The earlier famine and other forms of acute and severe food crises can be identified, the sooner programmatic responses can be designed and implemented. Often, however, these earliest stages fall into a grey area: where food insecurity is too severe to be considered a development problem but not severe enough to be considered a humanitarian one. In turn, in the period when programmatic interventions could be most useful, funding levels are often at their lowest. This opens up an urgent need to provide timely information on deteriorating food security trends and to confirm that this deterioration constitutes a crisis.
How these trends and confirmations can be captured, however, is an open question. Significant debate has taken place on how to accurately capture food security, even in non-crisis contexts. While food consumption diaries are often considered as a ‘gold standard’ in the field, they are far from perfect and are often cumbersome and difficult to collect, even in more standard situations. Other approaches, such as collecting information on the amount spent on food, are less cumbersome but prone to other errors, including weaker inference. Households could equally spend more on food due to increases in income, or changes in preferences, as from food insecurity, for example.
In the context of both idiosyncratic and covariate food crises, where deteriorations often take place very rapidly, practical concerns arise with the collection of survey data that goes beyond these standard measurement problems. Standard surveys are often collected with years between waves; while even more frequent ones could miss the crisis onset period entirely. Similarly, given the close relationship between conflict and food insecurity, collecting any data in person could be tricky. In this White Paper, we explore the requirements of a survey-based early warning system for the onset of severe food crises, then consider specific variables that should be collected in order to populate this system.
We argue that any system needs to comprise two stages: the first is designed to ascertain that trends are deteriorating, indicating that a crisis could be imminent; the second aims to confirm whether or not these deteriorating trends constitute a food crisis. Further, it implies that what is most important – in this context – is to correctly identify deteriorating trends in given locations. This requires data that can be collected at high frequency from a panel of individuals, which in turn imposes restrictions on the collection methodologies, and which variables should be considered. We conclude that, due to survey fatigue, survey duration, and access concerns that data should be collected remotely, from a large number of panels. This allows data to be collected in an on-going, near-continuous, manner, while each individual in the survey answers questions much more infrequently.
This data collection design imposes restrictions on the variables that can be collected. Multi-response or in-depth food diaries, especially over extended time periods, cannot be collected in this manner, for example. We therefore propose a number of tweaks to standard survey questions that allow them to be collected easily, and remotely, whilst holding strong ties to either the identification or confirmation stages. From this, we conclude that modern survey techniques can be a strong part of the armory in the early identification of food security crises; but that how surveys are collected, how often they are collected, and the variables that are collected in them must be tailored to the task at hand.