Publication type
ISER Working Paper Series
Series Number
2007-27
Series
ISER Working Paper Series
Authors
Publication date
January 14, 2008
Abstract:
Estimating the effects of demographic events on households’ living standards
introduces a range of statistical issues. In this paper we analyze this topic
considering our observational study as a quasi-experiment in which the
treatment is expressed by childbearing events between two time points and
the outcome is the change in equivalized household consumption
expenditure. Our main question concerns how one can best estimate causal
effects of demographic events on households’ economic wellbeing. We first
provide a brief discussion of different methods for causal inference stressing
their differences with respect to the underlying assumptions and data
requirement. In particular, we contrast methods relying on the
Uncounfoundedness Assumption (UNA), such as regressions and propensity
score matching, with methods allowing for selection on unobservables, such
as the Instrumental Variable (IV) estimators. We stress the fact that these
methods are not equivalent in what they estimate. With Regressions and
Propensity Score Matching (PSM) we can identify and estimate the Average
Treatment Effect (ATE) and the Average Treatment effect on the Treated
(ATT), while IV methods give the Local Average Treatment Effect (LATE).
Since LATE is the average causal effect of the treatment on the sub-group of
compliers, it is generally different from ATE and ATT. Moreover, different
instruments identify the effect on different groups of compliers giving different
estimates of LATE. A problem for policy making is that the compliers are in
general an unobserved sub-group. However, IV methods estimate relevant
policy parameter if the instrument itself is a potential policy variable. We
demonstrate these issues with an application on data derived from the
Vietnam Living Standard Measurement Study.
Subjects
Notes
working paper
Paper download#511407