by using less-demanding survey modules, such as food frequency questionnaire, so that data collection would be less expensive and surveys more frequent. See https://resources.vam.wfp.org/data-analysis/quantitative/food-security/food-consumption-score for technical guidance on how to compute it. For a more comprehensive critical discussion of these and other methods to conduct assessments of the adequacy of food consumption, see Cafiero et al. (2014) and Cafiero (2020).
[4] Though the way it is computed would suggest it to be more a metric of dietary diversity and hence of the overall quality of the diet, as it is described, for example by Kennedy et al. (2010).
[5] One explanation, for example, may be that the two series represent two distinctly different attributes of the households: “food consumption” and “coping behaviors”, as the names of the indicators suggest, and that while in Nigeria households are still able to somehow protect their food consumption levels, in Afghanistan most households may have exhausted their coping abilities. The decreasing trends in the r-CSI coupled with constant or increasing trends in the FCS can thus be interpreted as coherent signs of a continuing deterioration.
[6] Especially useful are the books by Bond et al. (2020), which presents a very complete yet accessible treatment of the Rasch model described as the tool to achieve fundamental measurement in the human sciences, and the one edited by Fischer and Molenaar (1995), which provides a collection of technical contributions on the foundation, extensions and examples of applications, including by introducing statistical estimation algorithms. Of note, in the context of the discussion in this paper, is the very recent book by Engelhard, Jr. and Wang (2021) where the Food Insecurity Experience Scale is used as an extended example throughout the book, to demonstrate points related to scale construction, evaluation, maintenance and use.
[7] For a detailed treatment of measurement invariance, see Engelhard, Jr. (2012).
[8] This is a fundamental aspect of what has been defined stochastic conjoint additive measurement (Perline, Wright, and Wainer 1979).