To ensure that the identification strategy is valid, before presenting the main results, the first stage of the regression is discussed, and the validation tests of the instrument are analyzed. As expected, belonging to the group of municipalities eligible to receive the program increases the chances of being in the treatment group, with a coefficient of 0.2996 statistically significant at a level of 1%. The Hausmann Test has a critical value of 322.69 and rejects the null hypothesis that there is no difference between the estimates of Ordinary Least Squares (OLS) and of Instrumental Variables (IV). Therefore, VI is more consistent and accurate. Another important aspect is that the instrument must be a good predictor of the treatment variable. To verify that the instrument is strong, the F test must contain values above ten and the value found in the estimation was 4,946. The identification strategy is also based on the assumption that treated and controlled municipalities follow parallel trends prior to the program. To verify the similarity, Appendix A.2 presents the graphs of trends between the groups for the variables that the PCF had a significant impact. In view of these, when following parallel trajectories, it appears that the groups follow common trends.
With these assumptions met, Table 3 presents the results for the instrumented DD two-way fixed effects model. The first column is the estimation with only the fixed effects of municipality and time, column (2) includes the control variables and in (3) the specification of instrumented DD is presented.
Table 3 – Estimated impacts
|
(1)
|
(2)
|
(3)
|
Mortality rate
|
By preventable deaths – under 5 years old
|
0.054*
|
0.000
|
-0.098
|
(0.032)
|
(0.048)
|
(0.224)
|
By fetal deaths
|
0.588**
|
0.063
|
0.335
|
(0.244)
|
(0.225)
|
(1.073)
|
By infant deaths
|
0.714**
|
0.091
|
0.084
|
(0.301)
|
(0.238)
|
(1.140)
|
By maternal deaths
|
-0.007***
|
-0.002***
|
-0.010***
|
(0.001)
|
(0.001)
|
(0.002)
|
Nutritional status of the pregnant woman
|
Underweight proportion (%)
|
-0.006***
|
0.001
|
0.001
|
|
(0.002)
|
(0.003)
|
(0.016)
|
Proportion with adequate weight (%)
|
-0.008***
|
0.005
|
-0.019
|
(0.002)
|
(0.004)
|
(0.022)
|
Nutritional status of children aged 0 to 6 months - weight x height
|
Thinness proportion (%)
|
-0.007***
|
-0.019***
|
-0.082***
|
|
(0.002)
|
(0.004)
|
(0.025)
|
Appropriate weight proportion (%)
|
0.004
|
0.005
|
0.096**
|
(0.005)
|
(0.008)
|
(0.042)
|
Nutritional status of children aged 0 to 2 years - weight x height
|
Thinness proportion (%)
|
-0.002*
|
-0.001
|
-0.004
|
(0.001)
|
(0.001)
|
(0.006)
|
Appropriate weight proportion (%)
|
-0.000
|
0.004*
|
-0.004
|
(0.002)
|
(0.002)
|
(0.013)
|
Source: prepared by the authors.
Notes: 1. *, ** and *** represent 90%, 95% and 99% confidence, respectively. 2. Considering robust standard errors. 3. The overweight variables were removed, as they are not the main goal of the analysis of the study. 4. The variables of proportion of thinness and extreme thinness were added. Therefore, thinness refers to the bottom 10% quintile of weight. 5. The spatial dependence of the dependent variables was tested. The I-Moran test with the highest value was 0.095. This indicates that the spatial dependence is small and does not pose a problem for estimation.
The results indicate that the municipalities that received the PCF are able to reduce the maternal mortality rate and improve the nutritional status of children from 0 to 6 months of age. The proportion of babies who experience thinness is reduced, while they migrate to an adequate weight. By analyzing column (3), with the DDIV estimate, it is possible to verify that the treatment effect parameter increases as the model becomes more robust.
As the program is recent and the data are grouped by municipalities, it is expected that the impact will still be small or not significant – as is the case with nutrition for the group of children up to two years of age and mortality rates. It corroborates what was presented in the background section of the program, in which families with children up to six months old receive weekly visits, for the others, visits occur fortnightly. That is, there is a difference in the intensity of treatment between the cohorts. In addition, the program seeks to improve family ties and, up to six months of age, the baby's food depends exclusively on the mother - either through breastfeeding or because the mother is closer (she usually returns to the labor market after this phase) – which may be favoring the impact.
Studies that assess the impact of interventions to help pregnant women and children show positive results on birth weight and child health (23,29,31,32). Thus, in the case of the PCF, by contributing to the reduction of the maternal mortality rate, it means that, possibly, mothers are accumulating health capital and babies are also able to enjoy a better nutritional status.
As the nutritional status of babies improves with program actions, the long-term implications are important. For Victora et al. (2008) (19), height-weight adequacy is one of the main predictors of human capital. Thus, by avoiding losses and improving child health indicators, the PCF can contribute to the propagation of results in adult health, education and the income of the beneficiaries.
6.1 ROBUSTNESS ANALYSIS
To ensure the robustness of the results, this section seeks to bring additional estimates for verification. The first strategy, presented in column (1) of Table 4, is to exclude the municipalities of Rio Grande do Sul (RS), since they already have a similar program, the PIM. This is also carried out in the study of Munhoz et al., (2022) (42), in an attempt to avoid contaminating the results of the two programs, as they have convergent goals. In column (2) a doubly robust approach is proposed (DD weighted by the treatment propensity score). The idea is to generate a closer group in terms of observable characteristics. In columns (3.a) and (3.b) the time lag of the dependent variable is included in the specification, that is, the dynamic panel approach seeks to control the temporal persistence. In column (4) the treatment variable assumes a continuous character, to verify how the time of exposure to the treatment affects the variables of interest.
Table 4 – Estimated impacts
|
(1)
|
(2)
|
(3.a)
|
(3.b)
|
(3)
|
Mortality rate
|
By preventable deaths – under 5 years old
|
-0.017
|
-0.025
|
-0.074
|
0.008
|
-0.055
|
(0.234)
|
(0.057)
|
(0.075)
|
(0.013)
|
(0.124)
|
By fetal deaths
|
1.238
|
0.021
|
0.128
|
0.036**
|
0.186
|
(1.089)
|
(0.241)
|
(0.336)
|
(0.014)
|
(0.595)
|
By infant deaths
|
0.597
|
0.022
|
0.142
|
0.014
|
0.046
|
(1.163)
|
(0.298)
|
(0.372)
|
(0.014)
|
(0.633)
|
By maternal deaths
|
-0.010***
|
-0.002**
|
-0.001*
|
-0.037**
|
-0.006***
|
(0.002)
|
(0.001)
|
(0.001)
|
(0.014)
|
(0.001)
|
Nutritional status of the pregnant woman
|
Underweight proportion (%)
|
0.012
|
0.000
|
-0.002
|
0.069***
|
0.000
|
|
(0.017)
|
(0.003)
|
(0.004)
|
(0.013)
|
(0.009)
|
Proportion with adequate weight (%)
|
-0.012
|
0.011**
|
0.005
|
0.034***
|
-0.011
|
(0.023)
|
(0.005)
|
(0.005)
|
(0.013)
|
(0.012)
|
Nutritional status of children aged 0 to 6 months - weight x height
|
Thinness proportion (%)
|
-0.083***
|
-0.015***
|
-0.000
|
0.021
|
-0.044***
|
|
(0.025)
|
(0.005)
|
(0.007)
|
(0.014)
|
(0.013)
|
Appropriate weight proportion (%)
|
0.087**
|
0.005
|
-0.017
|
0.062***
|
0.052**
|
(0.042)
|
(0.009)
|
(0.011)
|
(0.015)
|
(0.023)
|
Nutritional status of children aged 0 to 2 years - weight x height
|
Thinness proportion (%)
|
0.006
|
-0.000
|
0.001
|
0.028**
|
0.002
|
|
(0.006)
|
(0.001)
|
(0.001)
|
(0.012)
|
(0.003)
|
Appropriate weight proportion (%)
|
0.004
|
0.002
|
0.002
|
0.074***
|
-0.002
|
(0.013)
|
(0.003)
|
(0.003)
|
(0.013)
|
(0.007)
|
Source: prepared by the authors.
Notes: 1. *, ** and *** represent 90%, 95% and 99% confidence, respectively. 2. Considering robust standard errors. 3. The overweight variables were removed, as they are not the main goal of the analysis of the study. 4. The variables of proportion of thinness and extreme thinness were added. Therefore, thinness refers to the bottom 10% quintile of weight.
With the exception of the dynamic panel estimation, the other specifications showed that the results are robust. Without the RS, in terms of magnitude, the difference was small. The same occurs with the inclusion of the treatment propensity score, with the addition of a positive effect on the adequate nutrition of the pregnant woman. Regarding the time of exposure to treatment, it is possible to verify that there are gains in the reduction of maternal mortality and reduction of the proportion of children in a state of thinness with a longer time of adherence to the program.
As Brazil has an extensive territory and diversity among regions, added to the context of greater adherence in certain regions to the program, the last analysis contemplates an exercise with heterogeneous effects. With the base category (omitted) the Northeast region, the effects of the other regions are shown in Table 5.
Table 5- Heterogeneous Effects by region - Category omitted: Northeast
|
Treatment
|
South
|
Southeast
|
Midwest
|
North
|
Mortality rate
|
By preventable deaths – under 5 years old
|
-4.448***
|
-4.959***
|
-1.761***
|
-0.990***
|
-0.059
|
(0.306)
|
(0.549)
|
(0.158)
|
(0.178)
|
(0.144)
|
By fetal deaths
|
-29.785***
|
-30.615***
|
-9.195***
|
-5.894***
|
-1.182*
|
(1.697)
|
(3.393)
|
(0.934)
|
(0.986)
|
(0.654)
|
By infant deaths
|
-28.526***
|
-30.218***
|
-9.231***
|
-5.545***
|
1.249
|
(1.815)
|
(3.455)
|
(0.964)
|
(1.027)
|
(0.793)
|
By maternal deaths
|
-0.138***
|
-0.142***
|
-0.049***
|
-0.036***
|
-0.021*
|
(0.031)
|
(0.034)
|
(0.012)
|
(0.013)
|
(0.011)
|
Nutritional status of the pregnant woman
|
Underweight proportion (%)
|
0.015
|
0.017
|
0.016***
|
0.009
|
0.016***
|
|
(0.013)
|
(0.015)
|
(0.005)
|
(0.006)
|
(0.005)
|
Proportion with adequate weight (%)
|
0.049***
|
0.045**
|
0.015**
|
0.005
|
0.023***
|
(0.017)
|
(0.019)
|
(0.007)
|
(0.008)
|
(0.006)
|
Nutritional status of children aged 0 to 6 months - weight x height
|
Thinness proportion (%)
|
-0.033*
|
-0.009
|
0.015**
|
0.013
|
0.003
|
(0.017)
|
(0.019)
|
(0.007)
|
(0.010)
|
(0.008)
|
Appropriate weight proportion (%)
|
0.044*
|
0.030
|
0.017
|
-0.006
|
0.004
|
(0.026)
|
(0.031)
|
(0.012)
|
(0.018)
|
(0.014)
|
Nutritional status of children aged 0 to 2 years - weight x height
|
Thinness proportion (%)
|
0.004
|
-0.001
|
0.001
|
0.013***
|
-0.000
|
(0.005)
|
(0.005)
|
(0.002)
|
(0.003)
|
(0.003)
|
Appropriate weight proportion (%)
|
0.022**
|
0.006
|
0.022***
|
0.005
|
0.013***
|
(0.010)
|
(0.012)
|
(0.005)
|
(0.006)
|
(0.005)
|
Source: prepared by the authors.
Notes: 1. *, ** and *** represent 90%, 95% and 99% confidence, respectively. 2. Considering robust standard errors. 3. The overweight variables were removed, as they are not the main goal of the analysis of the study. 4. The variables of proportion of thinness and extreme thinness were added. Therefore, thinness refers to the bottom 10% quintile of weight. 5. The package used to estimate heterogeneous effects is ivtreatreg.
The main effect parameter (trat.) shows evidence of improvement in mortality and nutrition indicators for pregnant women and children. The heterogeneous effects between regions are significant, especially among mortality rates. The group of municipalities omitted (from the northeast region) has the greatest effects, since the estimation of the other regions has negative coefficients. Similarly to what Rocha and Soares (2010) found for the PSF, the regions with the greatest impact of the PCF were the North and Northeast, followed by the Midwest, Southeast, and South. Additionally, the results found on maternal deaths and on the nutrition of children up to six months of age remained robust.