ZEF Bonn
Zentrum für Entwicklungsforschung
Center for Development Research
Universität Bonn
Doris Wiesmann, Joachim von Braun, Torsten Feldbrügge
Number 26

ZEF -- Discussion Papers on Development Policy
Bonn, April 2000

At international conferences in the 90s, the international community reconfirmed the elimination of hunger and malnutrition as high priority. The Nutrition Index (NI) presented in this paper has been designed to assess the effectiveness of actions taken towards achieving this goal. The Nutrition Index reflects complementary dimensions of a country’s nutrition situation by including the following indicators: 1) the percentage of undernourished in the population, 2) the prevalence of underweight in children under the age of five, and 3) the under-five-mortality rate, indicating the deadly synergy between inadequate food intake and unhealthy living conditions.

After data refinement and several estimation procedures to supplement lacking data, data availability permits NI calculation for the vast majority of developing countries and at different points in time (1981, 1992 and 1997). The NI serves as a comprehensive measure to analyse performance and trends in combating hunger and malnutrition in single countries and in regions. In North Africa and the Near East, as well as in Latin America and the Caribbean, a nearly satisfactory nutrition situation has been achieved. The nutrition situation is still bad in South Asia and no better than mixed in most of Southeast Asia, but very promising upward trends have been observed in this region since the beginning of the 80s. In large parts of Sub-Saharan Africa, the nutrition situation is bad or even extremely bad, and recent trends do not provide much scope for optimism either.

In further analyses, significant correlations between NI values and the prevalence of micro-nutrient deficiencies are found, and the tight connection of macroeconomic performance and people’s ability to be free from hunger and malnutrition is demonstrated. In spite of this, the comparison of NI and GNP per capita also reveals that much economic scope for policies to relieve hunger and malnutrition still remains untapped in many countries. The Nutrition Index, in its function as an international monitoring tool, is hoped to make countries more accountable to their commitments and to help speeding up necessary nutritional improvements.

Notable progress has been made in recent decades in combating hunger and malnutrition. Nevertheless, considering the goal of the United Nations’ World Food Conference in 1974 - that ‘no child should go to bed hungry’ within a decade - the gap between vision and reality is striking: at the end of the 90s, about 800 million people in the developing world were suffering from hunger, that is, they could not meet their minimum dietary energy requirements (FAO 1999a). In 1995, 167 million children under five were malnourished as measured according to their weight for age, and even more -- 206 million -- showed signs of growth faltering, while 49 million were wasted (WHO 1997). About half of the 11.6 million deaths among children under five in 1995 in developing countries are estimated to be associated with malnutrition (Bailey et al. 1998). At least 40 million children in the developing world suffer from Vitamin A deficiency, and the dietary iodine supply of nearly one third of the world population is at risk (WHO/UNICEF/ICCIDD 1993; WHO/UNICEF 1995). Two billion people in both developed and developing countries are affected by iron deficiency (FAO/WHO 1992a). These figures indicate a largely unsatisfactory global nutrition situation and provide strong arguments for renewing national and international commitment to improve nutrition.

At the World Summit for Children in 1990, the International Conference on Nutrition in 1992 and the World Food Summit in 1996, the international community reconfirmed the elimination of hunger and malnutrition as priority of utmost importance (UNICEF 1990a; FAO/WHO 1992b; FAO 1998). To assess the effectiveness of actions taken by national governments, NGOs and the private sector towards achieving these goals and target scarce resources to the neediest, there is a need to quantify and continuously monitor changes in a country’s or region’s nutrition situation. First steps were taken by UN agencies concerned with nutrition: FAO, UNICEF and WHO, however, use different lead indicators for which they regularly collect and publish data. So far, there has been no widely accepted single, comprehensive measure for a country’s or region’s nutrition situation.

The Nutrition Index (NI) presented in this paper has been designed to fill this gap. The NI comprises the percentage of undernourished, the proportion of underweight children under five and the under-five-mortality rate. The combination of several indicators in one index has obvious advantages: in contrast to the multitude of single indicators that are currently used (and that have merits of their own), the NI allows for a quick international and intertemporal overview because of its condensed information content. The NI consists of indicators that reflect different, but complementary aspects of nutrition problems that should be viewed together.

In the following, the concept and theoretical underpinning of the Nutrition Index are presented first. In order to transparently derive the NI, methodology and data refinement procedures are described at length (chapter 3). Actual NI scores and outcomes of analyses with respect to micro-nutrient deficiencies and basic determinants are reflected in detail. The reader just interested in the results and their implications is referred to chapters 4 and 5. Here, the performance of each country in terms of overcoming hunger and malnutrition is presented and discussed. It is hoped that such information on success and failure stimulates renewed policy action in under-performing countries and continued attention to the problem in those countries which show a laudable track record.

The purpose of the NI is the quantitative description of a country’s or region’s nutrition situation. Prior to any actual index calculation, decisions have to be made about the set of indicators that are to be included in the index, and about the mode of aggregation. These decisions have to be taken with care, because the set of indicators used and their aggregation have a strong impact on resulting index values. The choice of indicators should be guided by an idea of how a good nutrition situation can be characterised. The following criteria are proposed here for the judgement of a country’s nutrition situation: Vast majorities of the population, especially vulnerable groups like children,

• are free from hunger (have adequate dietary energy supply);
• have good nutrition status (as measured by anthropometrics);
• and are safe from death caused directly or indirectly by malnutrition.

Clearly, the nutritional well-being of the individual is at the centre of this concept. Besides the definition of an adequate nutrition situation, a basic understanding of the interrelationship of the main variables expressed by specific indicators is needed for an informed choice of indicators (which is in fact limited by data availability). Furthermore, the analysis of factors such as economic performance and their effect on the nutrition situation requires a concept of the causal relationships and pathways. In the following, an appropriate conceptual framework is outlined on which the choice of indicators and mode of aggregation are based.

The conceptual framework is presented in Figure 1. A distinction is made between three levels of causality corresponding to basic, underlying and immediate causes of hunger and malnutrition, and malnutrition-related deaths. In accordance with the above definition of an adequate nutrition situation, the focus is on immediate causes and effects at the individual level and the outcomes for individuals, though the pathways leading from basic to underlying and from underlying to immediate causes will be briefly described.

The basic determinants of hunger and malnutrition at the national level relate to the interacting fields of economy and technology, policy and culture, ecology and natural resources: scarcity of natural resources, lack of technology and economic assets, market and policy failure (including violent political conflicts and discrimination) lead to low macro-economic performance. The result is widespread poverty, especially in the absence of mitigating social policies. Poverty seriously interferes with household food security and caring capacity ¹ in the family and community, impedes investment in education and utilisation of health services. Furthermore, household food security is directly affected by the functioning of local markets, the use of technology in households (e.g. for food conservation), agro-ecological factors such as soil fertility and climate, and by policy measures such as food for work programs or food subsidies. The acquisition of knowledge is influenced by education policy and public health campaigns. Women’s education and their caring capacity are frequently hampered by culturally embedded and politically reinforced discrimination. Time resources as another important factor for care-giving are determined by the workload and division of labour, available technologies (for example, firewood-saving stoves), natural resources (e.g. water) and cultural habits. Health environments are shaped by the safety of workplaces, natural conditions and policy outcomes like public investments in the provision of water, sanitation, health care and public health programs such as immunisation campaigns (von Braun et al. 1998; Smith and Haddad 1999).

The impact of the underlying determinants (household food security, knowledge and caring capacity, and health environments) on the immediate causes of dietary intake and health status is obvious: what kind of food, and how much, is accessible to an individual household member is determined by the availability of food in the household, by the recognition of own needs and the needs of others, and by the caring capacity of the family members responsible for food preparation and feeding of children. Knowledge, time and material resources other than food are necessary to care for oneself and others by preventing and curing infectious diseases (UNICEF 1990b). The exposition to pathogens and therefore probability of illness is partly a function of the health environment. Depending on the living conditions as determined by food security, knowledge, caring capacity and the health environment, the individual can get trapped in a vicious circle of inadequate dietary intake and bad health status. The starting point can be either energy, protein, and/or micro-nutrient intake falling beneath requirements, or infection and bad health status. The consequences of lack of dietary energy and protein and of the most prevalent micro-nutrient deficiencies ² are shown in Table 1, with special reference to the effects mentioned in Figure 1. The consequence of acute insufficient dietary energy intake is reduced body mass (wasting). Moreover, chronic lack of dietary energy, protein and some micro-nutrient deficiencies cause growth faltering in children (stunting). Both wasting and stunting are reflected in low body weight with reference to a child’s age (underweight). Other effects of dietary energy, protein and/or some micro-nutrient deficiencies are reduced functioning of the immuno system and damages to mucous membranes. Barriers against infection are no longer intact, and resistance to pathogens is weakened. The risk of viral and bacterial infections rises, and a bad nutrition status prior to infection prolongs the duration and aggravates the severity of the infectious disease. Loss of appetite is a frequent consequence of infection, which in turn reduces dietary intake. Physiological changes specific to many infectious diseases are energy-consuming fever and catabolism of body protein that enhance the need for dietary energy and protein (Tomkins and Watson 1989).

* mostly energy and protein deficiency are combined, as the co-called Protein-Energy-Malnutrition (PEM)
Source: Biesalski 1995, Spittler et al. 1995, Barth et al. 1995, de Gruyter 1998

The final outcome for the human being depends on genuine individual factors, care, health and nutrition interventions from the outside, and on the intensity of adverse effects in the vicious circle. At best, the individual gains full recovery after some time of suffering. Partial recovery is another possible outcome; in this case reduced body size in adulthood and irreversible damage to mental capacity, blindness or other disabilities are a remaining characteristic of the person affected.³ Premature death is the most serious outcome of the vicious circle (Pinstrup-Andersen et al. 1995).

The insights into the immediate causes, effects and final outcomes of hunger and malnutrition are integrated next into the choice of indicators.

The choice of indicators is guided by the concept of what constitutes an adequate nutrition situation and the framework just outlined. As the conceptual focus is on the nutritional well-being of the individual, the following indicators are selected:

• the percentage of undernourished as estimated by FAO, reflecting the share of the population with insufficient dietary energy intake,
• the percentage of underweight children under the age of five, indicating the proportion of children suffering from weight loss and/or reduced growth,⁴
• the under-five-mortality rate, reflecting to a large extent the fatal consequence of the synergy between inadequate food intake and bad health environments.

A comparative characterisation of the three indicators is given in Table 2. It is obvious that they refer to different, but complementary aspects of a bad nutrition situation. A common feature can be seen in the fact that undernourishment, underweight in children and child mortality are more or less strongly associated with micro-nutrient deficiencies. Even though no indicator for vitamin or mineral deficiencies was included in the index due to insufficient data availability, the NI is likely to reflect the lack of micro-nutrients to a certain extent.⁵

*vicious circle of inadequate dietary intake and bad health, see Figure 1.
Source: Tomkins and Watson 1989; Svedberg 1998; UNICEF 1998; Biesalski et al. 1995

The combination of the percentage of undernourished, that refers to the population as a whole, and the two indicators relating to children under five has more advantages as concerns the resulting index: both the nutrition situation of the whole population and the effects of inadequate nutrition (and concurrent factors) on a physiologically and socially very vulnerable group are captured by the NI. Besides the fact that children’s nutrition status deserves particular attention because malnutrition puts them at high risk of physical and mental impairment and death, relying on data about children is also justified from a methodological point of view: anthropometric data on children are favourable due to their international comparability. In contrast to the body size adults can reach, the growth potential of children under five most probably does not differ by ethnic origin (Svedberg 1998). Moreover, international data about the nutrition status of adults are not yet broadly available, whereas anthropometric data referring to children have been collected by WHO in a large, regularly updated database,⁶ so that data availability suggests their use.⁷ A shortcoming of anthropometric indicators can be seen in the fact that they do not reveal the most tragic consequence of undernutrition: premature death. Equal levels of malnutrition in a country as measured by low weight for age can have quite different effects on the proportion of malnutrition-related deaths among children, depending on the overall level of child mortality (compare the example of Niger and Yemen described in Box 1). This disadvantage of the percentage of underweight children is balanced by the inclusion of the under-five-mortality rate. Clearly, the mortality data comprise other causes of death than malnutrition, and the actual contribution of malnutrition to mortality is not easy to track, because the proximate cause of death is frequently an infectious disease (Pelletier et al. 1994). But about 54% of deaths among children under five in developing countries are estimated to be associated with malnutrition (Bailey et al. 1998).

Another type of complementarity between the indicators selected for the NI exists with respect to the kind of inequality in resource allocation that is considered.⁸ Differences between the indicators in the reflection of inter-household and intra-household allocation of food and other supplies are mainly rooted in the reference groups used and in the ways the data are derived. The primary source for the indicators referring to children are micro-data gained in representative surveys. FAO approaches the problem of quantifying undernourishment from the other side: macro-data like agricultural trade and production statistics and data or estimates about the national distribution of consumption are the point of departure. Concerns have been raised about the validity of these underlying data (Svedberg 1998). At least, errors in the underlying data are more likely to affect the absolute number of people estimated to be undernourished than the relative differences in the proportion of undernourished among countries or in succeeding periods of time; the latter are far more relevant for index calculation. On the other hand, the validity of the data about children is restricted by sampling and estimation errors, and, in the case of the under-five-mortality rate, partly dependent on the reliability of government statistics. Concerning data quality, there is certainly a lot of scope for improvement -- but the assessment of the nutrition situation should not be postponed until perfect data with global coverage are available.⁹

For the aggregation of the three indicators weighing factors have to be determined. Factor analysis was considered appropriate for the derivation of weights, because the relationships of the indicators suggest the existence of strong correlations, and because this technique allows the calculation of empirically founded weighing factors. The details concerning the aggregation of indicators are discussed in the next chapter together with other methodological issues, like data refinement and estimation procedures.

Before calculating NI scores, problems of data availability and comparability have to be tackled, and the aggregation of indicators has to be performed. Faced by a lack of direct nutrition-related information in some countries, ancillary criteria are defined in order to identify countries that are main candidates for nutritional surveillance. Thereafter, data gaps are filled by own estimations and estimates from other sources as far as possible. Refinement procedures are applied to the anthropometric data and reference periods are fixed before the data can actually be used in factor analysis and for NI calculation. The methodologies used are outlined in the following.

The NI cannot be calculated for every nation and any point in time due to lack of data for some countries and years. Whereas annual data for the under-five-mortality rate of all countries in the world are generally available, data availability for the percentage of undernourished and the proportion of underweight children is quite limited. For industrialised countries, virtually no data about undernutrition exist. Nutritional surveillance is no priority in most developed countries, because wealth and well-established social safety nets have led to a good nutrition situation for the overall population. The efforts of data collection and monitoring should focus on states where considerable proportions of the population are still suffering from hunger and are physically and mentally affected by malnutrition, and these countries are the main candidates for the following estimation procedures designed to close data gaps.

Thus, a first step undertaken in the construction of the index is the division of countries into those with problems of undernutrition and those without them. For countries with lack of direct nutrition-related data, the under-five-mortality rate and average dietary energy intake can serve as auxiliary criteria. A look at available data for average dietary energy intake (mean 1995-97) and undernourishment (1995-97) reveals that in the range of more than 3,100 kcal per capita and day, less than 2.5% of the population are considered to be undernourished in most countries, that is, they fall into the "very low hunger category" according to FAO (FAO 1999a).¹⁰ But taken as single auxiliary criterion, the national average of calorie consumption could easily conceal disparities in inter- and intra-household food allocation. Moreover, as has been outlined in the theoretical part, there is more to adequate nutrition than sufficient dietary energy intake. These facts are illustrated by the situation in some North African and Near Eastern countries - Egypt, Turkey and Syria - where average dietary energy intake even exceeds the mark of 3,100 kcal, whilst more than 10% of children under five are undernourished. At the same time, the under-five-mortality rate in these three countries amounts to more than 30 per 1,000 and is therefore considerably larger than in countries that have equal or even lower calorie consumption, but have reached a higher stage of economic development. In Japan, for example, the average citizen consumes less than 3,000 kcal per day, but the under-five mortality rate is as low as 6 per thousand live births, and there is no reason to assume malnutrition problems in this highly industrialised country.¹¹ Therefore, the under-five-mortality rate and average dietary energy intake are taken together for a rough assessment of the severity of nutrition problems. A country is assigned to the group with no severe nutrition problems, if average dietary energy intake exceeds 2,900 kcal per capita and day and the under-five-mortality rate is below 15 per thousand. These criteria are quite stringent in order to make sure that no country with persisting undernutrition escapes nutritional surveillance.¹² Of course, there are borderline cases, and it is accepted that some countries with lacking data, where nutrition problems might actually be smaller than assumed, may be erroneously classified as "countries with severe nutrition problems".

Estimating the percentage of undernourished for countries with probably severe nutrition problems, but missing data, is the next step.

Estimates of undernourishment are based on 3-year periods and have been published for 1979-81, 1990-92 and 1995-97 for developing countries only (FAO 1999a). In spite of rising concern for the nutrition situation in parts of Eastern Europe and in the countries of the former Soviet Union, no country estimates of undernourishment have been published yet by FAO for this region. For the Eastern European and Central Asian region and a few developing countries with less than one million inhabitants, estimates were therefore generated by a simple model to complement the data set.

FAO arrives at its estimates of the percentage of undernourished by using the average dietary energy intake, the distribution of food consumption according to actual consumption or income data, and the average minimum dietary energy requirement. If all this information was available for every country, the data set needed for the calculation of the NI could be completed following the methodology used and published by FAO. In the absence of some of these data, linear regression (ordinary least squares, OLS) was chosen as technique to produce estimates that are as close to the FAO figures as possible. To permit estimates for the bulk of countries with missing data, care had to be taken in the selection of indicators entered as independent variables into the regression model, to make sure that lack of data for explaining variables would not limit the expansion of the data set.¹³ Two readily and broadly available indicators likely to be strongly associated with the percentage of undernourished are GNP per capita and the average dietary energy consumption. Linear regressions were run with all possible combinations of these indicators, their logarithms and squares as independent variables (average values for the years 1995-97) and the percentage of undernourished as dependent variable (referring to the 1995-97 period). GNP per capita, its logarithm and square or combinations of these proved to be worse predictors than the variables derived from average dietary energy intake. Moreover, they had no additional explanatory power as against a set of the variables gained from dietary energy intake. The best fit, indicated by the highest adjusted R-square, was yielded by using a linear combination of the average dietary energy intake and its logarithm only. In a next step, regressions were run with the average dietary energy intake and its logarithm in the same way, but for different periods of time - with the variables referring to the years 1979-81 and 1990-92, respectively. The differences found between the coefficients in the three models that are based on different periods of time, and in the plots of the predicted values against the average dietary intake were marginal (see Figure 2).

To minimise the influence of random errors and to make sure that the final model is derived only from more recent data that are assumed to be of better quality than the 1979-1981 figures, country data for the years 1990-92 and 1995-97 were pooled for the final linear regression,¹⁴ again with the average dietary intake and its logarithm as independent variables and the percentage of undernourished according to FAO as dependent variable. The characteristics of this final model are shown in Table 3.

Note: Significance is two-tailed for both F and T-statistics
Source: own calculation

The significance of the overall model and of the singular coefficients and the constant proved to be very high, which is demonstrated by the significance levels given in Table 3. The probabilities that the whole model is invalid or that any of the coefficients or the constant is zero amounts to less than 0.005%.¹⁵ Average dietary intake and its logarithm, chosen as independent variables, predict 97% of the variation of the dependent variable (percentage of undernourished), which means that the model has an excellent goodness of fit. Of course, this is not surprising, given the fact that FAO derives the percentage of undernourished inter alia from average dietary intake and assumes a log-linear distribution of calorie consumption. The residuals produced by the model can be explained by the impact of specific countries’ food distribution data differing from the standard log-linear distribution on the percentage of undernourished and by the variation of average minimum dietary energy requirements. Though, as the residuals vary between a minimum of -6.6 and a maximum of 8.2, and the standard error is only 2.9, the influence of the latter factors is quite small.

If the predicted values were used for the aggregation of indicators instead of the actual figures published by FAO, the error introduced into the NI would be less than one percentage point for 73% of the countries in the sample (this is the case for all residuals smaller than 3, see also derivation of weights in this chapter - the percentage of undernourished accounts for one third of the NI).

The percentage of the undernourished as resulting from the model, the FAO figures and the residuals are plotted against the average dietary energy intake in Figure 3. The upward slope between 3360 and 3600 kcal per capita and day is likely to be a bias introduced by the fact that cultural traditions in some Near Eastern states like Turkey foster a high calorie intake at a moderate level of development while social inequalities persist.

In contrast to that, in a country like Korea, traditional dietary patterns tend to keep calorie consumption at 3,155 kcal per capita and day (in 1997) - a level that is still relatively modest taking into account Korea’s economic development - while the percentage of undernourished is very low, partly due to a well performing social system. Consequently, the model designed here should not be used to estimate the percentage of undernourished for developed countries with high calorie intake, because the results would be affected by the distortion of the predicted values in the upward slope. The predicted values for countries with a calorie consumption over 3,300 kcal per capita and day are irrelevant here anyway. After all, the original data set was supplemented by own estimates for 27 countries that were generated by the model and for 16 of these countries the estimates entered the NI. A complete list of countries with the percentage of undernourished given by FAO and the supplementary own estimates can be found in Table A-1 in the appendix.

The most serious data availability and comparability problems exist for the anthropometric data that were taken from the WHO’s regularly updated database (WHO 1997 and 1999) and, for the period before 1993, partly from data by the International Food Policy Research Institute (IFPRI) published in the Second Report on the World Nutrition Situation (UN ACC/SCN 1993). The main problem is that the nationally representative surveys these data are drawn from are conducted irregularly, and that for some countries, no data are available at all. The data set for underweight children was therefore supplemented by estimates published by UN ACC/SCN in 1993. Care was taken not to include estimated figures that showed an obvious misfit with existing data, e.g. predicted values with residuals of 10 and more. No estimates of the proportion of underweight children are available so far for the more recent years 1993 to 1998, but fortunately, a great number of surveys providing up-to-date information have been done in this period.

Besides the limited availability of data for given periods of time, another problem is posed by the fact that the anthropometric data reported in the nutrition surveys refer to different age groups of children (though the most frequently surveyed group are children under five years). The prevalence of underweight varies between groups of children at different age. For example, nutrition status is usually quite good up to an age of half a year, but dramatic deteriorations can be observed after the weaning phase due to inadequate dietary composition and feeding practices. For this reason, the comparison of the prevalence of underweight in children under five years in one country with the prevalence found for the age group 0.25 - <3 years in another country, for example, may overestimate the relative severity of malnutrition in the latter. Similarly, the intertemporal comparability of survey results based on different age groups can be affected (the left side of Table 4 shows the age groups used in the surveys that served as data sources for the NI and their share in the overall number of surveys).

This problem can be taken into account by multiplying the data referring to age groups other than the under-fives with corresponding correction factors, as done in the Second World Report on the Nutrition Situation (UN ACC/SCN 1993).¹⁶ For the purpose of calculating the Nutrition Index, the corrected data as given in the UN ACC/SCN report for the period before 1993 were adopted in this study, ensuring consistency with the estimates taken from the same publication. To improve data consistency over time and internationally, new factors were calculated from existing surveys for which malnutrition prevalence was reported separately for different age groups. In the selection of nationally representative nutrition surveys for this purpose, preference was given to countries with nutrition problems where underweight prevalence has also been reported for other age-groups than the under-five-year-olds. Special emphasis in this group was put on surveys from countries where proportions of underweight children higher than 6% are known. The latter selection criterion takes into account that errors introduced into the NI by data referring to the "wrong" age group might be the larger in absolute terms the higher the underweight prevalence is,¹⁷ with larger distortions of the NI scores as a result.

* as reported in the nutrition surveys (WHO 1997 and 1999; UN ACC/SCN 1993).
** surveys where no correction factor was applied, because prevalence in the respective age group proved to be close to identical with prevalence in under-five-year-olds, were attributed to the 0 - <5 years group.
Source: own calculation

The correction factors for a specific country and survey were obtained by dividing the underweight prevalences for the age groups in question (0 - under 3 years, 0.5 - under 3 years, 0 - under 4 years, and 0.5 - under 5 years) by the prevalence reported for all under-fives 18 . The box-plots in Figure 4 show the range, median and quartiles of correction factors found for different age groups. Evidently, the variation of factors among countries is high for the two age groups of under-three-year-olds (0-3 years and 0.5-3 years) and very low for the under-four-year-olds and the 0.5 - <5 group.

Correction factor patterns prove to be astonishingly stable over time for a given country, at least for countries where nutrition surveys are available for several years. A small time series of factors could be calculated for these countries (compare Figure 5 and Table A-2 in the appendix, that also reveal regional peculiarities of the South and Southeast Asian region).

In conclusion, the procedure described below was followed to improve data consistency:

• No correction factors were applied to data referring to the under-four age group, because the mean value of correction factors turned out to be one and the standard deviation proved to be very low, making it unlikely that large deviations from the underweight prevalence in under-fives might occur.
• Correction factors were always applied to data referring to the 0.5 - <5 group, because the calculated factors showed to be consistently greater than one with low standard deviation. In cases where a correction factor could be derived for the same country at a different point in time, this factor was used. If this was not possible due to lack of data, the average value of factors from countries of the same region was taken. Where neither factors for the respective country nor for its region could be calculated, the average of all factors was chosen (only in one case, for the Azerbaijan data).
• Taking into account the high variation of correction factors for the 0 - <3 and 0.5-3 age group even within regions and their average value of one, only factors derived from data for the same country were applied to avoid the enlargement of an existing bias with the real, yet unknown prevalence among under-fives.¹⁹ Therefore, correction was impossible for some countries, which does not necessarily mean that the underweight prevalence finally used for NI calculation deviates from the real prevalence among under-five-year-olds (because the unknown correction factor could also be exactly or nearly one).
• For some countries, anthropometric data referred to the 0.5 - <6 age group. In this case, first the prevalences for the 0.5 - <5 age group were calculated from the age-specific prevalences reported in the survey. Then, the correction factor for this group was applied to estimate the probable under-five prevalence. Differences found between prevalences in the under-five and under-six age group were so small that adjustment was considered unnecessary.

After this data refinement, more than 90% percent of the survey data expressed the underweight prevalence in the under-five age group (see Table 4, right side).

Gaps in available data could be filled and international comparability was improved by estimation and refinement procedures. Nevertheless, the data on undernourishment and the anthropometric data cannot be matched perfectly because they do not refer to the same years. In particular, the anthropometric data are spread over a wide time range. Therefore, anthropometric data for NI calculation cannot be drawn from one year, but the latest available information out of a fixed period of time has to be used. Justification can be found in the fact that quick changes in the percentage of underweight children rarely occur. There is, so to say, a built-in lag in this indicator: partly due to the fact that the eldest children in the sample are usually nearly five years old, and their weight for age is not only determined by current conditions, but also the outcome of their nutrition situation since and even before birth. Moreover, some factors that have been identified as relevant for children’s nutrition status -- for example female education and access to safe water (Smith and Haddad 1999) -- surely require long-term investments before showing an impact on a nation-wide level.

There is also empirical evidence for only slow changes in underweight prevalence (compare WHO 1997, pp. 40-42, where countries with more than one nutrition survey in the 1965-97 period are listed, together with the year of survey, reported prevalence and annual change rates for the time between the surveys). About 65% of the reported annual change rates are in the range of -1 to +1% points, and nearly 80% in the range of -1.5 to +1.5 % points. Only 7% of all observed change rates -- a total of six cases -- amount to an absolute value of more than 4% points per year. In this context, it should be noted that in five out of the six cases with relatively high annual change rates, these rates have been calculated from data out of surveys that were completed in subsequent years. Because inevitable measurement and sampling errors gain the more influence on calculated annual change rates the shorter the time period between the surveys is, these five rates are likely to be more distorted by random errors than the change rates computed from surveys with larger time intervals.²⁰

Taking into consideration that annual change rates in underweight prevalence are quite small for the majority of countries, six-year-periods have been chosen as reference periods, from which the latest available information about underweight prevalence in children was included into the NI. Furthermore, the bulk of data, including the estimates, refer to the more recent years in these periods (compare Figure A-1 in the appendix), e.g. more than three quarters of the data collected in the 1993-1998 period are from the years 1995-1998. Therefore, the latest information from 1993-1998 is considered approximately representative for the year 1997, for example. Similarly, the percentage of undernourished referring to 1995-1997 is indicative for the year 1997. The under-five-mortality rate has been directly taken from 1997. The complete listing of time ranges or years the single indicators and the respective NIs refer to is found in Table 5.