Contents - Next

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

Editorial policy

The Food and Nutrition Bulletin is intended to make available policy analyses, state-of-the-art summaries, and original scientific articles relating to multidisciplinary efforts to alleviate the problems of hunger and malnutrition in the developing world. It is not intended for the publication of scientific articles of principal interest only to individuals in a single discipline or within a single country or region. Notices of relevant books and other publications will be published if they are received for review. The Bulletin is also a vehicle for notices of forthcoming international meetings that satisfy the above criteria and for summaries of such meetings.

The Food and Nutrition Bulletin also serves as the principal outlet for the publication of reports of working groups and other activities of the UN ACC Sub-committee on Nutrition (SCN) and its Advisory Group on Nutrition. The SCN itself is a focal point for co-ordinating activities of FAO. WHO, UNICEF, the UNU, Unesco, the World Bank, the World Food Programme, the World Food Council, the United Nations Environment Programme, and other bodies of the United Nations system which have an interest in fond and nutrition.

Unsolicited manuscripts of articles of the type published in this and previous issues may be sent to the editor at the Cambridge office address given above. They must be typed, double-spaced, with complete references and must include original copy for any' figures used (see the "Note for contributors" in the back of this issue). All articles submitted will be reviewed promptly and the author will be notified of the editorial decision. Any disciplinary or conceptual approach relevant to problems of world hunger and malnutrition is welcome, and controversy over some of the articles is anticipated. Letters to the editor are encouraged and will be printed if judged to have an adequate basis and to he of sufficient general interest.

It is expressly understood that articles published in the Bulletin do not necessarily represent the views of the United Nations University or of any United Nations organization. The views expressed and the accuracy of the information on which they are based are the responsibility of the authors. Some articles in the Bulletin are reports of various international committees and working groups and do represent the consensus of the individuals involved; whether or not they also represent the opinions or policies of the sponsoring organizations is expressly stated.

Human nutrition

The assessment of caloric adequacy

Susan Randolph, Richard D. Ely, Lindsay H. Allen, Adolfo Chávez, and Gretel H. Pelto



The assessment of caloric adequacy on the basis of caloric shortfalls from nominal needs misclassifies individuals (undernourished as well-nourished, and vice versa) unless important inter-individual variations in energy needs are controlled for. This paper explores the extent of misclassification resulting from inter-individual differences in activity choice.



Although many factors influence health and nutrition status, nutrient intake has tended to occupy centre stage as the principal one. It is common practice to assess nutrition status on the basis of whether caloric intake meets or falls short of some nominal energy requirement. In this assessment, controlling for differences in energy needs between individuals in a meaningful way is essential. Failure to do so results in the misclassification of individuals and impairs our understanding of the extent and causes of malnutrition and who is most at risk. Despite agreement on this issue in principle, considerable controversy surrounds the question of which of the many possible factors influencing need can safely be ignored and which must be addressed.

Several articles have suggested that controlling for inter-individual differences in activity levels may be a matter of some importance. After reviewing several time-allocation studies from Indonesia, New Guinea, Côte d'Ivoire, and India, Edmundson and Sukhatme [1] concluded that the poor spend more time on economically productive work and are more likely to be engaged in heavy physical labour than are the wealthy. Strauss [2] also pointed out that higher earnings are likely to be associated with increased leisure consumption. Both factors imply that the poor are likely to be misclassified as well-nourished and the wealthy as poorly nourished. Failure to control for activity may bias results on nutritional gains during the course of national development, given changing occupational structures and employment opportunities [3]. However, the direction of the bias is not obvious.

There seems little question but that estimates of the extent of undernutrition will vary depending on whether inter-individual variations in activity level are accounted for. From the policy perspective, however, the critical question is whether ignoring these differences leads one significantly astray. What proportion of the population is likely to be misclassified if they are ignored? Is there a real danger of improperly identifying target groups? We addressed these questions in the context of four subsistence farming communities in the valley of Solis, about 75 miles northwest of Mexico City.


Methods and data

This study estimated energy adequacy on the basis of the ratio of observed intake to nominal need. If energy intake falls short of (exceeds) this level, the ratio will be less than (greater than) 1.0. Nominal need is assessed in two ways: on the basis of published WHO norms for activity levels in subsistence farming communities, and on the basis of individual-level activity data. To isolate the influence of inter-individual differences in activity mix, any variation between individuals in basal metabolic weight, observed weight change, and energy need for lactation is controlled for. The resultant energy adequacy measures are as follows:

Here Ii is the energy intake of individual i measured in kilocalories per 24 hours; R is the basal metabolic rate (BMR); W is the kilocalorie equivalent of daily weight change (8.27 kcal per gram), including the observed weight changes accompanying childbearing; and L is the kilocalorie expenditure for lactation. Nominal energy adequacy (NEA) uses a BMR multiplier, Si, of 1.78 for men and 1.76 for women; these are the WHO norms for activity levels in subsistence farming communities [4]. Corrected energy adequacy (CEA) uses (1/1,440) SjAijNj in place of Si in the estimate of energy needs. Here the variable Aij is the time (in minutes) individual i expends in each of the activities j over the 24 hours; Nj is the WHO multiplier of BMR for activity j [4]; and 1,440 is the number of minutes in 24 hours.

Basal metabolic rates were estimated on the basis of age, sex, weight. and height using WHO norms [4]. No further adjustment was made for pregnancy, in line with the findings of a study that concluded that pregnancy does not increase BMR beyond that accounted for by increasing weight during pregnancy [5] Data from a study on the average milk production of mothers by infants' weight [6] were combined with data on the average monthly weight of nursing children in the study population to estimate the average millilitres of milk produced during each month of lactation. The estimated requirement was converted to kilocalories at the rate of 0.7 kcal per millilitre using an 80% conversion rate.

An NEA value of 1.0 means that intake was sufficient to enable the individual to perform the activities typical in subsistence communities at nominal intensity levels. Individuals devoting less time to energy-intensive productive activities and more time to leisure activities than is typical in subsistence communities will have NEA values of less than 1.0, as will those in less strenuous occupations such as the village shopkeeper and schoolteacher. Such individuals are misclassified as malnourished on the basis of NEA. Similarly, persons engaging in more hours of physically strenuous activity than is typical in subsistence communities may have NEA values greater than 1.0 even if their energy intake is inadequate to enable them to engage in these activities at nominal intensity levels. Thus, they may be misclassified as well-nourished. The CEA measure avoids such misclassification by ensuring that the need assessment reflects the observed activity mix. A CEA value less than 1.0 implies that energy intake was inadequate to perform the activities undertaken if performed at nominal intensity levels.

The study's data were drawn from the Mexico-CRSP data base, a full description of which is available [7]. Marginal malnutrition was the basis of community selection and is evident in anthropometry; children tend to be stunted but not wasted. Households with children in any of three target groups and both parents present were invited to participate in the study; nearly 98% did so. Activity data were collected on the parents of the target children for two consecutive days out of each month. Interviewers asked the adults to recall how they spent their time during the previous day to the nearest minute, on two consecutive different days of the week each month. Individual intake data were collected for two consecutive 24-hour periods each month using a combined record-recall-weighing method. Food composition data specific to Mexico [8] were used to calculate the caloric content of foods consumed. Adults' heights and weights were measured every three months; for pregnant and lactating women, weights were recorded monthly.

Data collected during the calendar year 1984 were used for this study, as the activity data were most complete for that period. In the analyses, each intake observation day for each individual was initially treated as a single observation. Data on activity, weight, and lactation were interpolated for all days during the study period and matched to each intake observation day. A quasi-hermetic spline technique was used to interpolate weight and a cubic spline function to interpolate the activity data. The median (non-pregnant) height value was used to estimate BMR. These procedures resulted in an average of 14 observations on each of the 222 women and 11 observations on each of the 179 men in the analysis data set. The mean values of each energy adequacy measure were used in the analyses that follow to control for intra-individual variation in intake and activity.



Table 1 shows the extent and pattern of misclassification both for the population as a whole and separately for men and women. Overall, 13% of adults in the valley of Solis were misclassified on the basis of NEA, with 10% erroneously classified as well-nourished. These people tended to have a more energy-intensive activity mix than is typical of subsistence communities. This was particularly true for men, 19% of whom were misclassified as well-nourished on the basis of NEA, compared to 3% of women. In fact, NEA is more likely to misclassify women as undernourished than well-nourished. The gender disparity in misclassification is serious enough to result in the improper identification of target groups. Women were identified as facing a greater risk of malnutrition on the basis of NEA, whereas, once activity levels were controlled for, men were found to he at greater risk. If one chooses a lower cutoff value to differentiate between the well- and poorly nourished, the proportion of the population misclassified on the basis of NEA increases. A cutoff value of 0.9 resulted in the misclassification of 16"/o of the population (23"/o of the men and 11% of the women). A cutoff value of 0.8 resulted in the misclassification of 18% of the population 28% of the men and 10% of the women). As in the case of choosing a cutoff value of 1.0, when cutoff values of 0.9 or 0.8 were employed, the gender facing the greatest risk of undernutrition was misidentified on the basis of NEA.

TABLE 1. Extent and pattern of misclassification for the total population, men, and women percentages)

Nominal energy adequacy Corrected energy adequacy
Undernourished (CEA < 1.0) Well nourished (CEA >= 1.0) Total
Undernourished (NEA < 1.0) 77.6 2.7 80.3
Men 73.7 0.6 74.3
Women 80.6 4.5 85.1
Well-nourished (NEA >= 1.0) 10.0 9.7 19.7
Men 19.0 6.7 25.7
Women 2.7 12.2 14.9
Total (N= 401) 87.5 12.5 100
Men (N= 179) 92.7 7.3 100
Women (N=222) 83 3 16.7 100

Table 2 provides an alternative way of looking at the effect of inter-individual variation in activity mix on the assessment Of nutritional status. This table compares individuals' decile ranks as assessed by NEA with those for CEA. Positive entries imply that, on average, individuals classified in a given NEA decile rank improved their rank when reclassified on the basis of CEA. Negative entries imply that, on average, individuals lost rank - that is, were seen to be less well nourished - when reclassified on the basis Of CEA.

TABLE 2. Summary measures of decile rank change after controlling for activity, by sex

NEA decile(a) Percentage unchanged Mean absolute change Mean algebraic change
Men Women Men Women Men Women
1 85.7 61.6 0.24 0.39 +0.24 +0.39
2 42.9 46.2 0.79 0.81 +0.07 +0.81
3 8.3 39.3 1.25 0.96 -0.75 +0.89
4 7.1 15.4 1.64 1.54 -0.21 + 1.31
5 10.5 28.6 1.79 1.18 -0.95 + 1.00
6 15.0 20.0 1.75 1.35 -1.25 +0.45
7 17.6 17.4 1.47 1.13 -0.88 +0.96
8 6.3 12.5 1.81 1.25 -1.69 +0.33
9 26.1 41.2 1.39 0.76 -1.22 +0.06
10 39.1 87.5 1.17 0.13 -1.17 -0.13
Overall(b) 27.9 34.7 1.32 1.00 -0.82 +0.67
      (1.20) (095) (1.58) (1.20)

a.The first dccilc is thc most undernourished
b. Figures in parcnthcscs represent SD

On average, only 28% of men and 35% of women maintained their decile ranking after accounting for inter-individual differences in activity mix. Misclassification was especially pronounced for individuals who did not fall in the extreme NEA decile positions. Controlling for activity mix was of somewhat more importance for men than for women. Men moved an average of 13 percentiles and women an average of 10 percentiles after accounting for activity. If needs are not adjusted for intra-individual variation in activity, the nutritional well-being of men tends to he overstated and that of women understated; on average, the decile rank of men decreased by 0.8 and that of women increased by 0.7. These findings reinforce those presented in table 1 concerning the misidentification of target groups when interindividual variations in activity are ignored.



Failure to control for differences in energy needs arising from differences in the activity mix of individuals resulted in a substantial underestimation of the number of individuals with energy shortfalls in the Solis valley. The study population was relatively homogeneous; the extent of misclassification is likely to be even greater for more heterogeneous populations. The energy intensity of men's activities exceeded that of women's by a considerable margin. and this difference could lead to the improper identification of target groups if ignored. In other settings, the energy intensity of women's activities may exceed men's. While it may not be practical in most circumstances to control for differences in energy needs between individuals on the basis of activity-level data, the use of community-specific, or at least region-specific, norms in the assessment of energy adequacy would appear essential.



  1. Edmundson WC, Sukhatme PV. Food and work: poverty and hunger? Econ Dev Cultural Change 1990;38:263-80.
  2. Strauss J. The impact of improved nutrition in labor productivity and human resource development: an economic perspective. Economic Growth Center Discussion Paper no. 494. New Haven, Conn, USA: Yale University Press, 1985.
  3. Ravallion M. Income effects on undernutrition. Econ Dev Cultural Change 1990;38:489-516.
  4. World Health Organization. Energy and protein requirements: report of a joint FAO/WHO/UNU expert consultation. Geneva: WHO, 1985.
  5. Durnin JVGA. The energy requirements of pregnancy: an integration of the longitudinal data from a five-country study. Lancet 1987;2:1131-33.
  6. Chavez A, Martinez C. Growing up in a developing community. Mexico City: Instituto Nacional de la Nutrición, 1982
  7. Allen LH, Pelto GH, Chavez A. Marginal malnutrition and function in Mexico. Storrs, Conn, USA: University of Connecticut Press, 1987.
  8. Hernandez M, Chavez A, Bourges H. Valor nutritivo de los alimentos Mexicanos. Mexico City: Instituto National de la Nutrición, 1980.

Public health nutrition

Is positive deviance in growth simply the converse of negative deviance?

Meera Shekar, Jean-Pierre Habicht, and Michael C. Latham



The term "positive deviance" has been widely used to describe children who do not show evidence of protein-energy malnutrition when many others living in a similar unfavourable environment are malnourished. Implicit in this concept is that the determinants of positive deviance are something more than the converse of the determinants of poor growth. We modified and operationalized this concept using data on child growth from rural southern India. We divided children on the basis of anthropometry into positive deviants and what we called negative deviants and median growers. Our analysis suggests that the mechanisms producing positive and negative deviance are not always opposites or mirror images of each other. This finding has important implications for targeting and intervention strategies.


Editor's note

The concept of positive deviance in the capacity of underprivileged mothers to cope with health problems of their families has received a great deal of attention, including the book Positive Deviance in Child Nutrition [1], published as a supplement to this journal. There has been an assumption in most of this work that positive deviance is the opposite of negative deviance and that the behaviour responsible for negative deviance needs simply to be reversed to achieve the positive results. This article presents evidence that the mechanisms responsible for negative and positive deviance are not necessarily the same, and that factors associated with negative deviance may not be inversely correlated with positive effects. This non-uniformity of effect should be looked for whenever the concept of positive deviance is used to guide intervention strategies. The Bulletin considers this article to be a significant conceptually new advance and will welcome other studies exploring this approach.



Child growth and associated factors have been studied in much detail in numerous environmental and cultural settings. The usual design has looked at the whole spectrum of children, from those who are poorly nourished to those who are well nourished, as a single continuum. This approach also has been the basis for the design and evaluation of interventions.

The term "positive deviance" [ 1-3] in nutrition was introduced to reflect "adequate child growth in adverse environmental settings as an adaptive response to limited food availability" [1]. All previous attempts at studying positive deviance concentrated on contrasting the best-nourished with the worst-nourished, or contrasting the best-nourished with those "not malnourished." In principle, the results were no different from those achieved with the usual design. We have used data from Tamil Nadu, India, to modify and operationalize these concepts. The results have programmatic and strategic implications.


A framework for positive and negative deviance

The literature to date has addressed itself mainly to the study of positive deviance. We have added the terms "negative deviance" and "median growth," and defined them in the context of observed growth patterns in poor environments, thus obviating the need for reference to "adequate" or "inadequate" growth. Growth patterns of children (6-36 months of age) were graphed as a weight-for-age plot. Within this distribution, the uppermost end of the spectrum was identified as consisting of positive deviants and the lowermost end of negative deviants. Children growing at or around the median for the sample population were median growers. Thus' all definitions were relative to each other, and references to external growth standards were not necessary.

This framework does not imply that positive deviants are well nourished (a fact that may not be true for many poor environments, including that described here) but simply that they grow bigger and faster than others living in a similarly deprived environment. It is important to stress this point to avoid confusing this concept of positive deviance with the "small but healthy" concept [4; 5].

Another point that is implicit in this definition is that identification of a child as belonging to any of the three categories might better be based on longitudinal growth patterns than on a one-time cross-sectional measurement. Children identified as positive deviants on the basis of a single anthropometric measurement may not be correctly so categorized because of the dynamic nature of child growth, possible errors in anthropometry, and consequent chances of misclassification. Therefore, we used consistency in longitudinal growth patterns to classify the children. One factor that could confound the proposed definition and identification of deviance is inherent genetic variability within populations [6]. However, in populations living in poor environmental conditions where growth does not approach the expression of full genetic potential, differences in growth are mainly attributable to environmental influences rather than to genetic potential [7-9].

We propose that negative and positive deviance be viewed as two different conditions rather than as two ends of a continuum running from top to bottom wherein one is the converse of the other. Accordingly, the mechanisms operating to produce the two kinds of deviance are hypothesized to be different. A factor that may be associated with negative deviance need not necessarily be inversely associated with positive deviance. Similarly, some factors may be associated with positive but not with negative deviance.



A nested case-control study design was used [10]. Data were collected on 3,122 children from 42 villages selected randomly (out of a total of 68 villages) in Kottampatti Block, Madurai District, in the state of Tamil Nadu in southern India, within the project areas of the World Bank-aided Tamil Nadu Integrated Nutrition Project (TINP). Each village had one TINP centre catering to a population of about 1,500. Longitudinal growth data (weight for age) for 12 consecutive months (together with some socio-demographic data) were collected from existing TINP records for all children 6-36 months old enrolled in the 42 centres. A detailed description is available elsewhere [11].

Criteria for selecting positive and negative deviants and median growers from the total sample of 3.122 children were based on their rank according to growth velocity and attained growth over a 12-month reference period (April 1986-March 1987). The children who consistently tracked at the extreme top end of the growth spectrum (i.e. those who were best off in terms of both growth velocity and attained weight) were identified as positive deviants. Those who consistently tracked at the bottom of the spectrum and those tracking in the middle of the distribution were identified as negative deviants and median growers respectively.

The most extreme 100 children of each deviant group and 120 children of the median group who were available in their homes were included for a follow-up study. Supplementary data for all the 320 children were collected through follow-up interviews with their mothers. Information was collected on socioeconomic and socio-demographic factors; infant and young child feeding and dietary practices; health and hygiene; maternal knowledge, attitudes, and practice; maternal and paternal networking; and TINP programme participation variables.

Data were analysed using the SYSTAT statistical package. Positive (P) and negative (N) deviants were contrasted with median (M) growers (P versus M, and M versus N) to compare the size and statistical significance (p<.05) of these contrasts by t-tests or chisquare tests [12]. The probability that the differences between these two contrasts were the same was tested by a t-test of the differences between the sizes of the two contrasts (P versus M, and M versus N). The criterion for identifying the difference (P - M) (M - N) was a t value equivalent to a p< .05 to take into account the loss in power in assessing interactions, where p is the probability that the determinants of better growth are the converse of the determinants of poor growth.



Twenty-three of 52 variables examined differentiated the positive deviants from the median growers, and 12 differentiated the negative deviants from the median growers.

For two of the variables that differentiated the median growers from both positive and negative deviants, the magnitudes of the effect on positive deviance (P - M) were similar to those on negative deviance (M - N). Thus they lie on the same linear continuum, such that they seem to be the converse of each other. For example, as the number of visits to the doctor during pregnancy increased (from 2.35 to 3.77 to 4.93), the chances of moving from negative to median to positive deviant status increased proportionately (see FIG. 1. Number of visits to the doctor during pregnancy of mothers of negative deviants, median growers, and positive deviants).

For 22 of the variables the effect was not uniform as one moved across the growth spectrum. For some, the magnitude of the effect as one moved from negative to median to positive deviance was similar to that seen in drug response, where at higher levels a greater dose is required to obtain a same effect. An example is family wealth (see FIG. 2. Quantity of house land owned by the families of negative deviants, median growers and positive deviants), for which a small increment was associated with median growth and a much larger added increment with positive deviance. Appropriate transformation of these variables (e.g. logarithmic) makes the relationship linear. an indication that the response may be related to the proportion of the increment in dose. Other variables clearly were not dose-related, even though they could be scaled to be linear. For example, maternal wealth (proxied by whether the mother wears gold jewellery or not) (see FIG. 3. Percentages of mothers of negative deviants, median growers, and positive deviants who wear gold jewellery. and percentage of children of each category who are male (variables unlikely to be dose-related)) showed a much greater increment associated with negative than with positive deviance. The sex of the child was another linear variable that was not likely to be dose-related.

For other variables (see FIG. 4. Number of sons and of daughters desired by mothers of negative deviants, median growers, and positive deviants) we observed a peaking or dipping at the median level, wherein mean values were highest (peaking) or lowest (dipping) for median growers. Positive and negative deviants had more similar values, as in the case of maternal desire for an ideal number of sons (dipping) and daughters (peaking). This peaking or dipping also cannot be explained by a non-linear dose response. because no mathematical transformation can make such a response linear.

All the variables presented in figures 2 through 4 differentiated between the median growers and the positive deviants in a fashion that was different from the differentiation between the median growers and the negative deviants. For these variables, the negative deviants were not simply the converse of the positive deviants. Thus, moving from the exemplary variables to groups of variables that characterized deviance, some of the important characteristics of positive deviance were greater family wealth (proxied by quantity of land on which the house is situated, number of rooms in the house) (p < .01), greater consumption of prestige foods such as coffee (p<.01), lower maternal age at marriage (p<.05), and fewer hours spent by the mother at work (p < .05). The converse was not necessarily true; that is, negative deviants were not very much worse off than median growers in terms of family wealth, older maternal age at marriage, and maternal work hours.

Female sex was associated with negative deviance (p < .05), but male sex was not significantly associated with positive deviance. Similarly, lower maternal wealth (proxied by the mother's jewellery) was associated with negative deviance (p<.01), but higher maternal wealth did not predispose toward positive deviance.



The results support the primary hypotheses of the study, that positive and negative deviance are not necessarily the converse of each other. and that different factors characterize them.

Implications for programme targeting

For targeting, the mechanism of an effect is less important than its magnitude. Thus, it is important to know whether or not a variable differentiating the positive and negative deviants operates uniformly as one moves from positive to median to negative deviance. For instance, family wealth in this very poor population differentiated much more poorly between the negative deviants and median growers than between the positive deviants and median growers. Family wealth is therefore not useful in targeting negative deviants. Instead, the gender of the index child (female) may be a better criterion for the most needy negative deviants.

With regard to the dietary variables, prestige foods like biscuits, coffee, and milk (which is used in such small quantities that its nutritional contribution per se is unlikely to be of any consequence) were more commonly used in the household among positive deviants than among median growers. Parallel differences were not obvious between the negative deviants and median growers. The use of these foods would therefore be a poor targeting characteristic for the negative deviants. On the other hand. low consumption of pulses (the major protein source in the local diet) was the most important dietary factor differentiating the negative deviants from the median growers and would be a useful criterion for the most needy negative deviants.

Implications for programme design and evaluation

Non-uniformity of effect is important not just for targeting but also for identifying possibly different mechanisms of impact as one moves from positive to median to negative deviance. For instance, girls are more likely than boys to be negative deviants rather than median growers, but gender did not differentiate between positive deviants and median growers. Furthermore, of all the mothers those of median growers had the greatest desire for daughters and the least preference for sons, while the mothers of both positive and negative deviants expressed an overt preference for male children and less desire for females. It would appear that the mothers of median growers were therefore least likely to discriminate against a female child in feeding or child care and that this helped to protect their female children from negative deviance. Mothers of positive deviants (like the mothers of negative deviants) preferred more male children and fewer female children, but the disadvantages of this gender preference were outweighed by other factors such as greater family wealth, which promotes positive deviance.

When negative deviants form the only target group, the correct intervention strategy would not only seek to improve the nutritional status of all negative deviants but would also complement this effort with appropriate gender-specific interventions to reduce gender bias in child care. These additional gender-targeted interventions will, however. have no effect on improving the growth of the median growers to make them positive deviants. Therefore, if an intervention is targeted to be gender specific, it must also be evaluated with these facts taken into consideration in order to he able to assess the true impact, which would otherwise be diluted and hence not be picked up even if it exists.



Our analyses support the hypothesis that, for many variables, positive and negative deviance in growth exist as distinct conditions promoted by different mechanisms, and they are not the converse of each other. The data indicate that, as compared to median growers, positive deviant children come from wealthier families that consume more prestige foods and their mothers work fewer hours. These factors do not' however, predispose to negative deviance. which is characterized by female sex, lower maternal wealth, and lower consumption of protein-rich foods such as pulses. These findings have clear implications for programme targeting. design, and evaluation.



Funds for this research were provided by the Thrasher Research Fund and the Nestlé Nutrition Research Grant Fund. Edward Frongillo, Jr., contributed to the analytic plan for the study. We are grateful to the TINP Project Office, Government of Tamil Nadu, for permission to conduct the research.



  1. Zeitlin MF. Ghassemi H, Mansour M. Positive deviance in child nutrition. Tokyo, Japan: United Nations University Press, 1990.
  2. Chavez MM, Arroyo P, Gil S et al. The epidemiology of good nutrition in a population with a high prevalence of malnutrition. Ecol Food Nutr 1974;3:223-3(1.
  3. Wray JD. Editorial. J Trop Paediatr Environ Child Health 1972;18(3):279.
  4. Martorell R. Body size, adaptation and function. Humn Org 1989;48:15-20.
  5. Beaton G. Small but healthy? are we asking the right question? Humn Org 1989;48:30-39.
  6. Lowentin RC. The analysis of variance and the analysis of causes. Am J Humn Genet 1974;26:400- 11.
  7. Habicht J-P, Yarbrough C, Malina M, Klein RE. Height and weight standards for pre-school children: how relevant are ethnic differences in growth potential? Lancet 1974;1:611-15.
  8. Stephenson L, Latham MC, Jansen A. A comparison of growth standards: similarities between NCHS, Harvard. Denver and privileged African children and differences with Kenyan rural children. Cornell International Nutrition Monograph Series. no. 12. Ithaca, NY, USA: Cornell University, 1983.
  9. Editorial. Lancet 1984;1:142-43.
  10. Kleinbaum DG, Kupper LL, Morgenstern H. Epidemiologic research: principles and quantitative methods. 10th ed. London: Lifetime Learning Publications, 1982:71-73.
  11. Shekar M. Positive and negative deviance in child growth: a study in the context of the Tamil Nadu Integrated Nutrition Project. PhD dissertation, Cornell University, Ithaca, NY, USA, 1990.
  12. Snedecor GW, Cochran WG. Statistical methods. 7th ed. Ames, Iowa, USA: Iowa State University Press, 1982:290-91.

Contents - Next