Contents - Next

This is the old United Nations University website. Visit the new site at

Nutrition and health

Could improvements in child survival mask improvements in anthropometric indicators in nutrition programme evaluations?
Socio-economic determinants of child nutritional status: boys versus girls
Mortality levels and patterns in the oral therapy extension programme areas of the bangladesh rural advancement committee
Nutritional aspects of obesity and diabetes and their relation to cardiovascular


Could improvements in child survival mask improvements in anthropometric indicators in nutrition programme evaluations?

John G. Haaga

Cornell Nutritional Surveillance Program, Division of Nutritional Sciences, Cornell University, Ithaca, N.Y., USA

It has sometimes been suggested that evaluations of primary health care (PHC) and supplementary feeding programmes have failed to show improvements in children's heights or weights because the programmes are promoting survival of severely malnourished children [1] . It is argued that these children who, in the absence of the programme, would have died have worse nutritional status than those who would have survived anyway, and thus they bring the average down when the nutritional status of programme beneficiaries is compared with that of non-beneficiaries, or when the post-programme nutritional status is compared to the pre-programme status. This paper shows that this argument is applicable only in extremely unusual circumstances.

A thought experiment will show why. Suppose that a PHC/ feeding programme is begun in an extremely poor country where the life expectancy at birth is very low, say 42 years (the estimate for Chad in 1982). There are 5,000 three-month-old children who are measured when they begin the programme and who are then measured again when they leave the programme at age five years. Thus, while they are in the programme, they pass through the ages when about half of infant mortality and nearly all of childhood mortality occurs. Assuming that the age pattern of mortality recorded in the United Nations Model Life Tables for Developing Countries, General Pattern [2] applies, and that there is no change in mortality rates, 886 of the children would be expected to die before age five. Suppose the programme were dramatically successful, cutting mortality by three-quarters, so that an additional 664 children survived to be measured at age five. If the 4,114 children in the programme who were going to survive anyway had also improved in nutritional status (say from an average weight-for-age of 70 per cent of the standard to an average of 75 per cent), what would the average weight-for-age among the 664 saved children have to be in order to disguise this improvement, keeping the whole population average down to 70 per cent? The answer is 39 per cent (table 1). The children whose lives were saved by the programme would have to weigh a mere 39 per cent of the standard, on the average, at age five, in order to mask even this modest improvement in weight-for-age due to the programme.

TABLE 1. Example of calculationa

Survival to age one year = 0.85222 (average of male and female life tables for life expectancy = 42 years)
Survival to completion of half of infants who would have died without programme deaths = 0.92611
Survival to age five years = 0.761895 Deaths during programme participation (half of infant + child deaths normally expected for 5,000 children) = 5.000 * (0.92611 - 0.761895 /0.92611) = 886
75 per cent mortality reduction = 664 lives saved (instead of 4,114 at age five, 4,778 still alive)
If 4,114 improved to 75 per cent weight-for-age, and the population average is still 70 per cent weight-for-age, then the average weight-for-age for the 664 saved must be:
(4,114 x 75) + (644 X) = (4,778 x 70) X = 43.5
a. Weight-for-age is expressed as a percentage of the median, not as a percentile of the reference population.

But 39 per cent weight-for-age is not observed in real, living children. Analysts of survey data see such values crop up from time to time, of course, and recognise thereby an error in age or weight estimation or recording.

The assumptions used in this "thought experiment" are extreme: most programmes operate in countries with higher life expectancy (lower child mortality) to begin with, and most, unfortunately, effect far less dramatic reductions in mortality. The first row in table 2 shows what the required average weight-for-age among the children whose deaths are averted by a programme would have to be if an improvement of from 70 to 75 per cent of standard weight-for-age among the other survivors were to be masked. If the mortality reduction in the first example were "only" by half, the children saved would have to average 24 per cent weight-for-age to hide the weight gain in the others. If mortality fell by a quarter, the children saved would have to have negative weights to hide the effect. If life expectancy is initially 53 years, corresponding to the situation in the Philippines in 1982, there are no circumstances in which the children whose deaths are averted can mask an improvement in the others from 70 to 75 per cent of standard.

TABLE 2. Required average weight-for-age (WA) in children whose deaths are averted, if weight gain in other survivors is to be cancelled out

Initial life expectancy (years) Percentage reduction of child mortality (percentage) Increase in WA among programme children
70 to 75 per cent of standard 90 to 92 per cent of standard
42 25 - 229 53
50 24 71
75 39 78
53 25 - 100 22
50 - 15 56
75 13 67
62 25 - 243 - 35
50 - 87 27
75 - 35 48

The required averages shown in table 2 are impossible, of course. If the nutritional improvement starts from a higher initial average, and the effect of a smaller change is investigated—say, from 90 to 92 per cent of standard—then the averages for the saved children need not be so extreme. The second row of table 1 shows the average weight for-age among the saved children required to swamp this nutritional improvement among the other survivors. Again, the values for a country with the mortality rates of the Philippines are all impossibly low (ranging from —35 to 48). For a 75 per cent reduction in mortality, when life expectancy is initially as low in Chad, it is just conceivable that the postulated "survivor effect" would disguise the improved weight. The saved children would have to average 78 per cent of standard. But this is a full 14 points below the average for the other programme children, assumed to be 92 per cent. In the Narangwal study [1], the initial difference in weight-for-age between children who subsequently died and those who survived was considerably smaller, 5 per cent of standard (72 per cent versus 77 per cent, calculated from grouped data).

At this point, one would start to question the policy importance of the anthropometric findings. If a programme could slash child mortality rates by three-quarters in countries with the highest mortality rates in the world, surely the information that it also raises the average weight for-age from 90 per cent of standard to 92 per cent would be of little additional interest. it can be seen that the survivor effect, while logically a possible explanation for a failure to find expected improvements in anthropometric indicators, is not a quantitatively significant effect under realistic conditions. If before-and-after comparisons or programme versus non-programme comparisons show no differences in children's weights, it is for some other reason, or because no differences exist.


This research was supported by Co-operative Agreement DSAN/CA-0240 between the Office of Nutrition, US Agency for International Development, and the Division of Nutritional Sciences, Cornell University.


1. A. A. Kielmann and C. McCord, "Weight-for-age as an Index of Risk of Death in Children," Lancet, 2: 1247-1250 (1978).

2. UN Secretariat, Department of International Economic and Social Affairs, Model Life Tables for Developing Countries (UN, New York, 1982) (ST/ESA/ser. A/77).


Socio-economic determinants of child nutritional status: boys versus girls

Abbas Bhuiya, Bogdan Woityniak, Stan D'Souza, and Susan Zimicki
International Centre for Diarrhoeal Disease Research, Dhaka, Bangladesh


The inverse relationship between household socio-economic status (SES) and child mortality has been observed in many developing countries [1-5]. Child nutritional status was postulated to be one of the major pathways by which SES affects child mortality [6], and the positive relationship between SES and child nutritional status has also been documented in many studies [7-13] . But in most cases the relationship was examined in a univariate situation which neither allows one to assess the total importance of SES nor singles out any particular variable while controlling the effect of others. Moreover, while boys and girls were both included in the groups studied, the sex of the children was not considered in the analysis. Yet in societies where parents have a strong preference for male offspring, one suspects that the relationship between SES and child nutrition is different for boys and girls.

In this paper, multivariate linear models of the relationship between child nutritional status and household socio-economic characteristics have been developed separately for boys and girls using data from a rural area of Bangladesh.


The data for this study come from five villages of Matlab Upazila (administrative unit below a district) in Bangladesh. The International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR, B) has been operating a Demographic Surveillance System in this area since 1966 [14]. The Matlab field research area is in the low-lying deltaic plain of Bangladesh, situated 45 km south-east of Dhaka, the national capital. The area is relatively dry during the months of December to February, with very little rainfall. During the May to September monsoon the water rises and inundates most of the non-residential land.

The five study villages are socio-economically and culturally similar to other Matlab villages. The average household size is six persons, and 86 per cent of the total population in the study villages is Muslim. Farming is the most common occupation of the heads of households. The educational level is low, with 55 per cent of the household heads having had no formal schooling. Most of the houses are made of jute sticks and mud with thatched or tin roofs. Sanitation in the villages is very poor.

Two separate surveys were conducted in these villages in 1981. The first, seeking socio-economic information, covered all households during the months of February to April. Anthropometric measurements on all 1,722 available children aged 2-60 months were collected during the second survey, from June to September. However, the present study sample comprises 1,408 children aged 12-60 months. Weight for-age has been used as an indicator of nutrional status.

Household socio-economic information was collected by interviewing the household head or (in the case of his or her absence) another senior household member. The variables used in the present study are: religion, education of head of household and of the mother of the child, household size, ownership of articles (quilt, hurricane lamp, bicycle, watch, and radio), amount of cultivable land owned, and tax paid to the union council (lowest administrative unit) in the preceding year. Education of the household head and of the mother was measured in terms of years of secular schooling.

Anthropometric measurements were made by workers trained to use scales. At the end of the training, the extent of the variation among workers' measurements and between the workers' measurements and those of a supervisor was examined and found to be negligible. The body weight of the lightly clad children was measured to the nearest 50 g, using 25 kg Salter scales. The weight-for-age percentage was calculated using the median values of the US National Center for Health Statistics standard growth curves [15] .

Analysis was carried out in two stages: first, the difference in mean weight-for-age for various SES and age categories was examined separately for boys and girls in a univariate situation; second, multiple linear regression models of weight-for-age on those variables were developed. Independent variables were dummy-coded and the coding scheme was as follows:

Land tax: If land <= 2 acres and tax < 5 take (1 take = $0.03), then
If land <= 2 acres and tax > 5 take, then land-tax 1 = 1; else land-tax 1 = 0.
If land >2 acres and tax < 5 take, then land-tax 2 = 1; else land-tax 2 = 0.
Possession of articles: None = 0; at least one = 1.
Household size: < 6 = 0; 7+ = 1.
Religion: Hindu = 0; Muslim = 1.
Education of head: < 5 years = 0; 6+ = 1.
Education of mother: No schooling = 0; some = 1.

Land and tax were combined in a three-category variable since no household was found to possess more than two acres of land and pay tax less than five take. Accordingly in the regression analysis this variable has been represented by two dummy variables.


Mean weight for ages of boys and girls in the study were 69.5 and 68.43 respectively. Table 1 presents mean values for different socio-economic characteristics of the households and age of the children. Education of head of household and land tax showed a significant positive relationship with both the boys' end the girls' nutritional status, while mother's education and possession of articles showed a similar relationship only for boys. On the other hand, age was found to be related to boys' nutritional status in a non-linear fashion.

Multivariate regression analysis was carried out by taking weight-for-age as a continuous dependent variable and socio-economic characteristics as independent, including all first-order interaction terms. Age and age^2 were also included as continuous independent variables.

TABLE 1. Mean weight for age of boys and girls by household characteristics including education of head and mother, and age

  Weight for age
  Boys Girls
Characteristics Mean SD N Mean SD N
Land tax  
Land-tax 0 68.02a 8.48 345 67.26b 9.19 284
Land-tax 1 70.19 8.67 294 69.02 8.69 320
Land-tax 2 72.47 8.74 99 70.62 9.77 65
Ownership of articles  
None 66.41a 8.29 141 67.09 9.06 140
At least one 70.21 8.66 598 68.78 9.05 529
Household size:  
<6 68.94 8.89 348 68.49 9.06 296
7 + 68.98 8.53 389 68.38 9.07 373
Muslim 69.67 8.66 624 68.27 9.08 564
Hindu 68.51 8.96 113 69.31 9.01 105
Education of head (years of schooling)  
0 68.92c 8.36 383 67.90b 8.97 352
1-5 69.57 9.53 240 67.93 9.26 214
6+ 71.24 7.86 114 71.26 8.55 103
Education of mother (years of schooling)  
0 68.64a 8.61 529 68.01 9.11 480
1+ 71.64 8.62 208 69.49 8.89 189
Age in months  
12-23 68.01c 9.23 209 67.91 9.87 197
24-35 69.94 8.21 156 69.37 9.80 143
36-47 70.69 8.91 169 68.66 8.28 156
48-60 69.65 8.22 205 68.03 8.11 173

a. p<0.002;
b. p<0.01;
c. p<0.05.

Regression models for boys and girls were developed by forward step wise procedure. In the case of significant interaction terms, the main effects involved were forced in the model. Subsequently, if main effects and interaction terms did not improve the explanatory power of the model (measured by significant change in R^2) they were not included in the final equation. Results of the analyses are presented in table 2. The relationship between nutritional status and SES observed in the univariate analysis has been found changed to some extent in the multivariate context.

Possession of more than two acres of land and collateral tax amounting to more than five take, the mother's education, the household size, religion and age were the factors significantly related to the boys' nutritional status. However, their effects in most cases were not simply additive but depended mostly on the level of other factors. This fact is expressed by the significant interaction terms; results of these interactions are presented in table 3. The mother's education was found to have a more pronounced impact on the boys' nutritional status in smaller households than in larger ones. The significant coefficients associated with age and age^2 indicated that age and nutritional status are related as an inverse "U" shape with the peak at the age of 45 months.

TABLE 2. Ordinary least squares regression estimates of weight for age on household characteristics, education of mother and head,and age of the children.

  Regression coefficients
Independent variables Boys Girls
Age1 0.365a (b1) 0.241d (b1)
Age2 -0 004c (b2) -0.003d (b2)
Land-tax 1 - - 2.172 (b3)
Land-tax 2 1.978c(b3) 3.886b (b4)
Possession of articles (ART) 1.244 (b4) -
Household size (HS) - 3.028c(b5) - 1.247d (b5)
Religion (REL) 2.002c(b6) - 3,489b (b6)
Education of head - 2.982b (b7)
Education of mother (MED) 7.783a (b7)  
REL*land-tax 1 - 4.844c (b8)
REL*MED - 4 000d (b8) -
HS*MED -3.840b (b9) -
HS*ART 4 899b (b10) -
Constants 59,000a (b0) 66.569a (b0)
R^2 0.0878a 0.0489a
N 739 669

a. p<0.001;
b. p<0.01;
c. p<0.05;
d. p<0.10..

TABLE 3. Significant interaction effects of the variables on nutritional status of boys and girls based on the regression models

Boys   Boys
Hindu Muslim None 1 +
0 b0 b0 + b6 <=6 b0 b0 + b4
59.00 61.00 59.00 66.60
1 + b0+b7 b0+b6+b7+b8 7+ b0+b5 b0+b4+b5+b10
66.78 64.78 62.44 68.54
B vs D, p<0.001   B vs D, p<0.05
C vs D, p>0.05   C vs D, p>0.001


HS Boys Land-tax1 Girls
<6 0 1 +   Hindu Muslim
b0 b0+b7 0 b0 b0+b6
59.00 66.78 66.57 63.08
7+ b0+b5 b0+b5+b7+b9 1 b0+b3 b0+b3+b6+b8
62.44 59.91 66.50 67.85
B vs D, p<0.001   B vs D, p<0.001
C vs D, p>0.05   C vs D, p>0.05

The mother's educational effect was significant for both Hindus and Muslims, though it was more pronounced in Hindu households. In other words, religion was not so important in differentiating boys' nutritional status if the mother had had some schooling.

The last interaction term in the boys' model implies that the negative effect of household size was not observed if the family had achieved some socio-economic level, as determined by possession of articles such as a quilt, a hurricane lamp, a bicycle, a watch, and a radio.

Although the model is highly significant (p<0.001), it explained only 8.8 per cent of the total variation in the boys' nutritional status.

A model developed for girls ended up with fewer variables than the one for boys. Possession of more than two acres of land and higher tax payment, education of household head, household size, religion, and age were the factors significantly related to the girls' nutritional status. The unfavourable effect of being Muslim was not important if the household had two acres or less of land and paid a tax of 5 taka or more. The girls' age, like the boys', also showed a similar, but weaker, pattern of curvilinear relationship, with weight-for-age peaking at the age of 40 months.

The predicting power of the girls' model was even smaller than that of the boys; only 4.9 per cent of variation in the girls' nutritional status could be explained by the household socio-economic characteristics and age.


The difference in the regression models of socio-economic variables on weight-for-age for boys and girls indicates that sex is an important factor which modifies the relationship between household socio-economic characteristics and child nutritional status.

Child nutritional status is determined by a variety of complex factors of which adequate food intake and proper health care, during and after sickness, are two of the important behavioural ones. Both factors are in fact related to the amount of resource allocation which, in turn, is influenced by the attitude of the decision-makers towards the children. In our case, the better nutritional status of boys and girls in the highest land-tax group households is a reflection of the positive impact of resource availability. In the same way, the negative effect of household size on both boys' and girls' nutritional status may be due to a scarcity of resources, modified by the effects of ownership of articles. If the negative effect is observed only, this may reflect some discrimination against female children in this particular group of households, which is richer than the local norm and possibly more open to modern innovations. Although evidence of discrimination against female children in intrafamily food distribution and medical care has been documented for this study area of Bangladesh [16], the existence of such discrimination in this particular group of households is strange.

The mother's education, which can influence child health and mortality in many ways [2, 17], was found to have a positive impact only on boys' nutritional status. Although a surviving son ensures a better status for the mother in the household in rural Bangladesh, it is difficult to believe that mothers with more education are more biased towards their male offspring. It should be realized that in most cases mothers are not the decision-makers; however, their concern for their children's welfare may get more support from the male decision-makers if the child is a boy. In this way boys may benefit more from having more educated mothers. Nevertheless, the situation is different when the household head has had more than five years of schooling, as indicated by the positive impact of the household head's education on the girls' nutritional status.

The present study suggests that the sex of the children should be considered as an important factor in studying the socio-economic determinants of nutrition land perhaps even mortality) in a society like rural Bangladesh. The relationship between household SES, including mother's education, and child nutrition and mortality may vary between boys and girls, as is observed in many studies.

Lastly it should be mentioned that, although the models are statistically significant, a large amount of variation in dependent variables was unexplained. This means that although household SES showed some relationship with the weight-for-age of the children in the present study, its overall contribution to the explanation of observed differences in nutritional status was rather small. The possible explanation may be that in a poor society like that of rural Bangladesh, stratification of households by the SES categories used in the analysis could not really make the difference between strata prominent enough to observe a wide variation in child nutritional status. Findings of the present study also suggest that factors other than SES may be more important in explaining differences in the nutritional status of children aged 12-60 months.


This study was carried out at the International Centre for Diarrhoeal Disease Research, Bangladesh (ICDDR,B).

ICDDR,B is supported by countries and agencies which share its concern about the impact of diarrhoeal diseases on the developing world. Current donors to the centre include: the Arab Gulf Fund, Australia, Bangladesh, Canada (Canadian International Development Agency and International Development Research Centre), Japan, Saudi Arabia, Sweden, Switzerland, the United Kingdom, and the United States (United States Agency for International Development).


1. A. Antonovsky and J. Bernsten, "Social Class and Infant Mortality, "Social Science and Medicine, 11: 454-470 (1977).

2. J. C. Caldwell, "Education as a Factor in Mortality Decline: An Examination of Nigerian Data," Population Studies, 33: 395-413 (1978)

3. J. Da Vanzo, "A Household Survey of Child Mortality Determinants in Malaysia," Population and Development Review, 10 (Supp.): 307-322 (1983).

4. S. D'Souza and A. Bhuiya, "Socio-economic Mortality Differentials in a Rural Area of Bangladesh" Population and Development Review, 8: 540-560 (1982).

5. A. A. Farah and S. H. Preston, "Child Mortality Differentials in Sudan," Population and Development Review, 8: 365-383 (1982).

6. W. H. Mosley and L. C. Chen, "An Analytical Framework for the Study of Child Survival in Developing Countries," Population and Development Review, 10 (Su pp. ): 25-45 (1984) .

7. M F. Smith, S. K Paulsen, W. Fouger, and S. J. Ritchey, "Socio-economic, Education and Health Factors Influencing Growth of Rural Haitian Children, "Ecology of Food and Nutrition, 13: 99-108 (1983).

8. R Bairagi, "Dynamics of Child Nutrition in Rural Bangladesh," Ecology of Food and Nutrition, 13: 173- 178 (1982).

9. P Desai et al., "Socio-economic and Cultural Influences on Child Growth in Rural Jamaica," Journal of Biosocial Sciences, 2: 133-143 (1970)

10. Institute of Nutrition and Food Science, Nutrition Survey of Rural Bangladesh 1975-76 (University of Dhaka, Dhaka, 1977).

11. D. Morley et al., "Factors Influencing the Growth and Nutritional Status of Infants and Young Children in a Nigerian Village," Transactions of the Royal Society of Tropical Medicine and Hygiene, 62: 164- 195 (1968) .

12. R. B. Tripp et al., "Farmers and Traders. Some Economic Determinants of Nutritional Status in Northern Ghana," Food and Nutrition, 8: 3-11 (FAO, Rome, 1982).

13. A Bhuiya, S. Zimicki, and S. D'Souza, "Socio-economic Differentials in Child Nutrition and Morbidity in a Rural Area of Bangladesh," Journal of Tropical Pediatrics (in press).

14. Cholera Research Laboratory, Demographic Surveillance System-Matlab, 10 (Dhaka, 1978).

15. National Center for Health Statistics, NCHS Growth Curves for Children, Birth-18 Years, United States, DHEW Publication 78-1650 (Washington, D.C., 1977).

16. L. C. Chen, E. Huq, and S. D'Souza, "Sex Bias in the Family Allocation of Food and Health Care in Rural Bangladesh, Population and Development Review, 7: 55-70 (1981).

17. S. Lindenbaum, M. Chakrabory, and M. Elias, "The Influence of Maternal Education on Infant and Child Mortality," Special Publication, no. 23 (International Centre of Diarrhoeal Disease Research, Dhaka, 1985).


Contents - Next