Versión de HTML Básico
Lorena Moreno
106
Analiti a, Revista de análisis estadístico, Vol. 13 (1), 2017
Table 6:
Intention-to-treat (parametric)
VARIABLES
(1)
(2)
(3)
Z
-0,342 -0,340
-0,406**
(0,214) (0,214)
(0,197)
X
-0,0397 0,132
-0,369
(0,0546) (0,978)
(0,919)
X2
-0,00304
0,00522
(0,0173)
(0,0162)
age (months)
-0,000249***
(9,39e-05)
sex (1: male)
-0,289***
(0,0900)
ethnicity (1: indigenous)
-0,283**
(0,111)
mother’s height
0,0451***
(0,00877)
mother’s education level
0,0161
(0,0722)
Constant
-0,115 -2,547
-1,357
(1,641) (13,83)
(13,06)
Observations
662
662
662
Robust standard errors in parentheses
*** p
<
0,01, ** p
<
0,05, * p
<
0,1
Dep var: HAZ (Y)
Note:
Coefficients of Z in (1), (2) and (3) are parametric estimates of the effect of the
eligibility status on stunting z-scores. This intention-to-treat estimate is only significant
for the last specification, where all but one covariate have significant associations with
the dependent variable.
The estimates account for an average decrease of around 2 standard deviations in HAZ,
though non-of them are significantly different from 0. For example, in specification (3) the
LATE of -2,147 is the result of -0,406/0,189. The average z-score for those whom are to the
left of the threshold in the specific band was -1,6 (stunting prevalence of 36,46%), which
would become even more deviated when adding the found negative effect. Though, this
interpretation is purely methodological since the estimate is not significant
7
.
7
Note that the estimates look very similar between models, which amounts to the argument of the
fulfilment of the RDD identification strategy
24