Versión de HTML Básico
Assessing the Effect of Conditional Cash Transfers in Children Chronic Stunting: The Human Development Bonus in Ecuador
97
13
Analiti a, Revista de análisis estadístico, Vol. 13 (1), 2017
and rural levels (3.076 households sample).
The estimation process of the original RSII index started with the merging of the 2012
HSSS database to a smaller one containing Unsatisfied Basic Needs (UBN) poverty percent-
ages of disaggregated geographical units of the 2010 Census. Then, a set of individuals,
assets, and housing variables, theoretically and empirically related to poverty, were cor-
related with the per-capita consumption to narrow this selection according to association
levels. Those variables with associations higher than 0,11 were upwardly recoded assign-
ing the value of 1 to their “worst” category. Then, the RSII index was built as a simple
sum of the re-scaled category quantifications of 34 variables obtained with the CAPTCA as
explained earlier (SIISE, 2014).
The replication strategy followed the exact procedure using the 2014 LSMS and the
Census
4
. One of the variables could not be constructed from the questionnaire; therefore,
only 33 variables were calculated. With this input, I run a version 2.0 CAPTCA algorithm
in SPSS 23, attempting to be as close to the original. In Appendix 1, I present the main
output of the analysis: iteration history, model summary and component loadings. The
2-dimension specification accounted for 36,5% (the original accumulated 33,7%) of the total
variance and had a 62,2% correlation with the monthly per-capita aggregate consumption
(formerly 62,4%). Figure 3 shows a histogram of the frequency of households by the created
index.
Also importantly, cutoff choice was originally done by selecting values of the index that
represented households with consumption poverty. For SELBEN the eligibility criterion was
to keep the first two poverty quintiles and for the RS it was the point estimate of the poverty
line consumption value from an OLS of the index and the logarithm of the aggregate per-
capita consumption (Ponce, 2013). The RSII followed a similar approach to the last, but
since the “graduation” strategy started, the critical value was estimated around the extreme
poverty, specifically, 28,2. I kept the official threshold since computed pseudo-scores used
the same methodology and it meant less manipulation of the design.
4.2.2 Effect estimation
The RDD, firstly discussed by Thistlethwaite and Campbell (1960) can be applied when
there is precise knowledge of the rules determining treatment and when the following basic
elements are present: an outcome, a continuous assignment covariate, a threshold and a
treatment variable (Cook, 2008).
The identification assumption of the design, under the Rubin Causal Model (RCM)
setup, is that the conditional expectations functions of the potential outcomes
Y
1
and
Y
0
,
4
Other approaches applied but discarded were: OLS coefficients of the aggregate consumption and all
the variables in the HSSS as category weights, regression based estimated cutoff, nuclear families instead of
households, complete sample without distinguishing by surveying-period, and different combinations of the
previous.
15