Dear Jayati,
The task is indeed a difficult one, given the availability of data in various countries. What you describe for India is very common in other countries too (and could even be considered a particularly fortunate situation, compared to cases where household surveys are less frequent and less detailed). Your observation confirms the fact that a method entirely based on household survey data (as advocated, for example, by IFPRI a few years ago) would be fraught by many almost insurmountable problems and prone to serious estimation errors for many countries.
The method developed by Sukhatme in the sixties, and later perfected by Naiken, is from all points of view a brilliant, sophisticated solution to get at something that is, admittedly, far from yielding a perfect “count” of the number of people suffering from chronic undernourishment, but that is certainly good enough for the purpose at hand of measuring progress on a global scale.
The qualifying aspect of the FAO method is that it uses information that comes from different independent data sources, such as aggregated national food balances, socio demographic accounts of the populations, and - where available – food consumption survey data, which has allowed FAO to made the assessment for so many countries during all these years.
In short, we postulate a probability distribution for the habitual daily dietary energy consumption of the “average” individual in the population, representing it with a parametric density function, and estimating its parameters based on information available from the above mentioned sources. Unfortunately - as some of Thomas’ observations reveal - the nature of the probability distribution and of the associated inference process is still largely misunderstood and under-appreciated, whereas it is, in fact, our best guarantee to avoid gross mistakes.
First, one must notice that the distribution we use is not simply the empirical distribution of food consumption in a population. It rather refers to the “average individual”, which is a statistical device used to represent the entire population. This is necessary, because we do not have the kind of data that would allow to match food consumption with likely requirements at the individual level, and therefore we can never apply what could be described as a “headcount” approach. What this implies is that as, opposed to what one may think appropriate for an actual individual, for such “average” individual there will be not a single value, but rather an entire range of values of daily energy requirements that are consistent with an active and healthy life. The range is induced by the existing variability in the population by sex, age, physical activity levels and metabolic efficiency.
Second, one must consider that the observed variability in dietary energy consumption in the population confounds three different aspects:
a) measurement errors in individual consumption (due, for example, to the fact that, as you mention, usually food consumption data are collected at household, not individual level, which is perhaps the lesser problem, considering also that surveys often miss food consumed away from home, or that there are problems in misreporting consumption)
b) the variability due to the fact that people differ in their food requirements, and, finally
c) the fact that some people consume either in excess or, what we are interested in, short of their requirement.
It is precisely as a means to separate the three sources of variability, to control for the other two and to focus on the third one, that the FAO method has been devised.
The use of an independent source for the average per capita dietary energy consumption is intended, for example, is meant to control for systematic errors that may affect the level of average per capita dietary energy consumption as measured through surveys (a recent paper demonstrates how the bias in measuring per capita caloric consumption can be of up to 800 kcals per day, only depending on the particular data collection module used).
The detail treatment of household food consumption data from survey (described for example in our recent book on “Analyzing food security using household survey data” ) is also intended to control for measurement errors and to improve on our estimate of the parameters that describe the variability of habitual food consumption levels.
Finally, and what has perhaps created the most trouble in understanding the method, because the estimated distribution will still include variability that has nothing to do with food insecurity, but that simply reflect the diversity of the population, the proper threshold to be used for estimating the prevalence of inadequate food consumption is the minimum of the range of normal requirements. The fact that such a minimum is associated with the level that correspond to the average requirements for a sedentary lifestyle has led many to mistakenly believe that the method would fail to count among the undernourished those who are undernourished even if consuming more than such a minimum level. Nothing is farer from the truth. The use of the MDER is only the way in which we control for factors that affect variability in food consumption but have nothing to do with food insecurity. Failing to do so would be akin to an experimenter who fails to control for exposure to sun, and who would attribute to different fertilization levels the difference in growth rate of plants that have grown in the sun or in the shade.
I thank you for your query, and I hope these few notes help in grasping the essence of the method.
Carlo
Carlo Cafiero, PhD
Project Manager of “Voices of the Hungry”
Statistics Division (ESS)
Food and Agriculture Organization of the United Nations (FAO)
The task is indeed a difficult one, given the availability of data in various countries. What you describe for India is very common in other countries too (and could even be considered a particularly fortunate situation, compared to cases where household surveys are less frequent and less detailed). Your observation confirms the fact that a method entirely based on household survey data (as advocated, for example, by IFPRI a few years ago) would be fraught by many almost insurmountable problems and prone to serious estimation errors for many countries.
The method developed by Sukhatme in the sixties, and later perfected by Naiken, is from all points of view a brilliant, sophisticated solution to get at something that is, admittedly, far from yielding a perfect “count” of the number of people suffering from chronic undernourishment, but that is certainly good enough for the purpose at hand of measuring progress on a global scale.
The qualifying aspect of the FAO method is that it uses information that comes from different independent data sources, such as aggregated national food balances, socio demographic accounts of the populations, and - where available – food consumption survey data, which has allowed FAO to made the assessment for so many countries during all these years.
In short, we postulate a probability distribution for the habitual daily dietary energy consumption of the “average” individual in the population, representing it with a parametric density function, and estimating its parameters based on information available from the above mentioned sources. Unfortunately - as some of Thomas’ observations reveal - the nature of the probability distribution and of the associated inference process is still largely misunderstood and under-appreciated, whereas it is, in fact, our best guarantee to avoid gross mistakes.
First, one must notice that the distribution we use is not simply the empirical distribution of food consumption in a population. It rather refers to the “average individual”, which is a statistical device used to represent the entire population. This is necessary, because we do not have the kind of data that would allow to match food consumption with likely requirements at the individual level, and therefore we can never apply what could be described as a “headcount” approach. What this implies is that as, opposed to what one may think appropriate for an actual individual, for such “average” individual there will be not a single value, but rather an entire range of values of daily energy requirements that are consistent with an active and healthy life. The range is induced by the existing variability in the population by sex, age, physical activity levels and metabolic efficiency.
Second, one must consider that the observed variability in dietary energy consumption in the population confounds three different aspects:
a) measurement errors in individual consumption (due, for example, to the fact that, as you mention, usually food consumption data are collected at household, not individual level, which is perhaps the lesser problem, considering also that surveys often miss food consumed away from home, or that there are problems in misreporting consumption)
b) the variability due to the fact that people differ in their food requirements, and, finally
c) the fact that some people consume either in excess or, what we are interested in, short of their requirement.
It is precisely as a means to separate the three sources of variability, to control for the other two and to focus on the third one, that the FAO method has been devised.
The use of an independent source for the average per capita dietary energy consumption is intended, for example, is meant to control for systematic errors that may affect the level of average per capita dietary energy consumption as measured through surveys (a recent paper demonstrates how the bias in measuring per capita caloric consumption can be of up to 800 kcals per day, only depending on the particular data collection module used).
The detail treatment of household food consumption data from survey (described for example in our recent book on “Analyzing food security using household survey data” ) is also intended to control for measurement errors and to improve on our estimate of the parameters that describe the variability of habitual food consumption levels.
Finally, and what has perhaps created the most trouble in understanding the method, because the estimated distribution will still include variability that has nothing to do with food insecurity, but that simply reflect the diversity of the population, the proper threshold to be used for estimating the prevalence of inadequate food consumption is the minimum of the range of normal requirements. The fact that such a minimum is associated with the level that correspond to the average requirements for a sedentary lifestyle has led many to mistakenly believe that the method would fail to count among the undernourished those who are undernourished even if consuming more than such a minimum level. Nothing is farer from the truth. The use of the MDER is only the way in which we control for factors that affect variability in food consumption but have nothing to do with food insecurity. Failing to do so would be akin to an experimenter who fails to control for exposure to sun, and who would attribute to different fertilization levels the difference in growth rate of plants that have grown in the sun or in the shade.
I thank you for your query, and I hope these few notes help in grasping the essence of the method.
Carlo
Carlo Cafiero, PhD
Project Manager of “Voices of the Hungry”
Statistics Division (ESS)
Food and Agriculture Organization of the United Nations (FAO)