![[Feedback]](/icons/banner/feedbackACT.gif){kind=icon}
![[For Subscribers]](/icons/banner/subscriberACT.gif){kind=icon}
![[Archive]](/icons/banner/archiveACT.gif){kind=icon}
![[Search]](/icons/banner/searchACT.gif){kind=icon}
![[Contents]](/icons/banner/tocACT.gif){kind=icon}
|
|
|
|
|
||
Department of Epidemiology and Population Health, London School of Hygiene and Tropical Medicine, London, UK and * Department of International Health, Emory University, Atlanta, GA
1To whom correspondence should be addressed. E-mail: saul.morris{at}lshtm.ac.uk.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
0.85). The school height census estimates of mean height-for-age Z-score at the municipality level were also valid, with the reference criterion the survey results for children 1 y of age (Spearman’s rank correlation = 0.74). School height censuses cannot provide reliable estimates of levels of growth retardation in individual schools. Wider use of school height censuses could make it much easier to identify communities that might benefit from targeted nutrition interventions.
KEY WORDS: • school height census • targeting • growth retardation • children • Honduras
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
). Similarly, for nutrition interventions, there are likely to be substantial advantages to focusing activities where malnutrition is most common (2
). Program planners would ideally direct interventions to areas in which nutritional deficiencies are widespread and, in addition, the target population is able to respond to interventions to improve growth (3
). Information on responsiveness to nutrition interventions is, however, not widely available.
Stunting (inadequate length/height-for-age) in children < 5 y old is an indicator thought to be "suitable for targeting a wide range of interventions because it both reflects the cumulative effects of socioeconomic, health, and nutrition problems ... and varies widely from place to place" (4
). Information on national prevalences of stunting is now highly accessible because anthropometric surveys have been carried out in virtually every country in the world and are routinely collated by the WHO (5
). Many of these surveys are sufficiently large to permit disaggregation at least to the level of major subnational regions. However, only a few of the largest surveys permit disaggregation to the level of first-level administrative divisions (e.g., states, departments), and none are likely to permit the characterization of second-level administrative divisions (e.g., municipalities). This creates difficulties for program planners because most interventions are implemented at the level of communities or local administrative units, where there may be insufficient information available to enable the effective prioritization of the most needy areas. The school height census is an alternative, low-cost approach that does provide disaggregated data at the local level. However, not all nutrition planners are convinced of the interpretability of the data that are produced. This paper provides a theoretical argument for using the school height census for targeting nutrition interventions at a relatively fine level of geographic disaggregation. Then, on the basis of empirical evidence from the Central American nation of Honduras, the reliability and validity of the method are assessed.
Theoretical basis for the use of school height censuses in targeting nutrition interventions
School height censuses provide summary statistics on the distribution of growth retardation within and between geographic units. The basic methodology has been described by Valverde and colleagues (6
), and has been implemented extensively in Latin America (7
). Customarily, all children in y 1 of primary education are measured, and for ease of interpretation, the raw height measures are transformed into age- and sex-standardized measures based on the National Center for Health Statistics/WHO international growth reference population (4
). These measures are referred to as height-for-age Z-scores. Widely used summary statistics for population groups include mean height-for-age Z-score, which we will denote HAZ,2
and the proportion of children with Z-scores of less than -2, which we will denote %STUNTED. Because height-for-age Z-scores are normally distributed with relatively little cross-community variation in their SD, there is usually a near-perfect correlation at the community level between these two measures. Thus, for targeting purposes, they may be used interchangeably, and in the remainder of this paper we will refer only to HAZ.
HAZ is highly sensitive to the age and sex composition of the group of children being measured.
Typically, in developing countries, HAZ falls dramatically in y 1 of life and stabilizes over the course of y 2 (8
), as shown in Figure 1
. Unlike in anthropometric surveys, the age range of the children measured in school height censuses will typically be fairly narrow. Suppose, for example, that all of the children measured are 6 y old at the time of the Census. In this case, the cohort of children born 6 y before the census date will have experienced a continuous evolution of its HAZ over time (Fig. 2
). The horizontal axis of this figure represents time in years relative to the date of the census, whereas the vertical axis shows the (instantaneous) rate of change in HAZ experienced by this cohort of children, denoted 
HAZ/AGE. The graph illustrates that virtually all of the evolution (change) in these children’s HAZ occurred >5 y before the date of the School Height Census. This observation has important implications for the interpretation of school height census data.
|
|
When the coverage of the primary school system is not high, school height censuses are likely to give rise to biased estimates. This is because children who do not attend school tend to differ systematically from those who do attend school, in ways that are most often associated with anthropometric status. Variable coverage of the school system can thus distort interregional comparisons of anthropometric status even when most regions of the country have high coverage.
If age-specific growth rates have been evolving rapidly over time, there is no reason to expect that estimates of HAZ derived from the school height census will mirror those derived from an anthropometric survey. This is because the children in the school height census experienced the bulk of the evolution in their HAZ 5 y ago, whereas the younger children experienced the bulk of the evolution in their HAZ much more recently, when conditions may have been different. This situation might arise in countries (or parts of countries) in which natural disasters or economic shocks have seriously disrupted the livelihoods of the poor, such as has been documented in Brazzaville, Congo (9
). Additionally, if the anthropometric survey includes infants, the resultant estimate of HAZ is always expected to be higher than that derived from the school height census because the infants have yet to reach a "stable" level of growth retardation (as a group). The difference between the two estimates will depend critically on the proportion of infants present in the survey sample.
Even in the absence of bias, there may still be a discrepancy between estimates of HAZ derived from a school height census and those derived from a survey as a result of random variation. Random variation in surveys is directly related to the number of children measured (sample size), and lack of random error in epidemiologic measurements is referred to as "precision" (10
). Although sample size is not a problem for censuses, it remains the case that levels of growth retardation in any given area are not perfectly stable from one year to the next. This year-to-year fluctuation is caused by purely idiosyncratic features of the children present in first grade in any particular year, and also by evolving social, economic and environmental conditions. If these period-to-period fluctuations are large, the measurement procedure is said to be "unreliable" (11
), and the potential benefit of using it for targeting is greatly reduced.
|
SUBJECTS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
In each census, height measurements were undertaken by the children’s usual teacher according to a standardized methodology that has not changed. All teachers received a full set of equipment (a plumb line, a thick cardboard strip marked in 0.5 cm increments up to 150 cm, a pencil, a wooden right-angle for use as a head-piece, adhesive tape, thumb tacks, and forms) and an extensively illustrated manual explaining in detail: 1) how to choose an appropriate location for the measurement (smooth floor, no baseboard, smooth wall at right angles to the floor); 2) how to use the plumb line to affix the measuring strip exactly vertically, and with the base of the strip level with the floor; 3) how to prepare the child, removing shoes and hair ornaments, and correctly stand them with shoulders, knees, and heels against the wall; 4) how to measure the child (sliding the head-piece down against the top of the head, asking the child to then step aside crouching slightly, and reading the number exposed below the bottom of the head-piece); and 5) how to record the information correctly on the form.
Training was carried out in three tiers, with a central team training departmental supervisors, the departmental supervisors training district supervisors and the district supervisors training the primary school teachers in their district. The training included a standardization exercise based on the measurement of 5–10 first grade children. The census itself was conducted in the same week across the country, a few days after completion of the training. District and departmental supervisors visited a selection of schools in their area to ensure quality, and completed forms were sent back to the central office for data processing.
For the purpose of this analysis, implausible height-for-age Z-scores (those lower than -6 or greater than + 4) were recoded to missing. In 1 y (1994), the proportion of missing values was 4.3% of the total, but for all other years it was 
2%. Although specific permission was not sought from parents, it is common knowledge in Honduras that children in first grade will be measured, and that the findings will be made public at an aggregate level.
For the validity analysis, the 2001 school height census data were compared with data from an anthropometric survey conducted in 70 municipalities (out of a total of 299 in the country) between August and December 2000. The survey formed part of the baseline data collection for a large government program (12
) and was designed to be representative at the municipality level. The municipalities were chosen because they showed high prevalences of stunting in the 1997 school height census, ranging from 57 to 96% (mean 72%; SD, 8%). Within each municipality, eight census segments (as defined by the national population census) were selected by systematic sampling with probability proportional to size. The census segments generally comprised 
70 dwellings, and had been mapped for the whole country by the national census agency in the months immediately preceding the survey. In 40 of the 70 municipalities, the number of occupied dwellings in each segment was known exactly due to previous project activities. In the remaining 30 municipalities, this figure was estimated on the basis of the mapping activities of the national census agency. Within each segment, a random start point was then selected. The following procedure was used: a random number from 1 to Ns was selected, with Ns being the number of occupied houses in segment s. The dwelling with this number was located on the segment map, and its location relative to a major landmark (e.g., stream, football pitch or village square) was determined. The fieldworker then located the landmark and found the house with the predetermined bearings. This procedure ensured that even if the originally selected house had been knocked down, the chosen start point was truly random within the segment. Ten consecutive occupied dwellings were then visited, following the direction of the numbering on the segment maps. Up to three attempts were made to interview the chosen families. The family could not be contacted in 2.2% of all visits, and there was a 1.1% rejection rate. Additionally, 0.5% of interviews were incomplete. Anthropometric measurement (weight and height/length) was attempted for all children <5 y old in the survey households, and 96.6% of these children were successfully measured. Height was assessed using locally made equipment by nine specially trained anthropometrists. The anthropometrists were standardized using the procedures described by Habicht (13
). The purposes of the study and the details of the measurement procedures were carefully explained to each child’s principal care giver, and their consent was obtained for the measurement. Because neither the Honduran government nor the International Food Policy Research Institute has in-house capacity for ethical review, the study was certified as compliant with the regulations on the protection of human subjects by the relevant review board of the Instituto de Nutrición de Centroamérica y Panamá (INCAP).
For the assessment of reliability [defined as the extent to which the measurement procedure yields the same results on repeated trials (11
)], the five school height censuses were aggregated at each of the departamento (first-level administrative unit), municipality (second-level administrative unit) and school levels. An average departamento in Honduras had 4000–11,000 first-time, first-grade students (interquartile range), whereas a municipality had 800-2800 and a school had 25–100. Individual data were aggregated using the indicator mean height-for-age Z-score (HAZ). The total variance in HAZ in the aggregated file was then decomposed into its between-departamento and within-departamento components (subsequently, between- and within-municipalities, and between- and within-schools), using a random-effects ANOVA model (14
). This decomposition permitted estimation of the intraclass correlation 
, a measure of how stable the departamento-level (or municipality-, or school-level) estimates were over time. The intraclass correlation varies from = 0, which would indicate no stability at all over time, to = 1, which would indicate perfect stability. It can conveniently be interpreted as one minus the proportion of the total variation that can be (dis)regarded as "noise." It should be noted that this quantity is affected both by measurement error inherent in the method and by any true changes in patterns of stunting over the 5-y period.
To visualize more precisely the effects of the unreliability of the aggregate estimates of nutritional status derived from the school height censuses, departamento-level, municipality-level and school-level means (HAZ) were plotted against year of measurement, with observations from the same location connected by a continuous line. In the cases of the municipalities and the schools, plotting all of the lines would have resulted in an excessively busy graph; thus only judiciously selection lines were plotted, as suggested by Jones and Rice (15
). This method of selecting lines to be plotted involved the following: first, municipalities/schools were classified according to their mean value of HAZ averaged over the full 5 y of data. Municipalities/schools with <5 y of observations were excluded. Then those observations with the highest and lowest values of the classification variable, in addition to those at (or closest to) the 5th, 10th, 25th, 50th, 75th, 90th and 95th percentiles of the distribution, were plotted on a graph. Municipal/school-level series that are stable over time result in a set of straight, parallel lines, whereas series that are relatively unstable result in a jumble of criss-crossing lines.
For the assessment of criterion-related validity, the extent to which the instrument predicts values obtained from an external reference, the municipality-level estimates of HAZ from the March 2001 school height census were compared with the values obtained in the anthropometric survey described above. The degree of correspondence between an instrument and its reference is "usually estimated by the size of their correlation" (11
). However, in this case, we considered the rank correlation (16
) to be the most appropriate measure because geographic targeting usually requires only the ability to rank areas from best to worst, without specification of the precise levels of growth retardation in each area. It should be noted that any observed correlation between local estimates of HAZ derived from the survey and the school height census will not be as high as it would be if both methods were perfectly reliable (11
). That is to say, that both the unreliability of the school height census results and, in particular, the sampling error associated with the survey estimates will result in an attenuation of the "true" correlation between the two methods.
Occasionally, geographic targeting algorithms require actual levels of HAZ/%STUNTED to be known. In these cases, it would be important to know the form of the (linear) relationship between the estimates derived from each of two possible sources. Three situations can be distinguished:
 |
 |
 |
The first situation implies that there is no systematic tendency for survey estimates to differ from census estimates. The second situation, on the other hand, indicates that the survey estimates tend to differ from census estimates by a constant amount, 
. For example, survey estimates of HAZ may be routinely half a Z-score lower than census estimates. Finally, the third situation implies that a given change in HAZ in the census data will not be reflected in a change of the same magnitude in the estimates derived from surveys. To test these hypotheses using the Honduras data, the form of the linear association between the 2000 survey estimates of municipality-level HAZ, and the corresponding estimates from the 2001 school height census, was estimated using linear regression (17
). Wald tests (18
) were used to test the hypotheses = 0 and ß = 1. All analyses were conducted using Stata version 7.0 (Stata Corporation, College Station, TX).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Over the 5 y 1993–1997, there was a small, general deterioration in HAZ in the eighteen departamentos of Honduras (Fig. 3
). In spite of this, the ordering and relative distance of the 18 departamentos varied little from year to year. This stability was reflected in a very high intra-departamento (i.e., intraclass) correlation coefficient (HAZ) of 0.95.
|
= 0.85. To illustrate this pattern, nine municipalities were chosen because they were located at strategic points of the distribution of average HAZ, as described in the Subjects and Methods section. Once again, most of the lines were flat and ran almost parallel to one another (Fig. 4
). Only the municipality with the highest average HAZ showed significant year-to-year variation; it is a small, Caribbean island-possession of Honduras, with < 40 first-time, first-grade students each year.
|
= 0.51. Nine schools were chosen for illustration because they were located at strategic points of the distribution of average HAZ, as described above. The ranking of these nine schools clearly was not maintained from one year to the next (Fig. 5
).
|
There was a strong and clearly visible association between estimates of HAZ derived from the 2001 school height census and the 2000 anthropometric survey in 70 municipalities in the West of Honduras (Fig. 6
). The vertical axis of this figure shows the estimates derived from a representative anthropometric survey of children <5 y old conducted in August-December 2000, with vertical lines illustrating the sampling variability associated with these estimates (95% confidence intervals around the means). The location of each line on the horizontal axis indicates the value of HAZ that was observed for the same municipality in the 2001 School Height Census. It is immediately clear that the two-stage sampling strategy used in the survey, combined with the relatively limited number of children measured per municipality (mean 76.5, range 50–108), led to quite substantial sampling variability around the point estimates of HAZ for each municipality. The rank correlation between the survey-based estimates of municipal HAZ and the School Census-based estimates of the same indicator was r = 0.74. However, as indicated in the Subjects and Methods section, this correlation was almost certainly attenuated because of measurement error in both variables.
|
 |
This result, based on 70 data points, indicates that it was not possible to reject the hypotheses that 1) an increase of one Z-score in the census estimates was reflected in an exactly equal change in the survey estimates (P = 0.67), and 2) the true intercept was zero (P = 0.35).
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
800–2800 first-graders, and finally schools, with only 25–100. Data on individual children were aggregated using mean height-for-age Z-scores (HAZ) rather than the proportion of children below -2 Z-scores (%STUNTED), because the former measure uses all available information and is therefore more efficient in the context of small rural schools with a very small numbers of students. This choice is not expected to have substantially affected the study conclusions. Using data from five consecutive school height censuses conducted in Honduras between 1993 and 1997, we were able to show that age-standardized height data summarized at the level of departamentos showed very little year-to-year fluctuation. This means that given relatively stable conditions such as prevailed in Honduras between 1993 and 1997, such data can be used for geographic targeting of interventions without having to worry that it may become outdated before the completion of the project. Data summarized at the level of municipalities, on the other hand, did show some limited fluctuation from year to year. These fluctuations may in some cases represent shifts in particular municipalities’ long- or medium-term welfare ranking, but in other cases they will be essentially random, resulting only from the fact that different "batches" of children are measured in different years. They are important only if they lead to a poor correlation with the reference criterion, the anthropometric status of children < 5 y old. The present analysis, however, demonstrated a strikingly high correlation between school height census data and a representative anthropometric survey of children aged 12–59 mo conducted 3–7 mo earlier in 70 municipalities in the west of Honduras. This unquestionably indicates that the small amount of unreliability in the municipality-level, school height census estimates of HAZ is unimportant for targeting purposes.
As indicated by Carmines and Zeller (11
), the assessment of criterion-related validity is always complicated by the fact that the criterion itself may not be measured validly. In the present case, exceptional care was devoted to ensuring that the survey results were representative at the municipality level, and the validity of this measure is therefore thought to be high. However, the survey estimates were characterized by substantial levels of sampling variability, predictably resulting from the limited sample size of only 80 households in each municipality, and aggravated by the two-stage sampling design. The correlation between survey estimates and anything else will always be attenuated by the random error associated with the survey estimates. For this reason, Parillón and co-workers (19
) previously proposed downweighting nutritional survey data relative to school height census data when deriving a compound index of local nutritional vulnerability, in their case for Panamá. Nonetheless, these authors, like us, found a high level of correlation between survey-based and school height census–based measures of stunting at the district level. We believe that in Honduras, if the survey had not been limited to those municipalities with known nutritional problems, the observed correlation might have been even higher.
Another important, and perhaps counterintuitive finding from this analysis is that schools cannot be effectively targeted using data from a single school height census. Even though we based our analysis on means (HAZ) rather than proportions (%STUNTED) to minimize the effect of the imprecision of the school-level estimates, we still found that these estimates were highly unreliable. This unreliability was extremely obvious in Figure 5
, which shows how HAZ defined at the level of the individual school fluctuated wildly from year to year. Better reliability could be achieved by averaging over several different years of data, but this would be at the cost of sacrificing the sensitivity of the method to short-term local shocks. Alternatively, planners wishing to target schools may be forced to seek alternative, more stable indicators of risk at the school level.
One concern that applies to both school height census data and, to a lesser extent, specialized nutrition surveys is the fact that the indicators derived may be more informative about social processes operating several years before the measurement date than they are about current conditions. In the case of school height censuses, it may generally be assumed that the group of children studied evolved into their present state up to five years before measurement. This may not be a concern in isolated areas with little change in underlying conditions, but seriously mars the validity of the height census for small-area targeting in areas in which there is substantial movement of population, or rapid social or economic change. Because there is no obvious means of overcoming this problem, particular attention should be paid to trends that emerge from the comparison of several years’ height census data. For example, in Honduras over the period 1993–1997, there seems to have been a trend toward worsening nutritional status in the small cities and rural areas, a trend that may be expected to exacerbate rather than attenuate existing geographic differentials.
There has now been considerable experience with using school height censuses, especially in Central America and Panamá (20
). The quality of the height information collected by teachers has been shown to be highly reliable at the level of the individual child in Costa Rica, and at the group level in Guatemala and Nicaragua (6
). The current analysis has demonstrated that this instrument is both valid and reliable for small-area targeting in Honduras. Although validity and reliability could of course be lower in situations in which the measuring instrument was less precise, the training less complete or the teachers less committed, we believe that high quality data are achievable. Geographic targeting is a relatively simple way of concentrating resources in nutrition investments to reach a minimum effective input to beneficiary ratio. Previous commentators have shown that for social programs, geographic targeting can significantly improve poverty effect, and that the more narrowly defined the geographic unit, the greater the reduction in poverty (1
). We showed previously that the same positive effects are expected in the case of nutrition investments (2
). We are now able to confirm that school height censuses can provide a valid and reliable means of identifying communities of high risk preschool children, before these children actually reach school age and it is too late to intervene. We hope that in years to come, more countries will adopt this methodology as a key element of their strategic planning in nutrition.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
HAZ/AGE, the instantaneous rate of change in mean height-for-age Z-scores; %STUNTED, proportion of children with height-for-age Z-scores < -2. Manuscript received 16 November 2001. Initial review completed 3 January 2002. Revision accepted 25 February 2002.
|
LITERATURE CITED |
|---|
|
TOPABSTRACTINTRODUCTIONSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
1. Baker, J. L. & Grosh, M. E. (1994) Poverty reduction through geographic targeting: how well does it work?. World Dev 22:983-995.
2.
Morris, S. S., Flores, R. & Zúniga, M. (2000) Geographic targeting of nutrition programs can substantially affect the severity of stunting in Honduras. J. Nutr. 130:2514-2519.
3.
Ruel, M. T., Habicht, J.-P., Rasmussen, K. M. & Martorell, R. (1996) Screening for nutrition interventions: the risk or the differential-benefit approach?. Am. J. Clin. Nutr 63:671-677.
4. WHO Expert Committee on Physical Status (1995) Physical Status: the Use and Interpretation of Anthropometry. Report of a WHO Expert Committee. Technical Report Series No. 854 1995 World Health Organization Geneva, Switzerland .
5. () WHO Global Database on Child Growth and Malnutrition World Health Organization, Department of Nutrition for Health and Development http://www.who.int/nutgrowthdb. Last accessed, January 17, 2002.
6. Valverde, V., Delgado, H., Flores, R., Sibrián, R. & Palmieri, M. (1985) The school as a data source for food and nutrition surveillance systems in Central America and Panamá. Food Nutr. Bull. 7:32-37.
7. United Nations Administrative Committee on Co-ordination, Administrative Sub-committee on Nutrition (2000) Nutrition throughout the Life Cycle: 4th Report on the World Nutrition Situation 2000 United Nations Geneva, Switzerland .
8.
Shrimpton, R., Victora, C. V., de Onis, M., Costa Lima, R., Blössner, M. & Clugston, G. (2001) Worldwide timing of growth faltering: implications for nutritional interventions. Pediatrics 107:e75.
9. Martin-Prével, Y., Delpeuch, F., Traissac, P., Massamba, J.-P., Adoua-Oyila, G., Coudert, K. & Trèche, S. (2000) Deterioration in the nutritional status of young children and their mothers in Brazzaville, Congo, following the 1994 devaluation of the CFA franc. Bull. WHO 78:108-118.[Medline]
10. Rothman, K. J. (1986) Modern Epidemiology 1986 Little, Brown & Company Boston, MA .
11. Carmines, E. G. & Zeller, R. A. (1979) Reliability and Validity Assessment. Quantitative Applications in the Social Sciences Paper No. 17 1979 Sage Publications Newbury Park, CA .
12. () PRAF: The Family Allowance Program International Food Policy Research Institute http://www.ifpri.org/themes/praf.htm. Last accessed: January 17, 2002.
13. Habicht, J.-P. (1974) Standardization of quantitative epidemiologic methods in the field. Bol. Of. Sanit. Panam. 76:375-384.
14. StataCorp (1999) Stata statistical software: release 6.0. vol. 2: H–O 1999 Stata Corporation College Station, TX .
15. Jones, M. C. & Rice, J. A. (1992) Displaying the important features of large collections of similar curves. Am. Statistician 46:140-145.
16. Spearman, C. (1904) The proof and measurement of association between two things. Am. J. Psych. 15:72-101.
17. Kirkwood, B. R. (1988) Essentials of Medical Statistics 1988 Blackwell Scientific Publications Oxford, UK .
18. Judge, G. G., Griffiths, W. E., Hill, R. C., Lütkepohl, H. & Lee, T.-C. (1985) The Theory and Practice of Econometrics 2nd ed. 1985 John Wiley & Sons New York, NY .
19. Parillón, C., Valverde, V. & Delgado, H. (1988) Descripción de una metodología para localizar y cuantificar grupos de familias pobres y desnutridas en la República de Panamá. Arch. Latinoam. Nutr. 38:31-41.[Medline]
20. Delgado, H., Palma, P. & Fischer, M. (1991) The use of the height census of schoolchildren in Central America and Panamá. Food Nutr. Bull. 13:17-19.
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
