Contents - Previous - Next

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

Benefits of a school breakfast programme among Andean children in Huaraz, Peru

Enrique Jacoby, Santiago Cueto, and Ernesto Pollitt


A randomized, controlled trial measured the short-term impact of the Peruvian school breakfast programme on the diet, school attendance, and cognition of fourth- and fifth-graders. Ten schools on the rural outskirts of the Andean city of Huaraz were randomly assigned to a control or a treatment group. The programme significantly increased (p <.01) dietary intakes of energy by 15.2%, protein by 16.1%, and iron by 60%, and improved rates of attendance. Analysis of covariance also showed improved performance on a vocabulary test among heavier children, as indicated by a positive and significant weight x treatment interaction (parameter = 0.37; F = 4.97; p < .05). It is plausible that among stunted children, being relatively overweight (by the weight-for-height index) poses a cognitive risk that can be partially reversed by a school breakfast.


In 1993 the government of Peru, through the Fondo de Compensación pare el Desarrollo Social (FONCODES), launched a school breakfast programme in five poor Andean departments. The objectives were to promote better nutrition and educational achievement, and to improve attendance among children enrolled in public primary schools. These benefits were intended to compensate in part for the social costs of the economic adjustment policies implemented by the Peruvian government since late 1990. The nutritional and operational aspects of the programme were delegated to the Instituto de Investigacion Nutricional (IIN), a local private, non-profit organization devoted to research on nutrition. This arrangement allowed for the inclusion of a research and evaluation component, conjointly with implementation of the school breakfast programme. Such a dual process is rather atypical among these popular programmes and presents an ideal situation for public health programming.

Two studies were conducted in Huaraz, a city in the central Andes, to assess the impact of the programme. A field evaluation in 10 rural schools tested the proposition that the programme improved children's diet, attendance at school, and performance on psychoeducational tests. An experimental study of urban children assessed the effects of the school breakfast on information processing. This report is based on the first of the two studies.

Theoretical reasoning and empirical data support the assumption that breakfast facilitates learning in the classroom, particularly among children nutritionally at risk, through at least two mechanisms. One is by preventing or gradually remediating nutritional deficiencies (e.g., iron-deficiency anaemia) that interfere with school performance. Although information on this effect is scant [1], it is theoretically plausible. The other is by activating mechanisms that compensate for the decline in available brain fuel during the overnight fast. Under fasting conditions, the rate of the drop in glucose among children is greater than in adults [2], and under experimental conditions, healthy children process information better after eating breakfast than after a morning fast [3].

With exceptions [4], evaluations of school feeding programmes are damaged by serious methodological shortcomings, independently of whether they were conducted in developed [5, 6] or developing [7] countries. The methodological requirements for evaluating nutrition interventions also apply to the assessments of school feeding programmes [8, 9]. The gold standard is the clinical trial, including random assignment to experimental versus placebo conditions, and a double-blind assessment schedule. These conditions are almost universally lacking, partly because most evaluations are implemented in continuing programmes. A second distinctive requirement is assessment of the net increment in the habitual intake produced by the supplement. As clearly stated elsewhere [8], there is no reason to assume that food-distribution programmes produce significant increments in habitual intake.

A third condition is the sensitivity and specificity of the tests of outcomes. Most evaluations of the effects of school feeding programmes on school performance have paid little attention to the construct validity of the tests and their estimates of reliability [5]. Effects are not likely to be observed unless the tests measure what they are intended to measure, and the error variance in the test scores is narrow. These psychometric issues are critical for valid evaluations of effects on cognition and learning.

As mentioned, this is a report on the short-term effects of the school feeding programme in Peru on dietary intake, school attendance, and performance on a battery of psychoeducational tests. The postulated effects on performance are mediated by shortterm metabolic changes associated with adaptation to a morning fast in the school setting. A clinical trial design was used, with random assignment of schools.


Hypotheses and study design

It was hypothesized that the programme breakfast, in the short run, would enhance children's cognitive abilities and scholastic performance, produce a net increment in children's dietary intake, and improve school attendance rates.

Since the study was conducted under the realworld conditions of the school setting, it was an effectiveness trial. Ten schools enrolled in the school breakfast programme were randomly assigned to a treatment (Rx) or a control condition using a table of random numbers. This design did not prevent the control group from participating in the school breakfast programme, because the schedule for school participation was incremental. Table 1 shows the timetable of activities followed.

The school breakfast programme

The school breakfast programme started in April 1993 in the capital cities and rural periphery of five Andean provinces. The selection criteria combined the poverty level and the prevalence of infant and child malnutrition in the region [10]. All public schools were invited to participate, and virtually all were enrolled in each city. Implementation and operation were the responsibility of a network of school parent organizations, together with the teachers. All activities were organized and routinely supervised by programme field workers.

The IIN designed the nutritional content of the breakfast and tested several prototypes for palatability and ease of preparation. The final preparation consisted of four cookies and an instant drink. To avoid habituation, these were periodically alternated with a cake and drinks of different flavours and similar nutritional content. Quality control of the products at production, delivery, and storage was maintained through regular spot checks. For an average primary-school child (5-10 yr) the nutritional content of the breakfast (beverage and solid portion) provided 30% of the energy requirements as defined by WHO [11], 60% of the recommended dietary allowances (RDA) for minerals and vitamins [12], and 100% of the dietary iron needs [13] (table 2).

TABLE 1. Schedule of activities in the evaluation of the school breakfast programme in Huaraz, Peru

Group Time 0 Time 1 Time 2 Time 3
Sep 6-30 Oct 1-15 Oct 20-Nov8 Nov 12-30
Treatment Data collection Programme starts Data collection Data collection
Control Data collection   Data collection Data collection

Programme starts

Location, population, and subjects

The study was conducted among schools in the rural periphery of the city of Huaraz, located at an elevation of 3,300 meters 400 km north of Lima. Huaraz is part of Callejon de Huaylas, one of the three largest inter-Andean valleys of Peru, with nearly 90% of its 300,000 inhabitants being ethnically Qucchua. Most are bilingual except for older women, who speak primarily the Quechua language. Economic activity revolves around agriculture, with small-scale production in the hands of small landowners accounting for more than 80% of total marketable crops in the region. Houses are typically built with adobe walls and tile roof, have dirt floors, and contain no more than three rooms.

The schools selected had to be no more than one hour from the city by ground transportation and had to be coeducational public schools already enrolled for participation in the school breakfast programme, with at least 40 students in the fourth and fifth grades, and recipients of the Ministry of Agriculture lunch programme. This last criterion was adopted to maintain a nutritional intervention factor constant among groups.

The sample size was calculated based on the formula, shown below, proposed for randomizing by clusters [15]. In the present case, a school including all fourth- and fifth-graders was considered a cluster.

Subjects per group = 2(Za-Zb)2 s2 [1 +(n-1 ) p]/d2

We selected an 80% chance at a Z,, significance level of 5% of finding a significant difference (J) of 6 points on cognitive test scores between the two groups. From a previous pilot study, we estimated an intracluster correlation (p) of 0.2 and a variance of 36. The sample size per school (n) was set at 40 after examination of local information. The result yielded a sample size of five schools per study group.

It is worth noting that the sample size calculations are relevant only for analyses at the school level, as well as for drawing statistical inferences regarding all scholastic tests planned (math, vocabulary, and reading comprehension) and two of four cognitive tests (coding and digit span).

Of 25 schools that met the distance criteria, 15 were visited and 10 were selected. Five were excluded because they were already scheduled to begin the breakfast programme. Although an effort was made at the time of selection to minimize the interschool variability in the numbers of students and classrooms, some differences prevailed at the end.

TABLE 2. Nutritional content of the breakfast distributed by the school breakfast programme

Nutrients Beverage Solid Total Amount of requirement ( % )
Energy (kcal) 200 400 600 33a
Protein (g) 6.5 13.0 19.5 70b
Vitamin A (pa) 240   240 60c
Vitamin B1 (mg) 0.6   0.6 60b
Vitamin B6 (mg) 0.72   0.72 60b
Vitamin B12 (mg) 0.84   0.84 60b
Niacin (mg) 7.8   7.8 60b
Folic acid (µg) 60   60 60c
Iron (mg) 7.2 5.0 12.2 100c
Calcium (mg) 480   480 60b
Phosphorus (mg) 480   480 60b
Zinc (mg) 6.0   6.0 60b

a Ref. 11.
b Ref. 12.
c Ref. 13.
Source of table: ref. 14.

Outcomes, effect modifiers, and measurements

Dietary evaluation

To ascertain the school breakfast programme operation, we monitored breakfast consumption at each school three times: twice before administration of the tests, and on the morning of testing. Records were kept on whether the children received and consumed the breakfast, even partially.

In addition, as mentioned, it was deemed necessary to estimate whether the programme breakfast raised children's dietary consumption, hence providing support to the purported psychoeducational effects hypothesized by the study.

A subsample of 60 children by treatment group was selected to test the hypothesis that the school breakfast resulted in an increment of energy and protein consumption of at least 20% over the habitual intake. Twelve students, six per grade, were included in the subsample from each of the 10 participant schools.

A 24-hour-recall method was employed in the field, following standard guidelines [16], which included a procedure, developed at the IIN, to estimate the amount of food consumed. The target children answered questions regarding the school breakfast and snacks, and their mothers responded for the remaining meals. Data collection was scheduled during time 0 and time 2.

The amounts of nutrients consumed, particularly energy (kilocalories), protein (grams), and iron (milligrams), were calculated based on a table of food composition generated by the IIN. Inter-group nutrient consumption comparisons (t test) were performed at the two evaluation periods. In order to estimate dietary consumption before testing, these periods were arranged and compared according to the following daily schedule: 6.00 through 7.30 a.m., reflecting breakfast consumed at home; 7.31 through 11.00 a.m., consisting of snacks and the school breakfast at time 2, consumed before the psychoeducational testing; and 12.00 noon through 7.00 p.m., including all remaining foods eaten during the day.

Logistic and financial constraints prevented increasing the sample size to conform to the study's design, the school as a unit of analysis.


Teachers recorded each student's presence or absence on a form designed for this purpose, which they completed while taking roll. Information was collected during three time periods, allowing for inter-group comparison before the inception of the school breakfast programme (To), during the experimental period (T2), and during the universal implementation period (T3). The average number of school days in each of these terms, discounting days with no classes, was 16, 12, and 16, respectively. Rates were calculated as the ratio of days attended/ school days x 100.

Psychoeducational evaluation

The battery consisted of three tests of particular cognitive processes and three tests of complex mental abilities, which assess functions and abilities required of fourth- and fifth-graders. Those related to specific cognitive processes were digit discrimination and the coding and digit span tests of the Wechsler intelligence scale for children [17]. The digit discrimination test was devised for the present study. The task was to circle as many number 6's as possible in four minutes on a sheet of paper that included 1,175 digits listed in random order and arranged in lines of increasing length. The outcome is the ratio of the number of correct responses to the total number of digits reached. The coding test reveals visual perceptual organization and visual motor coordination, and the digit span test is a measure of working memory.

Tests of reading comprehension, vocabulary, and mathematics were used to evaluate complex mental abilities. The first two are part of the inter-American series designed to assess the reading abilities of Spanish-speaking children in Texas [18] that was used successfully in studies on nutrition and cognitive development [19]. The mathematics test was devised for this study based on the curriculum for public schools in Peru. It is worth noting that because testing was conducted in a group classroom setting, the tests for specific cognitive processes had to be adapted accordingly.

All test-retest correlation coefficients (four-week intervals) for the control group were greater than 0.50 (p <.001). The highest correlations were for vocabulary (r = .83) and mathematics (r= .78). The coefficients for number discrimination and digit span tests were too low (r = .56 and .50, respectively) to be considered sensitive to the treatment; therefore, they were dropped from all further analysis.

A second concern was the degree of convergence among the tests. Although indexes of performance in the six tests should be positively correlated, the magnitude of the covariation should be relatively low, except for tests that tap similar constructs. The size of the r2 was expected, and most coefficients were smaller than .30. An exception was the correlation between reading comprehension and vocabulary (r=.59), which was expected because both reflect verbal abilities.

Effect modifiers

A proxy variable of socio-economic status was constructed based on data recorded on the quality of the household, maternal education, and number of persons contributing to the family income in forms others than labour. These three items were combined into a single score following a strategy that was used successfully by others [19].

Household characteristics included type of roof, type of walls, number of rooms, window and door materials, water and waste disposal, electricity, and the presence of small commerce on the premises. The mother's education level was defined as the total number of years of formal education.

In addition, data were collected on the individual characteristics of the child, such as age, sex, and principal language spoken at home, and his or her educational attainment level. Also in this category were the child's weight and height, which were converted into the conventional indexes of nutrition status: weight-for-height, indicating current nutrition status or wasting, and heightfor-age, reflecting nutritional stunting. Both of these indexes were standardized by means of Z scores using the WHO-National Center for Health Statistics (NCHS) reference values.

As 26% of our sample exceeded the upper limit of the WHO-NCHS reference tables for the index weight-for-height, we decided to create a similar measure for wasting to take advantage of the information already collected. Thus, we regressed weight on age and height, obtaining a residual value that behaved similarly to the conventional wasting index (r= .89; p < .001). Hereafter this measure is refered to as weight-R.

The three school-related variables were the school entered as a nominal variable, the child's age at entering school, and current enrollment in the fourth or fifth primary grade.

Data analysis

The comparability of both groups regarding the confounder variables was first assessed. Next, we ascertained the experimental effects at the school level, by applying the least-squares method to the outcome variables, adjusted for their baseline value. The statistical test of choice was a t test. This and the leastsquares calculation were performed using the general linear model procedure (Proc GLM) available in the SAS software package.

Subsequently, to examine the treatment effects at the individual level, we used analysis of covariance (ANCOVA) by means of the Proc GLM in SAS. The outcome variables were the post-treatment test scores after partialing out their covariance with their respective pretreatment score. The predictor variables included in the model were all mentioned in the preceding section. In addition, a treatment variable (1 = Rx, 0 = control) with the school variable nested within each group was included.

Finally, the model included two interaction terms to test whether the effect of the school breakfast programme was a function of the children's attained stature (height-for-age x treatment) or current nutrition status (weight-R x treatment).

TABLE 3. Characteristics of the studied schools

Group/school Primary students in grades 1-6a Study sample in grades 4-5 Student/teacher ratio Distance school to Huaraz (km) Mean (SD) time school to home (min)
Experimental 850 233 22.3c 7.6b 23 (25)b
Atipayan 258 71 23.4 7 38 (36)
Uccru 74 11 18.54 11 21 (18)
Marian 182 49 26.0 8 23 (20)
Paria 222 68 24.6 8 16 (12)
Vichay 114 34 19.0 4 10 (10)
Control 842 169 23.2 10.0 23 (26)
Huamarin 220 41 24.4 13 38 (35)
Shansha 105 20 26.3 9 35 (25)
Palmira 200 55 25.0 4 16 (21)
Paltay 227 35 25.2 18 12 (18)
Quenuayoc 90 18 15.0 6 17 (12)

a Students registered according to school records.
b Group average.


Characteristics and comparability of selected schools

As shown in table 3, each treatment group had marked between-school variability in some of the school characteristics, particularly among those in the experimental group. For instance, the number of students enrolled in the school ranged from 74 in Uccru to 258 in Atipayan. Similarly, in Uccru only 11 children were enrolled in the study, whereas 71 were enrolled in Atipayan. The student/teacher ratio in the fourth and fifth grades ranged from 15 in Quenayoc to 26.3 in Shansha.

The distance from the school to the city of Huaraz is a rough indicator of the remoteness of the communities where the schools are located. However, in some instances this information is misleading. Some communities are relatively far from Huaraz but have access to rapid public transportation, whereas others are closer to the city but have little if any access to transportation. The range in the experimental group was 4 to 11 km, and that in the control group was 4 to 18 km.

The distance between school and home was also measured by the time it took to walk from one to the other. This time reflects the effort invested by the children to acquire an education. The range was from 10 to 38 minutes in the experimental group and from 12 to 38 minutes in the control group.

The success of randomization, as determined by intergroup (experimental-control) equivalence in the values of the main confounders, was satisfactory for age, gender, anthropometry, and social and economic background variables. However, there were differences in the language spoken at home, the age at which the child entered school, and the distribution of children in the fourth and fifth grades. The children in the experimental group were more likely to speak Quechua at home, to enter school earlier, and to be enrolled in the fifth grade than those in the control group (table 4).

TABLE 4. Selected characteristics of study subjects by study group [mean (SD)

  Treatment group Control group
Variable (n = 201) (n = 151)
Age (mo) 136.2 (18) 138.9 (20)
Male, no. (%) 90 (46) 80 (53)
Weight-for-height (Z score)a 0.51 (0.7) 0.43 (0.9)
Weight residualb


0.007 (2 9)
Height-for-age (Z score)


-2.20 (1.0)
Mother's years of schooling 2.25 (2 9) 2.7 (3 2)
Home characteristics (score) 15.28 (6.9) 15.72 (7.2)
Socio-economic status (score) 19.6 (8) 20.3 (9)
Quechua spoken at home, no. (%) 86 (43) 37 (24)c
Age of entry to school, no. (%)    
<=7 yr 190 (95) 127 (84)c
>7 yr 11(5) 24 (16)
Times child repeated a grade 1.10 (1 34) 1.23 (1.30)
Fourth-graders, no. (%) 81 (40) 77 (51)d

a. Includes data for 74% of the sample size.
b. Residual obtained after regressing weight on height and age.
c. p <.001.
d. p <.05.

Click here to continue

Contents - Previous - Next