Contents - Previous - Next

This is the old United Nations University website. Visit the new site at http://unu.edu

Lessons from the fluid-milk programme

Important lessons can be learned from the evaluation of the FMP and can shed some light on the feasibility of implementation at the national level of a food-coupon programme aimed at low-income families. In Brazil, programmes of this type have been the object of lengthy, and at times very heated, discussion. Therefore it would be useful to have recourse to the data of actual research that might provide relevant information for this debate.

Four important aspects of the programme were the suitability of milk for a subsidy, targeting, programme efficiency, and effectiveness, not only in nutrition terms but also in terms of the capacity to transfer funds that correspond to a significant portion of the budget of low-income households.

Product suitability

The suitability of milk for a subsidy programme is discussed on the basis of the results obtained by Gray [17]. The author presents an interesting comparison of the effect of a Cz$1 billion subsidy on the daily calorie intake of four consumer groups: the bottom 15% and 30% and the top 30% in intake distribution, and the top 20% in income distribution. She also calculated the budgetary gain of the families receiving the benefits of a subsidy of this size. The results refer to mutually exclusive alternatives: Cz$1 billion spent on wheat bread, rice, or milk.

The milk subsidy would yield less of an increase in caloric intake than that obtained from the distribution of rice but more than that obtained from the distribution of wheat bread for the lowest 15% and 30% in the caloric-intake distribution. Therefore, from the standpoint of nutritional value for these groups, milk would be more suitable for subsidy than wheat bread but less suitable than rice. The highest 30% in the caloric distribution and the highest 20% in the income distribution would benefit more from a milk subsidy than from a subsidy on either rice or wheat bread.

The impact of the milk subsidy on the family budget is smaller than that of an equivalent subsidy on wheat bread, rice, or manioc prices for the groups that consume fewer calories (the bottom 15% and 30% in the caloric-intake distribution); the subsidy on milk has a greater impact than the other commodities referred to in the highest consumption (top 30%) and income (top 20%) groups.

These results led Gray to comment that "to achieve two joint objectives, i.e., improve nutrition and transfer income, may not always require the same political prescription.... However, the benefits of a milk subsidy in terms of income are asymmetrical and much more favorable to the richer segment of the population... ," thus making it "a relatively weak instrument for the transfer of income to the poor, although its nutritional effect may be greater than an equally significant subsidy of wheat bread" [7, pp. 42-43].

This observation and the data presented by Gray indicate that milk may not be a good product to choose for a subsidy if what is desired is a nutrition programme that has redistributive characteristics. However, from a nutrition point of view, milk is more suitable for a subsidy than wheat bread, although it may be less suitable in terms of both nutrition and income than rice, for example.

These qualifications stem from the fact that the parameter used by Gray was the variation in caloric intake, a criterion that makes sense since the nutrition deficiency in Brazil is essentially caloric. However, with respect to children under 5 years of age, the protein requirement is very important (especially for those between 6 months and 1 year of age); in this case milk has no substitute from the nutritional point of view.

Thus milk is less suitable than rice or manioc flour if the objective is a general price subsidy that benefits the poor population, or even if the objective is to reach the biologically vulnerable group over 5 years of age. However, milk would seem to be the most suitable product for a programme targeted at children under 5 years of age who can be identified as being undernourished by screening procedures administered by health-centre personnel.

Evaluation of the targeting process

In a programme of this nature, two types of targeting errors may occur, similar to what are termed in statistics type-1 and type-2 errors [5]. A type-1 error would consist of giving aid to those who are not members of the target group, and a type-2 error would consist of denying aid to those who are members of the target group.

In the case of the FMP, it is impossible to quantify the proportion of type-2 errors, as no census was carried out on the population under 5 years of age to discover how many of those below the tenth percentile of the Santo André weight standard were not receiving aid at the health centres.

The data obtained do allow quantification of the type-l error. This was carried out by checking the number of children above the tenth and fifteenth percentiles who received food aid. The tenth percentile was chosen because it was the cut-off point used by the programme to distinguish children at risk of undernutrition from healthy children. As a second measure, the fifteenth percentile covered children who were on the borderline and were included for humanitarian reasons, or those who gave evidence in three successive measurements at the health centre of having consistently lost weight.

TABLE 3. Percentages of children receiving aid who should not be members of the target group (type-1 error)

a. Weight-for-age percentile.

The results presented in table 3 indicate that 30% of the children receiving milk should not be receiving it. Using the fifteenth percentile as a cut-off point, 20% who were receiving milk were not legitimate members of the target population. This is the best indicator we have of a type-l error; nevertheless, this rate may be overestimated, since one of the reasons for inclusion in the programme was that the child had been losing weight in the last three months. in which case he or she could be above the fifteenth percentile and not qualify as a type-1 error.

The age distribution shows that, in the case of boys only, the margin of error fell as the age increased for the successive ranges up to 24-36 months; the 36-48month range breaks this tendency, which reappears in the final range (48-60 months).

Programme efficiency

It was not possible to obtain cost-efficiency measures, since no data are available. Owing to this limitation, what is meant by efficiency here is programme management efficiency in terms of accessibility of the milk-distribution points, transportation costs incurred by the beneficiaries, and complaints regarding management.

The distance that the beneficiary mothers (or family members) had to travel to pick up milk each day did not represent a significant problem. In Santa Isabel, where milk had to be picked up at a central location run by the city government, the average number of blocks travelled was 14.7, with a mode of 10; 25% of the mothers had to come a maximum of 9 blocks, and 49% a maximum of 10. In Sorocaba, where mothers could pick up the milk at a local bakery or grocery store, the average number of blocks covered was considerably less-9, mode 2. (As there is no standard size for a block in Brazil, it is impossible to make any precise comparison of the distances covered in the two cases.)

Mothers of programme beneficiaries voiced no complaints as to the cost of transportation, which could reduce the attractiveness of the milk and thus limit participation. The majority of mothers were satisfied with the way the programme was handled, and no significant suggestions were made for changes in its management except that the amount of milk distributed should be increased (requested by 30% of the mothers).

TABLE 4. Analysis of children's nutritional status after approximately one year of participation in the programme

The data refer to the children's percentile position on the weight-for-age standard at the time of enrolment in the programme (t) and at the fourth weighing (t + 4). As the mother was supposed to bring the child to the health centre every three months. the fourth weighing corresponds roughly to one year after enrolment. although not necessarily, because some mothers delayed their visit to the health centre.

Programme effectiveness

The programme was evaluated in terms of its economic significance and its impact on nutrition status.

Impact on nutrition status

A clear evaluation of the programme's impact on nutrition was not possible, since we could not specify a control group. As a proxy, the proportion of children below the tenth percentile of weight for age at enrolment was compared with the proportion of children in this situation at the fourth weighing (table 4). The results are presented for the children sampled in Santa Isabel and Sorocaba, and also for an additional sample of 132 undernourished children for whom anthropometric data were gathered in Santa Isabel (referred to as "additional sample").

If the programme were effective, we would expect that the proportion of children below the tenth percentile after one year would be smaller than that at the time of enrolment. This hypothesis was accepted at the 5% level of significance for the two samples collected for children 12-60 months old. For children below 1 year of age the hypothesis is accepted at the same level of significance as for the original sample and for the total sample, but not for the additional group enrolled in the Santa Isabel health centre.

That result led us to investigate the importance of the variable weight at birth. Table 5 shows that the proportion of children below the tenth percentile who had birth weights equal to or greater than 3 kg is smaller than that of those above the tenth percentile for both age-groups; these differences are not statistically significant at the 5% level. Also not statistically significant are the differences in the proportion of children who were below and above the tenth percentile and who were born with low or insufficient weight. The chi-square tests conducted do not indicate any relationship (at the 10% level of significance) between weight at birth and nutrition status on entrance into the programme.

To investigate further the association between weight at birth and weight at enrolment in the programme, we also calculated the coefficient of uncertainty. The low values obtained for this coefficient indicate that the knowledge of weight at birth does not give us a priori knowledge of whether a child will or will not be below the tenth percentile in weight for age later on. (11)

TABLE 5. Relationship between birth weight and weight for age at enrolment in the programme

a. Weight-for-age percentile.

In conclusion, the FMP had a positive impact on nutrition status, as the proportion of children below the tenth percentile in the weight distribution scale was lower after one year than at enrolment in the programme. However, the statistical tests specifically showed that over time there was a significant change in the relative percentile occupied by boys between the ages of 1 and 2. This result is important because it is in this particular age-group that undernutrition is most severe in Brazil.

A final indirect measure of the effectiveness of the programme was obtained by asking the mothers' opinions about improvements in the nutrition status of their children. From their point of view the programme seems to have been successful, since only 18.3% of the 218 mothers who answered this question had noted no improvement in their children. The remaining 82% answered as follows: child improved (51%); child grew and developed (9%); and child's health improved (22%).

Economic impact

The economic significance of the FMP is presented in tables 6 and 7. Table 6 shows tabulations carried out for the purpose of checking the relative contributions to caloric and protein availability at the household level of the FMP and the PSA, and of the food bought directly by the family; it also shows the relative importance of these different sources of obtaining food to total food expenditure. The food-distribution programmes contributed 30% of the calories available to the families, and the remaining 70% was purchased. The FMP alone was responsible for 10%. Because milk is richer in protein than in calories, the FMP was responsible for 21% of protein intake, the PSA for 17%, and family purchases for 62%.

TABLE 6. Comparison of food received through the FMP and PSA and that purchased directly by the beneficiary families (percentages)

Tables 6 and 7 are based on information collected by surveying the mothers. As the quantity of food donated by the PSA and FMP became known it was added to obtain the total quantity or food available at the household level. The transformation of quantities into calories and grams of protein was done using food composition tables prepared by the School of Public Health of the university of São Paulo.

a. An imputed value was estimated for food received through the FMP and PSA.

TABLE 7. Comparison of calorie and protein costs from food supplied by the FMP and PSA and that purchased directly by the consumer

a. Actual cost at retail outlets.

In monetary terms the FMP was highly significant: it supplied 16% of the total value of the food used by the family, both purchased and donated. The PSA supplied nearly 7.7%, and the families directly purchased the equivalent of 76% of the total value of the food they used.

Cost-efficiency measurements

Two interesting cost-efficiency measurements are the unit cost per calorie and the cost per gram of protein according to the source of provision. The per unit cost of calories and protein when food is provided by the FMP or PSA is compared with the cost when the food is purchased directly by the consumer. This comparison allows one to infer the relative efficiency of the programme.

The data in table 7 indicate that the consumer bought an average of 1,000 calories at a cost of US$0.33, whereas the same 1,000 calories cost US$0.42 in the FMP and US$0.10 in the PSA. From this point of view, the PSA was the most efficient per calorie unit, its cost corresponding to 30% of the average cost of calories when purchased directly. In the FMP, this cost was 126% of the cost of calories purchased by the consumer, since milk is an expensive source of calories in the market-place.

The consumer purchased 1,000 g protein at an average cost of US$1.38, whereas the same 1,000 g protein cost US$0.75 in the FMP and US$0.46 in the PSA. Thus, using the cost per gram of protein as an indicator, the two programmes were both more efficient than direct consumer purchase, with the average cost of protein for the FMP equal to 54%, and for the PSA to 33%, of the cost when purchased directly. Once again, in comparison, the PSA was the most efficient in terms of the basket of goods selected.

It must be noted that the use of the "cost" here is not the same as that normally applied in the cost analysis of food and nutrition programmes. In this latter type of analysis, one calculates the basic cost of the food and adds on the cost of transportation, storage, packaging, administration, and leakages. In the exercise presented in table 7, we considered only the market value of the donated food, divided by its calorie or protein content. Actually, we are comparing the efficiency of the programmes relative to the consumer's ability to select a food basket with the lowest calorie and protein cost in the market-place. Therefore, the data presented in table 7 indicate that the PSA is more capable than the consumers of choosing a basket with a lower calorie unit cost evaluated at market prices, and the FMP is less capable, since milk is an expensive source of calories. In terms of protein, both programmes were more capable than the consumer. (The concept of cost used in this discussion precludes the presentation of international comparisons.)

Factors that distinguish undernourished from well-nourished children

To identify the variables that can isolate undernourished children from well-nourished children, a discriminant analysis was conducted on 158 children for whom a complete set of data was available (126 undernourished, 32 well-nourished). This statistical technique permits linear combinations of variables that can discriminate between the groups.

The variables that distinguish undernourished children from healthy children are, in order of importance, weight at birth, expenditure on water and electricity, expenditure on cigarettes, food expenditures, total number of family members, per capita caloric availability (per day), total family income, expenditure on products for cleaning and hygiene, and availability of animal protein per capita per day.

Some of these results were expected a priori. It is reasonable to expect that weight at birth would be an important discriminant variable, as well as total income, food expenditure, and calorie and protein availability per capita. The total number of family members was also expected to be an important variable, since the larger the family, the smaller the probability of finding well-nourished children.

The importance of expenditure on water and electricity can be attributed to the fact that in Brazil this variable serves as a proxy for the availability of durable consumer goods and is, in this sense, a proxy for wealth.

The importance of expenditure on products for cleaning and hygiene can be attributed to the fact that this reflects family attitudes toward hygiene.

It is difficult, however, to explain the importance of expenditure on cigarettes and the low rank of expenditure on education. However, one has to consider that, in Brazil, poor people usually go to public schools, so expenditure on education does not reflect the actual use of educational services by the families considered.

The results of the discriminant analysis were very good. Goodness of fit can be assessed by comparing the category in which each child in the sample fits with his or her clinical measurements. Using the results of the discriminant analysis, each child can be categorized as healthy or undernourished, according to the values of individual or family variables. This category is then compared with the child's effective classification and a rate of correct results is computed. This goodness-of-fit test indicated that for undernourished children the proportion of correct results was 91.8%, whereas for healthy children it was only 62.8%. (12)

Conclusion

On the basis of the experience of an actual programme, the FMP in the state of São Paulo, we draw the following conclusions with respect to whether a food-coupon programme would be feasible in Brazil. First, a targeted food-coupon programme is feasible and can be undertaken at the national level. To be well targeted, it must be integrated with health actions, as was the case with the FMP. The FMP was run efficiently and the population had no major complaints about its management. Problems expected a priori, such as distance from a bread-milk store or difficulty in finding milk at the grocery store, were not cited by a significant percentage of the beneficiaries. In addition, the programme had a significant impact on the budget allocated for food by the beneficiary families. Although it was not possible to specify a control group, anthropometric data showed a reduction in the proportion of children below the tenth percentile after one year of participation. There was a statistically significant change over time in the relative percentile occupied by boys between 1 and 2 years of age, the age-group in which undernutrition is most severe in Brazil. Finally, discriminant analysis indicated that weight at birth was the most important variable for distinguishing undernourished children from well-nourished children.

These results are relevant for other developing countries, especially those in Latin America, which face similar constraints with respect to the management of food programmes. The difficulty faced by health centres in Brazil in storing food and in imposing on their strained personnel a series of tasks related to food handling is typical of the reality of the health sector in Latin American countries. The solution found by the state Secretary of Health in São Paulo-a food coupon distributed at the health centre and traceable at any small vendor-has proved to be feasible and is compatible with the realities and the resources of other Latin American countries.

Notes

1. According to Batista Filho. Sigulem, and Nobriga [1], there are more than 20 definitions of malnutrition and undernutrition. These terms are used here with the meanings given by Aylward and Jul [2, p 13]: Undernutrition is a generic term given to caloric and/or protein deficiency, which may or may not be accompanied by deficiency of other nutrients. Malnutrition refers specifically to the effects of deficiencies in one or more nutrients or to a lack of balance among them.

2. Although the amounts devoted to the food and nutrition programmes are significant. they were only 27.6% as much as was spent on the subsidies for wheat and sugar, which has been estimated at US$3.6 billion [5]. The wheat subsidy was given at the producer and consumer levels. but the sugar price subsidy benefits only the producers in the north-east. There is no clear agreement in the literature about the equity effects of these subsidies, but it seems from the evidence gathered that, if part of this money were devoted to well-targeted programmes, the nutritional impact would be far greater than that obtained with the existing general price-subsidy programmes.

3. The cost of powdered milk constituted up to 65% of the current costs of the health centres, excluding labour (data provided by Dr. João Yunes, former Secretary, State Secretariat of Health, State of São Paulo).

4. It is important to note that these changes were made by the first democratically elected state government since 1965.

5. The Santo André study, carried out by a team of Brazilian researchers [13], covered 12,000 children from Santo Andre, an industrial county on the outskirts of São Paulo and part of the São Paulo metropolitan area. It sought to establish the Brazilian distribution pattern for weight and height according to socio-economic group. The growth pattern found by the study for the high socio-economic group in Santo Andre was quite similar to that found by the 1938 Harvard study. For further details see Batista Filho et al. [1].

6. The actual coverage could be even greater, since-on the evidence of the 1989 National Health and Nutrition Survey, which found a 30.7% incidence of undernutrition as opposed to 46.1% found by the 1975 National Household Expenditure Survey [15]-the frequency of undernutrition in the Brazilian population declined between the time of the IMPEP/IPE study in 1974 and the present study in 1987.

7. In fact, a third sample was also taken for the purpose of complementing the anthropometric data on undernourished children registered at the Santa Isabel Health Centre. Only anthropometric data were collected for 132 children; when added to the 47 cases comprising the first sample, these account for the total of 179 cases of undernutrition registered in Santa Isabel. The results of this sample are not reported here; the interested reader may refer to the chapter on sampling in Campino [16].

8. The monetary unit used by Gray is the cruzeiro of 1975, which had an exchange rate of US$1 = Cz$7.62 in Feb. 1975, the mid-point of the ENDEF study, conducted between Sept. 1974 and Aug. 1975. This unit is different from the cruzeiro used now. Since 1975 Brazil has experienced four different monetary units: the cruzeiro, the cruzado (1986), the cruzado novo (Jan. 1989), and the cruzeiro again since 1990.

9. In the supplementary sample of 132 children in the county of Santa Isabel, the situation in terms of fulfilment of the programme's objectives was much better: 18% of the boys and 29% of the girls were over the tenth percentile, and only 13% and 10% respectively were over the fifteenth percentile.

10. As one litre of milk per day is adequate for children 0-4 years old, who were the target population, this request indicates that the mothers wanted milk for other family members.

11. The coefficient of uncertainty (CU) is the proportion by which the uncertainty of the dependent variable is reduced by knowledge of the independent variable. It varies between 0, when knowledge of the classification of an individual in the independent variable does not reduce the uncertainty as to the individual's classification in the dependent variable, and 1, when there is complete elimination of uncertainty. Using weight at birth as the independent variable, we obtained CU values of 0.017 for children in the 0-11-month age bracket and 0.020 for children in the 12-60-month age bracket.

12. The smaller proportion of correct categorization of healthy children can be explained by the fact that it was possible to obtain a complete set of data for only 32 healthy children as compared with 196 undernourished children.

Contents - Previous - Next