|Early Supplementary Feeding and Cognition (Society for Research in Child Development, 1993, 123 pages)|
Theodore D. Wachs
Demographic data have indicated that undernutrition is a common occurrence for a large proportion of the world's children (Simeon & Grantham-McGregor, 1990). Evidence relating undernutrition to cognitive development has been available for at least half a century (Brozek, 1978). In spite of these facts, interest by behavioral researchers in the role of nutritional influences on development has been relatively limited, at least as indicated by available publications on this topic For example, a Psychlit scan between 1987 and 1992 revealed only 14 references on the topic of nutrition and cognitive ability, with only seven of these references involving children. In contrast, a Psychlit scan of studies relating genetics to cognitive ability yielded 67 studies, 42 of which involved children. It is not clear why behavioral researchers have tended to neglect the study of nutrition. Perhaps the fact that the bulk of studies on behavioral development have been done by researchers from developed countries (Schopflin & Muller-Brettel, 1990) may have something to do with this relative neglect of links between nutrition and behavioral development. It is to be hoped that the present Monograph will go a long way toward remedying this state of benign neglect.
Contributions of the Monograph
There are two major findings that emerge from this data set. First, there is evidence that early nutritional supplementation can have long-term developmental consequences for later cognitive performance. If developmental researchers wish to understand processes influencing variability in cognitive development in populations other than advantaged children in Western developed countries, the present pattern of results strongly argues for the development of models that include nutritional status as a critical parameter. 1
1 While we tend to think of undernutrition
as a phenomenon found primarily in less developed countries, it is important to
remember that pockets of undernourished populations are also found in Western
developed countries (Karp, in press). Hence, the present findings may also be
relevant for understanding group differences in cognition in developed countries
The second major finding of this study is that the relation of nutrition to concurrent and later cognitive performance does not fit a simple main-effects model. Rather, the relation of nutrition to cognitive performance is moderated both by the time period during which the nutritional supplementation was given and by the larger sociodemographic context (social status, school level) within which subjects resided. For example, children in the lowest sociodemographic groups benefited more from nutritional supplementation than did children from higher sociodemographic groups. Similarly, while there were still nutritional influences when supplementation was started after 2 years of age, the effect appears to be less powerful. What these results illustrate is the need to look at nutritional influences as part of a system of multiple determinants (Wachs, 1993).
The results outlined above also emphasize the need to tailor our data analysis strategies to recognize the operation of this multidetermined system. For example, ecological theorists such as Bronfenbrenner (1993) have contended that traditional analytic approaches for dealing with individual characteristics or contextual factors, namely, by covarying or partialing them out, are inappropriate. Covariance or partialing techniques are based on the assumption that the same developmental processes occur for different individuals or within different contexts (Bronfenbrenner, 1993). The present findings are a dramatic demonstration as to why it is important to look for different developmental processes across different contexts, rather than assuming that the same main-effect process is operating across contexts. Had Pollitt et al. not tested for moderating sociodemographic or grade-level influences, they might well have come to the conclusion that increased nutrition enhances development for their total sample; this conclusion, although parsimonious, would have been highly misleading. Given increasing evidence that most behavioral variability within normal ranges appears to be multidetermined, both at the molecular level (Plomin, 1990) and at a molar level (Wachs, 1993), the present results offer further confirmation as to why it is important to consider and not control for multiple influences on development.
The present project offers not only a set of findings that emphasize the need to bridge the gap between nutritional and developmental research but also a model of how a nutritional intervention field study should be carried out. This study is exemplary in terms of the care taken to identify potential confounders and, where possible, test for the effect of these potential confounders. Where it is impossible to test, Pollitt et al. are explicit in terms of how potential confounders might affect their results. There are multiple examples that illustrate why this study may be considered as a model nutritional intervention study. These include testing whether the supplement was truly a supplement and not just a substitution, looking for selective migration effects, considering whether attendance at the feeding station could act as a placebo, and testing for examiner effects. What we have here is not only a most provocative data set but also a set of guidelines for how future studies of this type should be designed and implemented.
One interesting finding is that nutritional interventions had a much more dramatic effect on psychometric measures of intelligence and achievement than on information-processing measures, such as reaction time. A similar pattern of differential influence is shown for the effects of SES and schooling. These differences could have important theoretical considerations. It has been argued that information-processing measures may be more universal (hard wired) and therefore potentially less responsive to extraneous factors such as nutrition, schooling, or sociodemographic differences (Kail, 1991). However, before assuming that this differential reactivity of psychometric versus information-processing measures reflects differential wiring, we must consider at least one alternative explanation. As Pollitt et al. show in their Table 13, the psychometric and achievement measures were more stable than the information-processing measures. The lower degree of intervention effects for the information-processing measures may not reflect differential reactivity as much as the fact that information-processing measures may be less stable. 2 While this study was not designed to test cognitive theory issues of this type, it is unfortunate that less stable information-processing measures did result in a loss of potentially interesting information. This may suggest the need to build in compensatory procedures such as aggregation (Rushton, Brainerd, & Pressley, 1983) to compensate for potential stability differences when researchers suspect the possibility of differential stability for measures from different domains.
2 As a preliminary test of this hypothesis,
I correlated stability coefficients for information-processing measures (Table
13) with the beta coefficients reflecting nutritional treatment effects on these
five measures (Table 19). The resulting correlation, although modest (r
=.26), is in the expected direction, suggesting stronger effect sizes for the
more stable measures.
A second question involves sex differences. While Pollitt et al. take great care to look for sex differences in the preschool period and in terms of village demographics, there is a surprising omission in the follow-up data. Specifically, there is no evidence presented on whether there are sex differences in either school attendance or school progress. This omission is particularly surprising given the fact that there are sex differences favoring males in the results (see Tables 18 and 19) as well as results indicating the relevance of school factors to subsequent cognitive performance. It could be argued that, since the sex x treatment interaction was nonsignificant, looking at sex differences in school functioning is, at best, only a side issue. However, as elegantly demonstrated by Wahlsten (1990), analyzing for interactions often results in lower statistical power than analyzing for main effects. Under these circumstances, if males have higher levels of school attendance and school progress, it would be useful to consider the potential role of sex differences in understanding processes underlying nutrition-cognition relations. This is particularly true if there were also sex differences in consumption of the supplement as opposed just to attendance at the feeding station.
In their discussion, Pollitt et al. emphasize the protective (buffering) effects of supplementation. In fact, what the present results strongly suggest is double buffering, in the sense that both supplementation and favorable sociodemographic context can each act as a buffer. Specifically, if we look within supplementation groups, the lack of SES differences in the Atole condition clearly shows nutritional buffering, as do the differences between the Atole and the Fresco groups at the lowest SES level. However, if we look between conditions, the lack of Atole and Fresco differences at the upper-SES levels clearly shows that sociodemographic contextual factors can also act as a buffer for those children who are not nutritionally supplemented. 3
3 Relations between social class,
nutrition, and cognition suggest the operation of a buffering process, wherein
enhanced nutrition can protect against the risk of low socioeconomic status
while the protective factors that covary with high socioeconomic status can
buffer against the detrimental influence of inadequate nutrition. In contrast,
when we look at the relation between school achievement level, nutrition, and
developmental outcome, buffering does not appear to be operating; children in
the Fresco condition who achieve high grade levels do not necessarily show
superior cognitive performance over Fresco children who do not achieve high
grade levels. Rather, what appears to be operating here is a synergistic
process, wherein maximum performance is achieved with a combination of
nutritional supplementation plus success in school. This illustrates how
different predictor-criterion combinations may be governed by different
The potential influence of context becomes even more critical when we look at the explanation offered as to why early nutritional supplementation influences were maintained and expanded across time. Basically, Pollitt et al. favor a two-process model, as shown in Figure C1a. First, they propose that early nutritional differences lead to differences in physical growth; physical growth differences, in turn, influence how children are subsequently treated (e.g., smaller children are treated as younger than their chronological age, whereas larger, more mature-looking children are allowed more autonomy and independence). It is this differential treatment that is proposed as one mechanism wherein the effect of early nutritional supplementation is maintained across time. Support for certain aspects of this model comes from data other than those described in the present Monograph. For example, in both Egypt and Kenya, variability in caregiver behaviors toward 18-30-month-old toddlers was associated primarily with the level of toddler nutritional intake rather than with the level of caregiver in-take, thus demonstrating that differences in children's nutrition do relate to how children are treated (Wachs et al., 1992). Supporting the hypothesis that less adequately nourished children are treated as if they were younger, in Kenya, and to a lesser extent in Egypt, inadequately nourished toddlers were carried and held more by caregivers; toddlers who were carried and held more showed lower levels of cognitive and behavioral competence.
The second aspect of the proposed model involves higher nutritional status resulting in increased physical activity. Increased physical activity in turn results in increased exploratory behavior, which, in turn, enhances subsequent cognitive development (see Fig. C1a). Again, support for this model is found in other sources. Previous research has clearly established linkages between nutrition and activity, between activity and exploration (Schürch & Scrimshaw, 1991), and between exploration and cognitive development (Wachs, 1992).
Where the proposed model may be problematic is not so much in its general outline but rather in terms of not also considering the possibility that higher-order contextual effects (e.g., culture) may influence how nutritionally at-risk children are treated. Specifically, in the two-country study referred to above, the data from Kenya indicate that poorly fed children not only are carried more but are also responded more to by caregivers; in contrast, in Egypt, caregivers were less responsive to poorly fed children (Wachs et al., 1992). These cross-cultural differences between Kenya and Egypt may reflect differences in level of food intake in the two countries - Kenyan toddlers had significantly lower food intake than Egyptian toddlers. The fact that the nutritional level in Guatemala appears to be closer to that of Kenya than of Egypt suggests that this aspect of the authors' model may be valid for nutritional-contextual situations in which there is moderate malnutrition and where cultures support caregivers' attempts to compensate for inadequate intake by special treatment of the physically smaller child. However, the model may be less applicable in a context like Egypt, where food is more available and where poorly fed toddlers may come from families that are less able to provide for the toddlers' needs in multiple areas of development, including both nutrition and adequacy of caregiving. The critical point is that how undernourished children are treated appears to be a function not only of physical growth but also of cultural differences, suggesting a model more like that shown in Figure C1b.
A similar point can be made in regard to the nutrition-activity-exploration link. In a number of societies, infants' physical activity and attempts to explore the environment are likely to be restricted by their caregivers, either as a function of heavy maternal work loads or as a function of naturally occurring environmental hazards (Brazelton, Robey, & Collier, 1969; Kaplan & Dove, 1987; McSwain, 1981; Super, 1981). In contrast to the model postulated by the authors, in which higher levels of nutrition result in higher levels of motor behavior, which result in higher levels of exploration, in some cultures more adequately nourished children might find their attempts at motor exploration sharply restricted by their caregivers (see Fig. C1b). As a result, we would not necessarily expect a developmental advantage in some cultures for more adequately nourished, physically active children. As the authors themselves note, the child's capacity to modulate higher activity on the basis of contextual demands may be more critical than high levels of activity per se.
The alternative model offered in Figure C1b should not be seen as contradicting the main point of the model offered by Pollitt et al. I believe that they are essentially correct in suggesting that one path linking early undernutrition to later deficits in cognitive performance is mediated via child behaviors and caregiver reactivity. What I am suggesting is that the model needs to be taken one step further, namely, integrating contextual factors. Higher-order contextual factors can influence whether caregivers respond in developmentally facilitative or inhibitory ways toward more adequately nourished children; these factors can also influence the degree to which caregivers support or inhibit the child's attempts at activity and exploratory behaviors.
The results of the present project are rich, not only in terms of the actual results, but also in terms of the implications of these results for future research, theory, and intervention with children at risk.
In their initial review of the study of nutrition-behavior relations, Pollitt et al. note the possibility that critical nutritional parameters may involve micronutrients (e.g., vitamins, trace minerals) rather than energy (kilocalories) or macronutrients (e.g., protein). They also discuss some of the reasons why more recent nutrition-behavior research has shifted to experimental field studies rather than correlational studies. However, if the critical nutrient parameters are micronutrients, this raises the question of which micronutrients or combinations of micronutrients are likely to be most salient in influencing developmental variability. Correlational studies may be initially useful in dealing with this question, through assessing which micronutrients or micronutrient combinations are most consistently related to developmental variability. To the extent that correlational studies can also measure and test the role of nonnutritional covariates (e.g., morbidity, sociodemographic risk factors, caregiver behaviors), these types of studies also may be extremely useful in illustrating the nature of the multidetermined system of influences encountered by the child who is at nutritional risk. Correlational studies could form the basis for future intervention-supplementation studies, designed to separate out correlational from causal relations between nutrition, nutritional covariates, and development.
A second implication involves the finding that the effect of nutritional supplementation will be moderated by contextual factors, such as sociodemographic status and school attendance. I have noted previously in this Commentary the importance of looking at supplementation effects at different contextual levels rather than assuming that one supplement feeds all. Such a strategy illustrates the process by context design, as described by Bronfenbrenner (1993). However, I would go further and argue that this pattern of results can also illustrate the importance of looking at individual differences in reaction to treatment (nutrition or otherwise), within a given contextual level. In spite of the fact that there are marked individual differences in response to similar treatment regimens, including both biological and psychological interventions, the study of these types of individual differences in reactivity to treatment has been a relatively neglected area in the behavioral sciences (Wachs & Plomin, 1991). It will be important to continue looking at the degree to which contexts moderate treatment effects. However, it will be of equal importance not only to look at mean differences in reaction to treatment but also to look for variability in response to treatment within a given context level. For example, for low-SES children who are nutritionally supplemented in the first 2 years, what are the characteristics that distinguish those children who benefit more from supplementation from those who benefit less? This next step leads into what Bronfenbrenner (1993) has called a person by process by context research strategy.
One obvious drawback of person by process by context research is sample size, in the sense that, the more subsamples, the fewer subjects at each subsample, and the lower the power. To some extent, we may be able to compensate for potentially lower power by increased use of aggregation, more precise measurement of critical variables, and utilization of statistical procedures targeted at specific subgroup/individual effects (for a discussion of these issues, see Wachs & Plomin, 1991). Further, as shown in both the present Monograph and previously published work by Werner on resilient children (Werner & Smith, 1982), what we lose in power may be more than made up for by the richness of data obtained.
As noted earlier, the present results clearly emphasize the importance of going beyond main-effect approaches when developing models for understanding the nature of nutrition-development relations. Multidimensional multidetermined systems approaches seem to offer a much better fit to the data. Examples of these types of approaches have been developed not only for the study of nutrition per se (Pollitt, 1988) but also for the more general area of "determinants" of development (Bronfenbrenner, 1993; Horowitz, 1987; Wachs, 1992).
The present results also have implications for the question of sensitive periods in human development. Available reviews suggest that there is no strong evidence for a critical period in human development and only limited evidence for sensitive periods (Bornstein, 1989). While the sample size for the late exposure group was relatively small, the present results at least suggest the possibility of the continued, although diminishing, salience of nutritional interventions when started after 2 years of age. These results do not support a critical periods notion, but they are not inconsistent with theories based on periods of maximum sensitivity, in the sense that, while earlier may be better, later may still offer some benefits.
Implications for Intervention
Particularly in less developed countries, theoretically based research is seen as having less value than research that has practical implications for individuals' day-to-day lives (Nsamenang, 1992). Pollitt et al. clearly share this concern, as exemplified by their discussion of public policy issues. A major policy issue that comes from the present research is the question of which children should be targeted for nutritional intervention. In an era of increasingly scarce resources, can we afford to target all children who are potentially at risk, or should interventions target primarily those children who are most at risk? If the latter, which children? In the present project, maximum risk appears to occur for those children who are simultaneously exposed to inadequate nutrition (no supplementation), low social class, and low levels of school attainment (the three-way interaction demonstrated) with Raven's Progressive Matrices).
Clearly, the present results suggest the importance of targeting low-socioeconomic-status children for nutritional intervention. Providing nutritional supplementation for these children can be seen as one way of breaking the naturally occurring covariance between inadequate rearing environments and inadequate nutrition. However, Pollitt et al. go beyond low SES and also argue for the importance of considering grade attainment as another potential risk factor that may call for nutritional intervention. Their argument is based, in part, on the synergistic interaction between treatment and grade level as well as on the assumption that, even fat the upper levels of the SES distribution, there may be only a limited amount of buffering that a poor rural environment can offer to a child. In contrast, I would argue that the data presented in this Monograph seem to suggest that supplementation should be directed primarily toward low-SES children.
There are two reasons for emphasizing SES. First, in contrast to the authors' argument that there is only so much environmental buffering that can be offered to children in a poor rural community, the SES buffering effects for Fresco children show that something positive is being offered to children, even within this relatively restricted context. Second, the results at least suggest a diminishing effect of nutritional intervention when started after 2 years of age. If the strongest effects of nutrition are shown for children in the first 2 years of life, then the only possible targeting is social class since at-risk children are not yet enrolled in school. What the present results could suggest is a two-stage process. In the first stage, nutritional intervention would be directed primarily toward children in the lowest social class groups. In the second stage, after children reach school age, there should be a secondary emphasis on economic aid to families, to allow supplemented children to remain in school as long as possible.
The present project is a major contribution to the literature, not only in terms of demonstrating that early nutritional supplementation can have long-term effects on cognition, but also in terms of illustrating potential processes whereby nutritional influences interact with the overall context within which the individual functions. There has been a slowly increasing emphasis in the literature on biological and contextual linkages. Most of our current efforts in this direction have involved linkages between genes and environments (e.g., Plomin & McClearn, in press). The present results suggest that an equally fruitful path may lie in exploring linkages between nutrition, context, and the implication of these linkages for developmental variability.
Bornstein, M. (1989). Sensitive periods in development. Psychological Bulletin, 105, 179197.
Brazelton, T., Robey, J., & Collier, G. (1969). Infant development in the Zinacanteco Indians of southern Mexico. Pediatrics, 4, 274-309.
Bronfenbrenner, U. (1993). Ecological system theory. In R. Wozniak & K. Fisher (Eds.), Specific environments: Thinking in contexts. Hillsdale, NJ: Erlbaum.
Brozek, J. (1978). Nutrition, malnutrition and behavior. Annual Reviews of Psychology, 29, 157-178.
Horowitz, F. (1987). Exploring developmental theories. Hillsdale, NJ: Erlbaum.
Kail, R. (1991). Development of processing speed in childhood and adolescence. Advances in Child Development and Behavior, 23, 151-185.
Kaplan, H., & Dove, H. (1987). Infant development among the Ache in Eastern Paraguay. Developmental Psychology, 23, 109 - 198.
Karp, R. (in press). Malnourished children in the United States: Caught in the cycle of poverty. New York: Springer.
McSwain, R. (1981). Care and conflict in infant development. Infant Behavior and Development, 4, 225-246.
Nsamenang, A. (1992). Human development in cultural contexts. Newbury Park, CA: Sage.
Plomin, R. (1990). The role of inheritance in behavior. Science, 248, 183-188.
Plomin, R., & McClearn, G. (in press). Nature, nurture and psychology. Washington, DC: American Psychological Association.
Pollitt, E. (1988). A critical view of three decades of research on the effects of chronic malnutrition on behavioral development. In B. Schürch & N. Scrimshaw (Eds.), Chronic energy deficiency: Consequences and related issues. Lausanne: International Dietary Energy Consultative Group.
Rushton, P., Brainerd, C., & Pressley, M. (1983). Behavioral development and construct validity: The principle of aggregation. Psychological Bulletin, 94, 18-38.
Schopflin, U., & Muller-Brettel, M. (1990). International Journal of Behavioral Development: Scope and trends. International Journal of Behavioral Development, 13, 393-406.
Schürch, B., & Scrimshaw, N. (Eds.). (1991). Activity, energy expenditure and energy requirement of infants and children. Lausanne: International Dietary Energy Consultative Group.
Simeon, D., & Grantham-McGregor, S. (1990). Nutritional deficiencies and children's behavioral and mental development. Nutrition Research Review, 3, 1-24.
Super, C. (1981). Behavioral development in infancy. In R. H. Munroe, R. L. Munroe, & B. Whiting (Eds)., Handbook of cross-cultural human development. New York: Garland.
Wachs, T. D. (1992). The nature of nurture. Newbury Park, CA: Sage.
Wachs, T. D. (1993). Determinants of intellectual development: Single determinant research in a multi-determinant universe. Intelligence, 17, 1-9.
Wachs, T. D., & Plomin, R. (1991). Conceptualization and measurement of organism-environment interaction. Washington, DC: American Psychological Association.
Wachs, T. D., Sigman, M., Bishry, Z., Moussa, W., Jerome, N., Neumann, C., Bwibo, N., & McDonald, M. (1992). Caregiver child interaction patterns in two cultures in relation to nutrition. International Journal of Behavioral Development, 15, 1-18.
Wahlsten, D. (1990). Insensitivity of the analysis of variance to heredity-environment interactions. Behavior and Brain Sciences, 13, 109-161.
Werner, E., & Smith, R. (1982). Vulnerable but invincible. New York: McGraw-Hill.