The 1996 World Food Summit (WFS) called for a 50 percent reduction in the number of undernourished people by 2015. In 2000, the Millennium Declaration (MD) recognized the value of hunger and poverty reduction by setting the MDG target of “halving, between 1990 and 2015, the proportion of people who suffer from hunger” (target 1.C).
To monitor progress towards the WFS and the MDG1 targets, FAO provides regular updates to the number and proportion of persons below the minimum level of dietary energy requirement (MDG indicator 1.9). Such estimates, which are produced at global, regional and country levels where reliable data is available, are presented annually in the State of Food Insecurity in the World (SOFI) report.
According to the FAO, undernourishment is a condition of “continued inability to obtain enough food”, and the Prevalence of Undernourishment (PoU) measures the “probability that a randomly selected individual from a population is found to be consuming less than her/his requirement for an active and healthy life”. This probability is assessed against a normative minimum threshold established by nutritionists for reference age and sex groups. While it is not possible to assess precisely the individual dietary energy requirement, the PoU is based on an inference at the population level in probabilistic terms. Indeed, the FAO methodology for estimating the prevalence of undernourishment refers to:
• a probability distribution of habitual Dietary Energy Consumption (DEC) of a representative individual in a population; and
• a cut-off point for intake adequacy – Minimum Dietary Energy Requirement (MDER) – specific for the same population.
In 2011-12 the FAO methodology for estimating the prevalence of undernourishment was deeply reviewed. The revised methodology was first introduced in the 2012 SOFI. Models adopted to describe the habitual dietary energy consumption in the population were reviewed and changed where national level survey data were available. The estimation of the parameters was also reviewed.
Recent revisions of the methodology
Parameters to characterize the distribution of food consumption employed to estimate the PoU are derived from different sources. To compute per mean or per capita DEC at a national level, FAO relies on Food Balance Sheets. The latest data from this source refer to 2011; therefore, additional sources were needed to estimate the DEC for the last 3 years, from 2012-14. The main source for 2012 and 2013 estimates were projections prepared by the Trade and Market Division of FAO. The Holt-Winters distributed lag model was instead used to project the DEC for 2014. In some cases, the same model was applied to compute projections also for 2012 and 2013, when data from the Trade and Market Division were not available. The Holt-Winters model uses a process known as exponential smoothing, which attributes higher weights to the more recent data and progressively less weight to the older observations. Weights decrease in each period by a constant amount, which lies on an exponential curve. For countries showing peculiar patterns, other simpler forecasting models were used, such as linear or exponential trends.
Two more parameters are required to characterize the distribution of food consumption: the Coefficient of Variation (CV) and the Skewness (SK). These are computed from national household surveys where they are available, which is the case for a wide sub-sample of the monitored countries. In recent years, thanks to the collaboration between FAO and National Statistical Offices, the Statistics Division has processed more than 100 surveys to obtain new parameters for about 50 countries that, together, cover more than 60 percent of the number of undernourished in developing regions.
When FAO started the regular monitoring of undernourishment, after the 1996 World Food Summit, the distribution of dietary energy consumption in the population was assumed to be log-normal. One of the main changes that FAO introduced in 2012 was the adoption, for the countries where adequate survey data were available, of a skew-normal distribution. This is fully characterized by the three parameters: the mean, the CV and the SK. The skew-normal distribution is more flexible than the log-normal and therefore better accounts for changes in the asymmetry that take place when food consumption changes, that is, when countries make progress toward greater food security.
In the SOFI 2014 edition, an even further refinement was introduced. The data are used to determine the appropriate distributional form for food consumption. In this way, the empirical SK from the distribution of per capita calorie consumption is derived from available national household surveys. The resulting model makes it possible to account for reductions in inequality of food consumption, such as those made by targeted food intervention programmes, thus ensuring a smooth transition towards a distribution in which food consumption is symmetric.
The 2014 edition of SOFI also introduced a new outlier detection method for consumption data derived from national household surveys. This is known as “leave-out-one cross-validation”, and allows for a robust calculation of the parameters in case of noisy data. Excess variability was also controlled for by using a linear regression, linking the log of per capita income to per capita calorie consumption, along with indicator variables for the month the survey was conducted to control for seasonality.
The MDER is estimated according to the normative standards set forth in the 2001 FAO/WHO/UNU Expert Consultation. To minimize the probability of overestimating undernourishment, the FAO method uses the minimum of the range of values consistent with adequate nourishment , therefore the lowest acceptable body weight for a given height and light activity. The cut-off point for a population is derived by aggregating sex and age-specific MDERs using the proportion of the population in the different sex and age groups as weights. Since the sex-age distribution of the population changes over time, the cut-off point is updated every year to reflect changes in the demographic structure of the population. This edition of SOFI uses updated population estimates from the latest revision published by the UN Population Division in June 2013. When data on population heights are not available, reference is either made to data on heights from countries where similar ethnicities prevail, or to models that use partial information to estimate heights for various sex and age classes.
Food losses occurring at the retail level were also introduced for the first time in 2012. Country-specific values regarding the average per capita loss of calories have been estimated taking into account data provided by a recent FAO study on food losses at various stages of the commodity chain.
SOFI 2014 also includes a new estimation of CVs for countries where reliable consumption data from surveys are not available. The method is based on a relationship between the CV due to income and macroeconomic variables – GDP per capita, and the Gini – and food prices. Finally, SOFI 2014 introduces a time-varying computation of variability in food consumption due to requirements in order to account for the world’s rapidly changing population structure.