![[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}
|
|
|
|
|
||
Europe, Middle East and Americas Division, United Kingdom Department for International Development, London SW1E 5HE, UK;
* Food Consumption and Nutrition Division, International Food Policy Research Institute, Washington, DC 20006;
Rollins School of Public Health, Emory University, Atlanta, GA 30322; and
** Ministry of Health, Brasília, DF, 70750–543, Brazil
2To whom correspondence should be addressed. E-mail: s-morris{at}dfid.gov.uk.
 |
ABSTRACT |
|---|
 TOP ABSTRACT SUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
KEY WORDS: • cash transfer programs • childhood growth • Latin America • Brazil • quasi-experiment
Stunting, the retardation of linear growth, is caused by factors that include lack of access to nutritionally rich diets (1), inadequate infant feeding practices (2), and repeated illness (3). All of these factors are related to poverty, with the result that, within countries, stunting consistently affects children from poorer families more than those who are better off (4). Significant improvements in living conditions, such as food supplementation or adoption, trigger catch-up growth in very young children at high risk of stunting (5). Therefore, it seems reasonable to suppose that a direct transfer of money to very poor families should lead to an improvement in the growth of their children, at least at those ages at which children are particularly susceptible to changes in their household environment.
On the other hand, improved child-caring practices do not always accompany an increase in household resources. Because of this, interventions may need to focus on empowering caregivers to provide more appropriate foods delivered in appropriate ways and to protect their children from disease. Health service personnel have been identified as a key resource for ensuring that caregivers of young children are able to recognize feeding problems and respond appropriately (6). Unfortunately, poor families often do not have frequent contacts with health services. If the frequency of contacts between these poor families and health services could be increased, the likelihood of health-service personnel making a real contribution to the reduction of malnutrition would be enhanced.
In Latin America, a number of governments have sought to boost demand for preventive health services and at the same time lessen the resource constraints faced by the poorest families by making direct payments to poor families on the condition that they keep up-to-date with preventive health measures (7). Mexico’s programa de educación y salud (PROGRESA) and Nicaragua’s red de protección social have both markedly increased the utilization of peripheral health services (8,9). It is as yet unclear, however, whether these programs can reduce the numbers of children affected by stunting. In Mexico, children receiving both the cash transfer and a multimicronutrient supplement grew about 1 cm more than those receiving neither intervention, but it has not been possible to disentangle the effects of the 2 interventions (10).
In Brazil, a national health-related conditional cash transfer program, Bolsa Alimentação, aimed to reduce nutritional deficiencies and to ensure that high-risk households were effectively linked into the national health service. The present analysis took advantage of a natural experiment to compare the growth of children in Bolsa Alimentação beneficiary households with that of similar children accidentally prevented from receiving the benefit.
|
SUBJECTS AND METHODS |
|---|
|
TOPABSTRACTSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
800,000 pregnant and lactating women and 2,700,000 children from all the 5561 municipalities in the country. Within a year of its launch at the end of 2001, benefits were reaching over 1 million children. Beneficiary households were selected in a 2-stage process. In the first stage, the federal Ministry of Health allocated program funding to municipalities in proportion to the estimated fraction of infants (aged 0–2 y) suffering from malnutrition (11). In the second stage, participating municipalities identified beneficiary households. The number of eligible household members determined the size of the grant a household received. Only families with a reported per capita income of less than half of the national minimum wage could be enrolled.
Monthly transfers ranged from 15 to 45 Brazilian reais (US$6.25 to US$18.70) per household per month, depending on the number of eligible individuals. Transfers were credited to a magnetic card that could be used to withdraw cash at offices of a federally owned bank, or, in very isolated municipalities, with lottery agents. Initial enrollment was for a period of 6 mo but could be extended if the family complied with its responsibilities under the program and remained income eligible.
In this study, program beneficiaries were compared with matched individuals from households that were originally selected to receive the benefit but who subsequently were excluded due to 1 of 3 quasi-random, administrative errors. The first of these involved the accidental separation of 2 electronic data files containing household identifying information. These files were supposed to be transferred simultaneously from participating municipalities to a central data-processing unit in the federal capital, Brasília, and when this did not happen, entire batches of beneficiaries were lost. The second error occurred when the primary beneficiary for a household had a nonstandard character (such as é, ô, or ç) in his or her name. The original version of the central data-processing system could not recognize these characters, and the affected records were rejected. The third error occurred when personal details recorded on a Bolsa Alimentação registration form did not match information already held on the same family by another federal program, Bolsa Escola. Unlike the previous 2 errors, this situation was more likely to arise if the family was already registered in the system and may therefore be considered quasi-random conditional upon participation in Bolsa Escola.
Excluded households were found in 67 municipalities. For logistic reasons and to ensure that enough time had elapsed since the beginning of the program to detect an impact on child growth, three additional criteria were established for the selection of the municipalities to be included in the evaluation study. These were that study municipalities should have at least 40 excluded households; be located in the northeast of the country, where 60% of program beneficiaries were resident; and have been participating in the program for at least 6 mo. In April 2002, when the data collection team went to the field, there were just 4 municipalities that met these criteria. All 314 excluded households in these 4 municipalities were to be visited and assessed as part of this study. It is important to note that all 4 municipalities belonged to a group of 20 municipalities where the program was first piloted; municipal secretaries of health were in charge of the selection of beneficiaries in these locations, and many beneficiaries were passed directly from a previous federal program intended to benefit malnourished children, the Incentivo para o Combate de Carências Nutricionais.
Because in these municipalities there were many more actual beneficiaries (2493, in total) than there were intended beneficiaries living in excluded households (506), it was decided to compare the intended beneficiaries living in the excluded households with an individually matched sample of actual beneficiaries, rather than the other way round. The matching criteria used were as follows:
Beneficiaries registered during their pregnancy were included in this exercise because they were expected to have given birth by the time they were located in the follow-up survey. Data collected during the registration process on household income, number of household members, rental payments, and the value of water, electricity and gas consumption, were used to characterize each household’s socioeconomic level. These variables were reduced to a single factor by using principal components analysis (12), and—in a random order—optimal matches were found for each excluded person based on the sum of the (squared) difference in socioeconomic score and (squared) difference in age. This technique is referred to as "nearest neighbor matching" based on Euclidean distances (13). After each match, matched beneficiaries were removed from the pool so that they could not be matched again to other excluded persons. Limits (or "calipers") were placed on the differences in either variable that were considered acceptable.
To increase the power of the statistical tests, 2 beneficiaries were matched to each excluded person in the sample. The study was designed to have 90% power to detect as statistically significant a 0.25 Z-score difference in weight-for-age between Bolsa Alimentação beneficiary and excluded children. For the retrospective cohort analysis based on routine record data, it was estimated that the study had 90% power to detect a 15% increase in the expected 6-monthly weight gain of 2 groups of children: those aged 0–6 mo and those aged 6–36 mo.
Data. Two complementary sets of data were used to assess child growth. First, a specially trained anthropometry team that visited the study municipalities 6 mo after the launch of the program assessed the attained growth of all children under 7 in study households. Weight was measured to the nearest 100 g by using a Filizola electric scale (Filizola). For children less than 2 years of age, length was measured recumbent to the nearest 1 mm by using a locally made infantometer. For children aged 2 years and older, standing height was measured to the nearest 1 mm by using a Seca Leicester Height Measure (Seca Vogel and Halke GmbH and Co.). Before going to the field, the 8 anthropometrists were trained by using the methods described by Gibson (14). Raw weights and heights were converted to age- and gender-standardized measures by using the National Center for Health Statistics reference standard, as recommended by the World Health Organization (15).
While these measurements conducted by the study team may have reflected program impacts, they may also reflect imbalances already present in the 2 samples before the program was launched. Therefore, we sought to reconstruct each child’s growth trajectory over time, based on routinely collected data. To do this, we analyzed the weight measurements recorded on each child’s Ministry of Health growth monitoring card (height not being routinely recorded in Brazil). The weights were copied to the study questionnaires, starting with the most recent recorded weight and going back in time. Up to a maximum of 10 weights were copied, together with the child’s age at the time of each measurement. Because the frequency of weighings decreased very markedly after 36 mo of age, the analysis of these data were limited to children < 36 mo old at the time of the interview.
The study was approved by the Ethics Committee of the London School of Hygiene and Tropical Medicine and by the Comissão de Ética em Pesquisa in Brazil. Written agreement to take part in the study was obtained from the principal informant in each family.
Statistical analysis. To avoid biases associated with the selective take-up of program benefits, all analyses presented in this paper are by beneficiary status defined by the program’s provision of benefits, not by the actual take-up. This is equivalent to analysis by "intention to treat" in a randomized trial.
Socioeconomic and demographic characteristics of beneficiary and excluded households were compared by using t tests (continuous variables) and 
2 tests (categorical variables) (16). The cross-sectional anthropometry data were also analyzed by using t tests. Ordinary least-squares linear regression was used to adjust the results for Bolsa Escola beneficiary status and for household demographic composition.
The retrospective cohort analysis used a regression model, described in the following paragraphs, to determine whether children’s growth velocity accelerated at the time their families began to receive program benefits. We assumed that this higher growth velocity was maintained throughout the time that the family was exposed to the program. The outcome variable for this analysis was weight in kilograms. The principal exposure variable was the length of time elapsed between the date that the child’s family was first supposed to have received a Bolsa Alimentação payment and the date of the weight assessment. This variable was set to zero for all children whose program status was "excluded" and for observations of beneficiary children that predated their families’ first receipt of the cash transfer (we checked carefully to verify that age-specific weight gain did not differ between these 2 sets of unexposed children). In those cases where the child’s family had first received a Bolsa Alimentação payment before the index child was born, because another child or pregnant woman in the family was enrolled as a beneficiary, this variable was set as equal to the child’s age. A second Bolsa Alimentação related variable was a dummy variable taking the value 1 for children from beneficiary families and 0 for children from excluded households. In the presence of the duration-of-exposure variable, this variable can be interpreted as the difference in weight between beneficiary and excluded children at the time of enrollment.
The principal control variable was the age of the child at the time the child was weighed. A combination of linear and log-linear terms captured 80% of the within-child variability in weight. The gender of the child was also included in the model. To account for the fact that being in receipt of Bolsa Escola increased the probability of being excluded from Bolsa Alimentação, the model included a dummy variable representing receipt of Bolsa Escola. The coefficients associated with this variable should not be interpreted as measuring the impact of Bolsa Escola, because the study was not designed to distinguish between impact and selection effects for any program other than Bolsa Alimentação.
Finally, it was necessary to allow for correlation in multiple weight measurements of the same children. This was achieved by including the identifier for each child as a random effect. The random effects model is appropriate when one is not interested in the particular study subjects per se but rather considers them to represent a more general population (17). To determine whether the effects of exposure to Bolsa Alimentação varied according the age of the child, we also tested a more complex model with interaction terms. P-values < 0.05 were considered statistically significant. All analyses were conducted by using Stata v.7.0 (Stata).
|
RESULTS |
|---|
|
TOPABSTRACTSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
2 test, P = 0.001). Almost all (263/282; 93.3%) households categorized as "excluded" were able to confirm that they were not receiving transfers, and almost all (673/717; 93.9%) households categorized as beneficiaries did report receiving transfers. There were few problems with obtaining anthropometric measurements, but 15% of children did not have any weight recorded on their growth monitoring cards. This proportion did not differ by analysis group. For children with weights recorded, the mean number of weights copied to the questionnaires was 6.4. A mean of 3.7 weighings were recorded for each child over the 6-mo period preceding the interview, with more observations for beneficiary children (3.8 ± 2.2, SD) than for excluded children (3.4 ± 2.2, SD; P = 0.027, t test).
|
|
|
|
2 test; P = 0.016). The reduction in weight gain associated with each month of exposure to the program (Table 4) was observed to increase in magnitude up to 12 mo of age (at which point it was equivalent to 274 g less weight gained over a 6-mo period) and then decrease at older ages.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
Either true negative impact of the program or the perpetuation of differences already present before the program was launched might explain the small differences between beneficiary and excluded children identified in the cross-sectional data. However, because of the way in which the exclusions occurred, we have no reason to suspect any systematic bias other than that related to receipt of the other federal program, Bolsa Escola. This effect has been adequately controlled in the analyses. On the other hand, the routine record data allow us to estimate the change in weight gain that accompanied introduction of the program and—separately—the average difference in weight between the 2 groups before the program was launched. The fact that the 2 methods give similar estimates of the net difference associated with participation in the Bolsa Alimentação program suggests that this differential cannot be attributed wholly to the idiosyncracies of the routine weight recording system, nor to the particular specification of the random effects regression model used to analyze these data. Because previous work (19) has shown that the program appears to have increased the availability of nutritious foods in the household, we are inclined to attribute the small negative impact on children’s weight gain to an incentive effect: mothers may have believed that their participation in the program was due to their child being underweight and that the benefits would be suspended should the child start to grow well. This rule was once enforced in a Brazilian federal program called Incentivo para o Combate de Carências Nutricionais, which made milk powder available to mothers of underweight children. Many (probably the majority) of the mothers in our sample had previously been beneficiaries of this program, and there have been anecdotal—and impossible to substantiate—reports of beneficiary mothers deliberately keeping their children malnourished to qualify for the benefits.
In spite of the current popularity of conditional cash transfer programs in Latin America, little is known about their impact on child growth. In Mexico’s PROGRESA program, children receiving both a cash transfer and a multimicronutrient supplement grew about 1 cm more than those receiving neither intervention (10). However, because there was no impact on children in beneficiary communities as a whole (some of whom received the supplement and some of whom did not), the authors of the evaluation study have concluded that the effect was due to the supplement, not the cash transfer. Little is yet known about the impact of comparable programs in Honduras and Nicaragua.
A major limitation of the current study is the lack of a baseline measurement. Evaluations that are based exclusively on data collected after the intervention can be misleading if the groups being compared were already different prior to the introduction of the intervention. We attempted to limit these differences at the design stage by pair matching excluded persons and beneficiaries. At the analysis stage, we conducted all analyses controlling for Bolsa Escola beneficiary status, which we knew differed between the 2 groups. Furthermore, we supplemented the cross-sectional analysis with a retrospective cohort analysis of routinely collected weight data, which included many measurements from before the time that Bolsa Alimentação became operational. However, 15% of children under 3 (in both groups) did not have any weight data on their growth monitoring cards, which may have affected our analysis. In addition, 10–20% of the initial sample could not be located at all (due to deficient administrative records). Even by using the routine data, we did not have any estimate of preprogram height (or length). This could be an important omission, because attained height (or length) is expected to influence subsequent weight gain. It is also unfortunate that we did not have routinely collected data on height gain, because stunting is the most relevant measure of poor growth in this population. However, in populations with minimal or no deficit in weight-for-height (which is the case in Brazil), age-adjusted weight and age-adjusted height respond to the same physiological processes (15). Finally, our findings only apply to the 4 pilot municipalities studied and may not have been replicated in the expansion phase of the program. In fact, it is very likely that families’ concerns about being suspended from the program will have diminished over the course of 2003, because it became obvious that mass suspensions were not occurring.
We conclude that nutrition program planners and implementers should do their utmost to avoid giving the impression that program participation is in some way dependent on a child being (or remaining) malnourished. Conditional cash transfers are a powerful tool for altering household behaviors, including some that are quite resistant to more traditional approaches to behavior change. Precisely because of this effectiveness, great care must be exercised by program planners to ensure that the incentives embodied in the program design are truly those that were originally intended.
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
Manuscript received 4 February 2004. Initial review completed 28 March 2004. Revision accepted 2 June 2004.
|
LITERATURE CITED |
|---|
|
TOPABSTRACTSUBJECTS AND METHODSRESULTSDISCUSSIONLITERATURE CITED |
|---|
1. Allen, L. H. (1994) Nutritional influences on linear growth: a general review. Eur. J. Clin. Nutr. 48(Suppl. 1):S75-S89.
2. Ruel, M. T. & Menon, P. (2002) Child feeding practices are associated with child nutritional status in Latin America: innovative uses of the demographic and health surveys. J. Nutr. 132:1180-1187.
3. Stephensen, C. B. (1999) Burden of infection on growth failure. J. Nutr. 129:534S-538S.[Medline]
4. Wagstaff, A. & Watanabe, N. (2000) Socioeconomic inequalities in child malnutrition in the developing world. World Bank Policy Research Working Paper No. 2434 2000 The World Bank Washington, DC.
5. Martorell, R., Khan, L. K. & Schroeder, D. G. (1994) Reversibility of stunting: epidemiological findings in children from developing countries. Eur. J. Clin. Nutr. 48(Suppl. 1):S45-S57.
6. WHO/UNICEF (1997) Integrated Management of Childhood Illness. Counsel the Mother 1997 World Health Organization Geneva, Switzerland.
7. Interamerican Development Bank Research Department (2003) A new generation of social programs. Ideas for Development in the Americas 1:1-4.
8. Gertler, P. J. & Boyce, S. (2001) An experiment in incentive-based welfare: the impact of PROGRESA on health in Mexico. Working Paper 2001 University of California Berkeley, CA.
9. IFPRI (International Food Policy Research Institute) (2002) Final report: Nicaragua Social Protection Network, pilot evaluation system, impact evaluation 2002 International Food Policy Research Institute Washington, DC.
10. Behrman, J. R. & Hoddinott, J. (2002) Program evaluation with unobserved heterogeneity and selective implementation: the Mexican PROGRESA impact on child nutrition. PIER Working Paper 02–006 2002 University of Pennsylvania Philadelphia, PA.
11. Benício, M.H.A. & Monteiro, C. A. (1997) Desnutrição infantil nos municípios brasileiros: risco de ocorrência 1997 NUPENS/USP/UNICEF Brasília, Brazil.
12. Hotelling, H. (1933) Analysis of a complex of statistical variables into principal components. J. Educ. Psych. 24:417-444, 498–520.
13. Rosenbaum, P. R. & Rubin, D. B. (1985) Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. Am. Stat. 39:33-38.
14. Gibson, R. S. (1990) Principles of nutritional assessment 1990 Oxford University Press Oxford, UK.
15. WHO (World Health Organization) (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.
16. Kirkwood, B. R. & Sterne, J.A.C. (2003) Essential medical statistics 2nd ed. 2003 Blackwell Science Oxford, UK.
17. Hsiao, C. (1986) Analysis of panel data. Econometric Society Monographs No. 11 1986 Cambridge University Press Cambridge, UK.
18. Shrimpton, R., Victora, C. G., de Onis, M., Lima, R. C., Blossner, M. & Clugston, G. (2001) Worldwide timing of growth faltering: implications for nutritional interventions. Pediatrics 107:E75.
19. Olinto, P., Morris, S. S., Flores, R. & Veiga, A. (2003) The impact of the Bolsa Alimentação program on food consumption 2003 Paper presented at annual meetings of the International Association of Agricultural Economists Durban, South Africa, August 2003.
This article has been cited by other articles:
|
|
 |
|
  I. Forde and D. Zeuner Financial incentives to promote social mobility BMJ, August 25, 2009; 339(aug25_2): b3219 - b3219. [Full Text]
|
|
|
|
 |
|
 M. Lagarde, A. Haines, and N. Palmer Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries: A Systematic Review JAMA, October 24, 2007; 298(16): 1900 - 1910. [Abstract] [Full Text] [PDF]
|
|
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
