The Statistics Division held a Technical Meeting (ESS-TM) on "The Major Limitations / Problems Related to the Accuracy of the Measurement of Food Deprivation"
Saturday, 29 June 2002.

The objective of this ESS-TM was to discuss selected issues related to the referred subject, in particular considering the inputs from the International Scientific Symposium (ISS) held in FAO from 26 to 28 June 2002. The agenda is organized as follows:

• Need for clarification on the concepts used for the variables under measurement.
• Methodological issues requiring improvement of further research.
• Data limitations and improvement perspectives.

Summary Record of Meeting on "Major Limitations/Problems Related to the Accuracy of the Measurement of Food Deprivation"
held in FAO - Rome
on Saturday 29 June, 2002 from 9.00 to 13.00 hrs

A. Introduction

In opening the meeting Mr. Haluk Kasnakoglu, Director, Statistics Division, indicated that the main purpose of the meeting was to discuss certain issues relating to FAO's estimation of the prevalence of undernourishment in the light of the inputs from the International Scientific Symposium (FAO, Rome 26-28 June 2002). In particular the aim was to consider the actions that ESS should take in attempting to improve the procedure for estimating the prevalence of undernourishment.

The meeting was chaired by Mr. J. Mernies, Chief of Statistical Analysis Service, Statistics Division. In his introductory statement the Chairman admitted that the agenda was long and therefore it may not be possible to consider all the points during the limited time available. In view of this he indicated the issues that should be given priority. The list of participants is in the Appendix.

B. Need for clarification on the concepts used for the variable under measurement

Under this item the meeting focused on the terminology to be used in referring to the FAO measure of food deprivation given the fact that many participants at the Symposium felt that the terms "undernourishment" "undernutrition" or "hunger" were not appropriate. In general it was agreed that the term "Dietary Energy Deficiency (DED)" was more appropriate as the measure refers to those with dietary energy consumption below requirement. However, it was recognized that in popular statements and for the purpose of advocacy the term "hunger" is more likely to be used.

C. Methodological Issues requiring improvement or further research

Under this item a number of issues were considered as reported below.

i) The bivariate distribution versus cut-off point approach issue
The issue arose in the context of an argument by Prof. Svedberg (2001) questioning the validity of the FAO basic methodological framework for estimating the prevalence of dietary energy deficiency, which is based on the specification of the frequency distribution of intake and a cut-off point reflecting the lower limit of the requirement distribution. Prof. Svedberg's argument was based on an application of the bivariate probability distribution model under the assumption that the joint distribution is lognormal and intake is correlated with requirement. On the other hand Mr. Naiken has shown that if the effect of correlation is appropriately taken into account in the same model the result should be the same as the FAO cut-off point approach.

During the discussion it was agreed that the bivariate distribution is the appropriate framework for mathematically expressing the proportion of the population with intake below their respective requirements, although under the assumptions considered by Mr. Naiken the bivariate formula reduces to the cut-off point formula. However, the problem is that the actual shape of the bivariate distribution of intake and requirement is unknown and the empirical specifications that have been used in evaluating the formula are too simplistic to yield realistic results. Further research is needed in order to arrive at better specifications of the bivariate distribution. Meanwhile FAO should continue to use the cut-off point formula which can be justified as reflecting an attempt to minimise the risk of overestimation. However it has to be recognized that the approach is subject to a risk of under-estimation, the extent of which cannot still be assessed.

ii) Sampling design issue in relation to frequency distribution data from household surveys
This issue arose out of the consideration that the objective of most sample surveys is to estimate the mean and not the other parameters that determine the frequency distribution. Consequently the sample designs usually adopted in these surveys lead to biased estimates of the variance and the higher moments of the distribution. In this connection Mr. David had expressed the view that in the context of estimating the proportion of households below certain fixed cut-off point the issue of bias in the variance or higher moments of the distribution does not arise as one would normally estimate the proportion directly from the household level data in the sample. Consequently the resulting estimate should be treated just as a mean or a proportion.

During the discussion it was however explained that the proportion estimated directly from the sample households' data reflects an inference on the cumulated frequency distribution or percentile distribution. Although this distribution is non-parametric it cannot be claimed that it is free from the effect of the sample design bias referred to above.

iii) Feasibility of improving the definition of the cut-off point (minimum energy requirement)
It was generally agreed that the cut-off point should be improved taking into account the recommendations of the Joint FAO/WHO/UNU Expert Consultation on Energy in Human Nutrition that was held in October 2001. In this connection Mr. Shetty reported that, although the report of the Consultation has not yet been finalized, it seemed clear that a) while the energy requirement of children have been scaled down to some extent, those for the adults have risen and b) the physical activity level (PAL) index applied to the BMR in order to arrive at energy required for light activity is likely to be specified as ranging from 1.40 to 1.69 as opposed to the past approach of giving a fixed value.

It was noted that the new approach regarding PAL for light activity poses a practical problem in defining the cut-off point. From the point of view of FAO's aim of minimizing the risk of overestimation, the lower limit corresponding to 1.4 BMR should be adopted as the cut-off point. However, all other things being equal, this would imply a significant reduction in the proportion as well as the number of undernourished with obvious negative consequences in so far as comprehension of the changes by the public is concerned. On the other hand, the latter problem could be avoided by opting for the average of the range which would actually boil down to the currently used level (1.55 BMR). This would however imply an inconsistency with the basic principle justifying the use of the cut-off point formula. In this connection it was argued by Mr. Vercueil that after all the choice of the cut-off point is generally viewed as being rather arbitrary so that not much harm will result from sticking to the currently used one.

It was evident that the pros and cons of the two alternatives should be taken into account in deciding on the PAL value to be used.

iv) Need to revise/up-date the CV as data become available
It was overwhelmingly recommended that, for the sake of credibility in the estimates, the CV should be revised and up-dated regularly as new survey data become available. In the case of the countries where the CV was estimated by regression models this would imply re-estimation on the basis of revised models. This recommendation implied that not only the presently used CV be revised as necessary but also that any change in the CV over time be taken into account.

v) Feasibility of allowing for changes in time in the CV
While recognizing the need to take into account changes in the CV over time, it was clarified that its feasibility would depend on: a) the availability of time series data referring to the whole time-span covered by the FAO estimates (i.e. for 1969-71 to current period) and b) the time and effort that would needed to properly undertake this work.

vi) Frequency of up-dating the estimates of undernourishment
In this context the general view was that FAO should refrain from annually up-dating the national estimates particularly in the light of the backward revisions necessitated by revisions in the basic data used for estimation. Furthermore the errors that the basic data are subject to may be more that the annual changes shown. The general feeling was that a three-year cycle of up-dating the national estimates with reference periods corresponding to the period when the three-year average of the per caput DES do not overlap would be more appropriate for reflecting changes over time. However, global and regional figures would continue to be reported on an annual basis.

vii) Consideration of the effect of inter-household variation in sex-age composition on the variation in energy requirements
This issue which was raised by Prof. Svedberg was relevant in the context of the Sixth World Food Survey where the CV was estimated on the basis of the distribution of household per caput food consumption data provided by household surveys. As the inter-household variation includes the variation due to differences in sex-age composition, it was logical to expect that this should be taken into account in defining the range of variation in household per caput requirement and hence deriving the cut-off point.

However, the procedure for estimating the CV has been significantly modified since the Sixth World Food Survey due to the problems associated measurement errors in the household level data and sample survey design. In the currently used procedure, the CV of per caput household dietary energy consumption is defined as being composed of two components of variation: one due to income and the other due to requirement. The income component is estimated on the basis of survey data referring to the average per caput consumption groups of households classified by income. As a consequence of this simplification, the effect of differences in sex-age composition of the households is by definition ignored in deriving the distribution of intake. In view of this the issue of taking into account the effect of differences in sex-age composition of households in determining the variation in energy requirements does not arise.

D. Data limitation and improvement perspectives

i) Feasibility of reconciling FBS estimates of mean with household survey estimated mean
It was agreed that as part of the efforts to improve the FBS estimates of the mean food consumption level, the available household survey data pertaining to per caput quantities of food items purchased/acquired by households should be confronted with the corresponding information derived through the supply/utilization accounts (SUA). This would provide a means of calibrating the SUA estimates. It was also recommended the FAO Publication entitled "A Comparative Study of Food Consumption Data from Food Balance Sheets and Household Surveys" (1983) be up-dated.

ii) Feasibility of up-dating the height data provided in the James/Schofield book
Mr. Shetty reported that the height data published in the James/Schofield book ( "Human Energy Requirements: A manual for Planners and Nutritionists" by W.P.T. James and E.C. Schofield, Oxford University Press, 1990) will be revised/updated and published in a FAO report dealing with the practical application of the principles/norms recommended by the recent FAO/WHO/UNU Expert Consultation on Energy in Human Nutrition.

APPENDIX

List of Participants