Versión de HTML Básico
Assessing the Effect of Conditional Cash Transfers in Children Chronic Stunting: The Human Development Bonus in Ecuador
115
13
Analiti a, Revista de análisis estadístico, Vol. 13 (1), 2017
variance estimator, there would have been incorrect indicative signs of a counterproductive
effect of the transfer.
The RDD design itself is built upon the internal validity of the results for the compliers
units in an arbitrary bandwidth. This could be argued for this case, under the evidence
accounting for an as if random assignment around the threshold in the selected vicinity,
without forgetting the challenges that the study faced. The external validity was tested by
running the same models for the complete evaluation sample. The results for a parametric
approach (Appendix 5) still showed a significant first stage, of around 0,14 to 0,15, not too
far from the vicinity sample. Though, neither the ITT, nor the IV estimates showed similar
patterns. Therefore, the national validity of the results is not likely. Analysing a wider con-
text, we could attempt to make a light comparison with other Latin American programmes.
For instance, the Colombian
Familias en acci´on
, as well as the Mexican
PROGRESA
, and
the Nicaraguan
Red de Protecci´on Social
, have shown significant positive effects over HAZ in
several studies. While, the
Bolsa Alimenta¸c˜ao
in Brazil has evidenced larger gains in anthro-
pometric measures for non-treated children (detailed in section 2.3). The first ones relied
on experimental evaluations which ensured no selection bias, while, the last implemented a
meticulous matching of administrative data trying to resemble an experiment. This could be
pair-wised to the intricate conditions under which the present research had to be conducted,
and give some insight to the reasons of the findings.
Importantly, the previously mentioned programmes that found to be beneficial for chil-
dren, where evaluated in a period of two to three years after the first implementation and
the major gains in height for age were reported for those, in fact, under 24 or 36 months
old. This is also the case for the estimates of the relevant studies of the effect of BDH on
nutritional outcomes in Ecuador previously outlined. Even though, these initial effects of
the programme cannot be estimated with the present design, I analysed if there might be a
differentiated effect depending on the intensity of the treatment. I computed the difference
between the years of the treatment a household received and the age of the kid. Around
89% of the children had received the transfer all their life, which was expected due to the
date of the evaluation. Therefore, the number of individuals that reported less than full life
treatment was only 34, from which only 16 had received it 2 to 3 years. Even though, I
could not proceed to an evaluation with that number of observations, this is a suggestion
for further research that could be done if data more robust could be accessible.
A second exercise I made to analyse this relationship between age and treatment effect,
and also attempting to test the well evidenced argument that the earlier the intervention the
better the expected outcomes (Behrman and Hoddinott, 2005), was to estimate results by
single age groups. Given the advantages showed by the non-parametric approach and due to
the related sample reduction, I computed local polynomial robust estimates for specification
(3). Findings are still negative and non-significant, though; for instance, the group age of
0 to 1 years old has the lowest negative value (Appendix 6). This could incentive further
discussion, since it is somehow aligned to the higher benefits empirically reported for the
33