Methodology for Calculation
Methodological Description The indicator is computed at the population level. For this purpose, the population is represented by a “representative” individual, for whom the probability distribution of habitual daily calorie consumption is modeled through a parametric probability density function (pdf). Once the parametric probability density function is characterized, the indicator is calculated as the integral probability that habitual daily calorie consumption (x) falls below the lower bound of the range of normal dietary energy requirements for that representative individual (MDER), as shown in the formula below: PoU = \int_{x<MDER} f(x \mid DEC; CV; Skew)\,dx where DEC, CV, and Skew denote respectively the mean, coefficient of variation, and skewness characterizing the distribution of habitual calorie consumption levels in the population. Until 2012, the probability distribution f(x) was modeled as a log-normal parametric probability density function relying on only two parameters: the arithmetic mean and the coefficient of variation. In its most recent formulation, it is modeled as a three-parameter parametric probability density function capable of representing different degrees of skewness, ranging from the symmetric normal distribution to the positively skewed log-normal distribution. This flexibility in capturing varying degrees of skewness is required given that human energy consumption is naturally bounded by physiological limits. As average consumption levels increase, the skewness of the distribution is expected to decrease, gradually shifting from a positively skewed log-normal distribution—typical of populations with relatively low average food consumption—toward symmetric normal distributions. Both normal and log-normal skewness families are allowed to represent all intermediate degrees of positive skewness (see http://www.fao.org/3/a-i4046e.pdf for further details). A user-developed R function is available from FAO’s Statistics Division to compute estimates of the prevalence of undernourishment, accounting for four parameters: dietary energy consumption (DEC), coefficient of variation (CV), skewness (Skew), and minimum dietary energy requirement (MDER). Different data sources may be used to estimate the various model parameters. ⸻ Dietary Energy Consumption (DEC) The mean of the distribution of calorie consumption levels for the representative individual in the population (DEC) corresponds, by definition, to the per capita daily food consumption of the population. DEC may be estimated using food consumption data obtained from surveys that are representative of the population group of interest. Depending on survey design, such data may allow estimation of DEC at national or subnational levels, or by geographic areas and socio-economic population groups. Unfortunately, although the situation is improving rapidly, representative food consumption surveys are still not available for every country and cannot always be conducted annually. For the national population, DEC may also be estimated using information on total food supply and utilization of all food commodities within a country, where the contribution of each commodity to food availability for human consumption is expressed in dietary energy terms. The sum of these contributions is then divided by the total population. The primary data source for national food balances is the Food Balance Sheets (FBS), maintained by FAO for most countries worldwide (see http://www.fao.org/economic/ess/fbs/en/), based on official data submitted by Member States and disseminated through FAOSTAT (http://faostat3.fao.org/download/FB/*/E). ⸻ Coefficient of Variation (CV) Surveys containing information on food consumption at the individual and household levels are the only sources capable of directly estimating the coefficient of variation (CV) of habitual food consumption for individuals in the population. However, food consumption survey data are subject to numerous limitations that complicate reliable estimation of CV. In principle, repeated observations of daily intake for each individual in the sample are required to estimate habitual consumption levels and measurement error. Additionally, data should be collected from the same individuals or households across different seasons to capture possible seasonal variation in calorie consumption. Due to their high cost, national surveys of individual food consumption with these characteristics are extremely rare and virtually absent in most developing countries. As a result, the most commonly used data sources for estimating CV are multi-purpose household surveys—such as Living Standards Measurement Studies, income and expenditure surveys, or household budget surveys—which also collect food consumption information. When using household-level data, careful attention must be paid to differences between food acquisition and actual consumption (including waste) during the reporting period, as well as accurate recording of the number of individuals sharing consumption. Moreover, household-level data conceal variability arising from intra-household food distribution. For these reasons, the coefficient of variation calculated from average per capita daily calorie consumption recorded for each household in a survey is not a reliable estimate of CV, which should reflect differences in habitual (rather than random) daily calorie consumption at the individual (rather than household) level. Empirical CV estimates derived from household survey data tend to be inflated due to spurious variability caused by measurement error, differences between random and habitual intake, discrepancies between acquisition and actual consumption, and seasonality. In addition, they do not reflect variability related to individual characteristics of household members (e.g. sex, age, body weight, and physical activity level). Therefore, when using household survey data, CV is best estimated indirectly by accounting for spurious variability and adjusting for inter-individual (in addition to inter-household) variability. The simplest approach is to classify households into homogeneous groups and calculate the coefficient of variation of average per capita daily calorie consumption across these groups. This yields an estimate of the inter-household component of CV, denoted as CV_H. The inter-individual component, denoted CV_I, is estimated for each population group based on its sex, age, and body weight composition. The two components are then combined as follows: \hat{CV} = \sqrt{(CV_H)^2 + (CV_I)^2} For countries and years without household survey data, an indirect estimate of CV (CV_IND) is obtained through regression analysis linking CV to GDP per capita, the income Gini coefficient, the relative food price index (FPI), and regional effects (REG): \hat{CV}_{IND} = \beta_0 + \beta_1 GDP + \beta_2 GINI + \beta_3 FPI + \beta_4 REG Regression coefficients are estimated using a dataset covering countries and years for which information on CV, GDP, GINI, and FPI is available. ⸻ Skewness Because skewness is not strongly affected by spurious variability, it is estimated directly from household-level average daily calorie consumption data, with the sole adjustment of removing rare extreme high or low values. If the empirically estimated skewness exceeds the value corresponding to a log-normal distribution with the given mean and coefficient of variation, the parameter is disregarded and a two-parameter log-normal distribution is used for f(x) (see http://www.fao.org/3/a-i4046e.pdf for further details). ⸻ Minimum Dietary Energy Requirement (MDER) Human energy requirements are calculated by multiplying normative basal metabolic rate (BMR) requirements (per kg of body weight) by the ideal body weight of a healthy individual of given height, and then by a physical activity level (PAL). Thus, ranges of normal dietary energy requirements are calculated for each sex and age group, recognizing that a range of body mass index (BMI) values—from 18.5 to 25—corresponds to healthy status. Consequently, any given height may correspond to a range of healthy body weights and, therefore, to a range of BMR values. Taking into account average height and the possibility of different physical activity levels within groups, minimum, average, and maximum dietary energy requirements are calculated for each sex and age category, with additional allowances for growth among individuals aged 0–21 and for women during pregnancy and lactation (see ftp://ftp.fao.org/docrep/fao/007/y5686e/y5686e00.pdf). The MDER for a given population group, including the national population, is obtained as the weighted average of the lower bounds of the energy requirement ranges for each sex and age group, using the population size of each group as weights. In estimating the prevalence of dietary energy inadequacy, confusion has often arisen between MDER and the average dietary energy requirement (ADER), and the appropriate threshold for calculating the probability of inadequacy. The probability of dietary energy inadequacy must be calculated relative to MDER rather than ADER (which may instead be used as an estimate of the population’s average recommended intake), because populations naturally exhibit variability in energy needs. Using ADER as a threshold would substantially overestimate undernourishment by including healthy individuals whose intake is below the average simply because their requirements are lower. Where needed, ADER should be used to compute calorie consumption gaps. ⸻ Disaggregation Because estimates of average calorie consumption rely on national food balance sheet data, global monitoring of MDG 1C and the World Food Summit target is based solely on national-level estimates of the prevalence of undernourishment. In principle, the indicator may be calculated for specific population groups provided sufficiently accurate information is available to characterize model parameters for those groups—namely, data on food consumption, age and sex structure, and possibly physical activity levels. Thus, the scope for disaggregation depends critically on the availability of surveys representative at subnational population levels. Given current household survey practices, reliable information beyond macro-regional (urban/rural) residence and major administrative regions is rarely available. Since most surveys are designed primarily to measure income distribution, undernourishment estimates may be derived for different income groups. Gender disaggregation is limited to identifying and grouping households by gender-related characteristics (e.g. sex of household head or male–female composition). ⸻ Treatment of Missing Values At the country level: When recent household food consumption survey data are unavailable, modeled estimates of PoU rely on DEC derived from food balance sheets, indirect estimates of CV based on GDP per capita, income Gini coefficient, relative food price index or other development indicators (such as under-five mortality rates), and MDER estimates based on UN Population Division World Population Prospects data. At regional and global levels: Missing values for individual countries are implicitly imputed as the population-weighted average of estimated values for countries within the same region. ⸻ Regional Aggregates Regional and global PoU estimates are calculated as: PoU_{REG} = \frac{\sum_i PoU_i \times N_i}{\sum_i N_i} where PoU_i represents undernourishment prevalence for each country with reliable data in the region, and N_i denotes the corresponding population size. ⸻ Sources of Discrepancies Many countries have produced and reported estimates of undernourishment prevalence, including in national MDG reports, but almost invariably using methodologies different from that developed by FAO, rendering national figures incomparable with FAO estimates used for global monitoring. The most common approach in national reporting has been to calculate the share of households whose average per capita daily calorie consumption falls below fixed thresholds—typically 2,100 kcal—based on household survey data. In some cases, lower thresholds around 1,400 kcal were used, likely in response to implausibly high undernourishment estimates produced by the higher cutoff. In nearly all cases, excess variability in calorie consumption data is ignored, resulting in reports that show limited or no progress in reducing undernourishment over time. As noted in the methodological section, results derived from these alternative approaches are highly unreliable and almost certainly biased upward. Therefore, coordinated efforts are needed to promote the use of FAO methodologies in national reporting, and FAO stands ready to provide the necessary technical support.
