This is the old United Nations University website. Visit the new site at http://unu.edu
Descriptive analyses
A total of 1,436 children met the criteria for inclusion in these analyses; most were included in more than one age interval (table 2).
TABLE 2. Sample sizes by village, age interval, and participation status
Age (mos)  Atole villages  Fresco villages  
Conacaste  San Juan  Santo Domingo  Espíritu Santo  
P  NP  P  NP  P  NP  P  NP  
03  154  58  106  57  131  66  82  62 
36  165  54  112  53  151  52  113  34 
69  173  58  112  56  155  59  119  34 
912  186  54  115  59  165  60  122  33 
1215  194  59  117  68  174  63  125  35 
1518  196  66  128  63  180  66  131  36 
1821  201  72  133  65  185  72  132  42 
2124  203  74  141  63  200  65  137  43 
2430  222  69  153  58  210  62  148  36 
3036  235  74  162  66  228  63  159  38 
3642  260  69  178  64  243  64  167  41 
4348  262  72  177  59  233  62  157  39 
4860  233  63  166  49  224  44  141  33 
6072  222  55  148  47  209  34  137  26 
7284  208  56  130  36  206  35  131  15 
P = participants; NP = nonparticipants.
Participation rates were between 65% and 75% during the first two years of life, increasing to 75% to 85% by year five (see FIG. 1. Participation of eligible children in the supplement programme, by age and village). Comparing the four villages, except for significantly lower rates for San Juan at 912 and 1215 months (ANOVA, p < .03, p < .005 respectively), differences in participation were not statistically significant. Aggregating the villages over supplement type, those receiving atole had significantly lower rates of participation (about 78%) for children 57 years old than those receiving fresco (about 86%) (t test, p < .01).
Of children who participated at least once during the measurement interval, those in the atole villages attended the centre a significantly higher percentage of days than those in the fresco villages until the age of 3.5 years (see FIG. 2. Attendance of participants at the feeding centre (percentage of days in interval), by age and village). Participating atole children attended the centre 50%65% of the days in the interval at all ages. Fresco children attended only 10%30% of the days during the first year of life, steadily increasing their attendance until age 4, when they too were attending about 60% of days. Differences between the supplement types are statistically significant at all measurement intervals until 42 months of age; there are in general no significant differences between villages of the same supplement type. When this analysis was done including nonparticipants, the interpretation of the results was unchanged.
The mean quantity of supplement consumed mirrored attendance patterns during the first three years of life (see FIG. 3. Mean volume of supplement consumed per day, by age and village (nonparticipants included)). In fact, differences between the two supplements were magnified because children consumed very small amounts of fresco during these early years. Consumption of atole increased rapidly between the first and second years of life, from a mean of 70 ml per day to about 125 ml per day. From 4 to 7 years of age, however, intakes of atole steadily decreased to about 100 ml per day. Intakes of fresco rose steadily as the children aged, peaking in Espíritu Santo at a high of 292 ml per day (ANOVA, p < .01).
The energy derived from the supplements is presented aggregated over supplement type in figure 4 (see FIG. 4. Mean energy intake per day from supplement, by age and supplement type (with and without nonparticipants)); values are presented both including and excluding nonparticipants. Because the caloric density of the atole was nearly three times that of the fresco, the greater volumes consumed during the younger ages result in even greater differences in the energy derived from the product. At age 1, for example, participating children were consuming about 80 kcal per day from atole but only about 6 kcal per day from fresco. The energy from atole peaked in the second year of life at 157 kcal per day. Children consuming fresco derived increasing amounts of energy from the supplement as they aged; by age 7, they reached a mean of 82 kcal per day. Differences in caloric in takes between the two groups are statistically significant at all age intervals.
Supplement values for energy are also presented based on the entire sample, including nonparticipants. As seen in figure 4, including nonparticipants affects atole more than fresco values; for example, at age 48 months the mean values are 47 kcal per day lower for the atole group and 12 kcal per day lower for the fresco group. It should be noted, however, that as a percentage of total energy from supplement, these values represent approximately a 20% decrease in both cases.
Changes in attendance and intakes over the duration of the study are presented in figures 5 (See FIG. 5. Attendance of children 13 years old (nonparticipants included), by year of study and village) and 6 (see FIG. 6. Mean energy intake per day from supplement, children 13 years old (nonparticipants included), by year of study and village). For children 13 years old, attendance was erratic during the first half of the study, but three of study villages showed significantly increased rates of attendance after 1973 For the atole villages, increasing amounts of energy were obtained from the supplement as the study progressed, due to higher attendance. In the fresco villages, however, there appears to be little change in the amounts ingested by year, even though attendance rates increased. This suggests that children were consuming smaller amounts of fresco at each visit, a result also identified in the multivariate models.
Multivariate models
Results of six separate analyses, by supplement type and age, identify predictors of attendance at the supplementation centre (table 3). For atole children, being closer to the centre, having more people in the family, and lower SES were all significantly associated with higher attendance for each of the three age categories. The magnitude of the estimates suggests that the influence of the six covariates on attendance did not vary much as children aged. The exception was the decreasing importance of family size with age.
TABLE 3. Factors for predicting attendance at the supplementation centre (percentage of days during age interval) using Tobit analysis—separate models by age category and supplement type
Variable  Atole  Fresco  
Birth1 yr  13 yrs  37 yrs  Birth1 yr  13 yrs  37 yrs  
Number^{a}  289  333  503  255  329  424 
Leftcensored^{b}  39  33  41  27  25  14 
Intercept  28.6 (20.4)  46.2 (12.8)***  57.2 (7.7)***  1.2 (11.8)  16.9 (11.5)  59.2 (8.1)*** 
Distance to centre  14.3 (3.0)***  17.6 (2.3)***  17.7 (1.7)***  5.9 (1.6)***  11.9 (1.9)***  14.3 (1.7)*** 
Family size  4.3 (1.1)***  3.3 (0.9)***  2.3 (0.7)***  2.3 (0.6)***  3.9 (0.7)***  3.0 (0.6)*** 
SES  7.0 (2.1)***  8.3 (1.7)**  6.3 (1.3)***  1.1 (2.4)  2.5 (1.8)  0.8 (1.4) 
Sex  5.4 (4.0)  3.7 (3 2)  5.1 (2.5)*  1.9 (2.4)  2.2 (2 9)  1.0 (2.5) 
Village  4.8 (4.3)  2.7 (3.2)  4.7 (2.6)  5.6 (2.7)*  4.3 (3 2)  8.4 (2.7)** 
Birth year  2.5 (1.7)  2.1(0.9)*  0.3 (0.5)  2.0 (0.9)*  2.1(0.8)**  0.4 (0.5) 
Log likelihood  1274.5  1679.0  2234.6  1017.0  1446.7  1917.5 
Only children who were able to attend for at least 12 consecutive months in an age category are included. Figures in parentheses are parameter estimates (standard error).
a. Sample size in model.
b. Number of leftcensored observations (i.e., children who did not attend).
*p < .05. **p < .01. ***p < .001.
For fresco children, besides the expected differences in the magnitude of the estimates, two main results stand out. First, SES did not affect attendance rates. Second, the importance of some of the covariates differed by age; for example, children in Espíritu Santo were less likely to attend at younger ages but more likely to attend at older ages.
The varying influence of SES on attendance by supplement type was confirmed through additional multiple variable analyses using a pooled sample of atole and fresco children. In these analyses, supplement type was represented by an indicator variable, and an interaction term between supplement type and SES (dichotomized at the median) was included. This interaction term was significant at p<.01 for each of the three age categories. The only other significant interaction identified was between supplement type and distance from the centre, with attendance being more influenced by distance for atole than for fresco children; this interaction was only significant for the 13year (p<.02) and 37year (p < .08) age categories.
Comparing the above results to the twostage alternative approach discussed earlier, the parameter estimates and levels of significance obtained in the second stage of the approach (i.e., ordinary least squares after eliminating nonattenders) are nearly identical to those presented in table 3. While the direction of the parameter estimates were the same for the first stage as well (in which the predictors of attendance versus nonattendance were modelled using logistic regression), estimates were often not statistically significant due to the relatively small percentage of nonattenders and, therefore, low power.
Factors predicting energy intakes from supplement, controlling for attendance, are presented in table 4. In this case, children living farther from the centre consumed more supplement once they succeeded in arriving. The SES is a negative predictor of energy consumed from atole at each visit for children 13 years old. Children in San Juan consumed more atole at each visit than those in Conacaste.
TABLE 4. Factors for predicting (the square root of) caloric intake from supplement (kcal per day) after controlling for attendance at the centre—separate models by age category and supplement type
Variable  Atole  Fresco  
Birth1 yr (N = 255) 
13 yrs (N = 409) 
37 yrs (N = 603) 
Birth1 yr (N = 334) 
13 yrs (N = 390) 
37 yrs (N = 424) 

Intercept  0.167 (1.30)  3.160 (0.88)***  1.831 (0.74)***  1.276 (0.37)***  4.434 (0.46)***  3.488 (0.36)*** 
Attendance  0.133 (0.00)***  0.151 (0.00)***  0.164 (0.00)***  0.046 (0.00)***  0.076 (0.00)***  0.103 (0.00)*** 
Distance  
to centre  0.306 (0.20)  0.383 (0.17)**  0.470 (0.17)***  0.019 (0.05)  0.079 (0.08)  0.208 (0.07)*** 
Family size  0.108 (0.07)  0.123 (0.06)**  0.048 (0.06)  0.076 (0.02)***  0.079 (0.49)  0.068 (0.03)*** 
SES  0.142 (0.14)  0.297 (0.11)*  0.172 (0.12)  0.054 (0.05)  0.032 (0.07)  0.056 (0.06) 
Sex  0.203 (0.26)  0.173 (0.21)  0.321 (0.23)  0.045 (0.07)  0.214 (0.11)  0.143 (0.10) 
Village  0.624 (0.27)  0.043 (0.22)  0.469 (0.24)*  0.213 (0.08)*  0.354 (0.12)***  0.723 (0.11)*** 
Birth year  0.062 (0.10)  0.103 (0.06)  0.203 (0.26)  0.183(0.03)***  0.354(0.03)***  0.245(0.02)*** 
Rsquare  0.80  0.80  0.74  0.71  0.81  0.88 
Only children who were able to attend for at least 12 consecutive months in an age category are included. N= sample size in model.
Figures in parentheses are parameter estimates (standard error).
* p < .05. ** p < .01. *** p < .001.
For children consuming fresco, family size had a positive effect on intakes at younger ages but a negative effect at older ages. Again, differences are seen between the two villages. Finally, that the estimate for year of birth was negative was most likely due to a large decrease for children in Espíritu Santo between 1971 and 1973; in fact, the figure of the descriptive statistics suggests consumption rose again in the second half of the study.
Each of the models was also run including nonparticipants, and not controlling for attendance. In the first case, although the intercept value was somewhat reduced, the magnitude and significance of the estimates were nearly identical. In the second case, leaving attendance out of the model gave results very similar to those obtained for the models presented in table 3. This is due to the extreme correlation between percentage of days attending and energy obtained from supplement. For the atole group, the Pearson correlation coefficients between these two variables at each of the three age categories are over .92 (p<.0001), and they are similarly high for the fresco group.