Understanding the Determinants of Health in Russia: A Dynamic Panel Attrition Analysis using the RLMS

sugarannoyedΠολεοδομικά Έργα

16 Νοε 2013 (πριν από 3 χρόνια και 8 μήνες)

89 εμφανίσεις


Understanding the Determinants of Health in Russia:

Panel Attrition Analysis using the RLMS

Christopher J Gerry, Carmen A Li and George Papadopoulos

13 November 2011

Extended abstract

The high and fluctuating rates of morbidity and mortality that have emerged in Russia during
the last half century and peaking during the 1990s transitional period have been well
documented. The most recent decade, dominated (until the 2008 global financia
l crisis) by
high levels of economic growth, reductions in poverty, increased public spending and the
explicit prioritisation of health by the Russian government has witnessed a gradual
improvement in most health indicators. Despite this, levels of morbidi
ty and mortality remain
markedly higher than in similar middle
income countries and though explanations abound for

in health, better understanding of the causal explanations for the

of ill
health remains elusive. In this paper we

address all three of these concerns
through examining

how socio
economic variables affect the dynamics of individual (self
assessed) health using waves 11
18 of th

Russian Longitudinal Monitoring Surv

In doing this we note

the importance of sample non
response for this and other
health related
survey data

and so

examine non
response patterns
, test for
attrition bias, and
apply the inverse probability weighting (IPW) to correct for this
We further discuss

limitations of this approach including its sensitivity to the choice of the weighting model and
compare estimation strategies in the presence of attrition.

Individual level longitudinal

data are important for understanding the

determinants of health.

They provide data from which it is possible to estimate health production functions while also
allowing the effects of persistence to be captured separately from the effects of key socio
economic and demographic variables.

This is
important because failure to account for
persistence in health outcomes can result in erroneous results (Gerry, 201

In econometric


therefore, longitudinal data allows us to

control for the presence of unobserved

while at the same t
ime capturingthe dynamics of the
wever, there is no free lunch: not only is panel data difficult and
expensive to
collect, but it prompts a set of panel and subject specific statistical challenges stemming from
the possibilities of non
random sample attrition. That is, by design, longitudinal survey data
presents the possibility that the initially selected sample of respondents will cease to
participate over time in ways that are causally related to their characteristics.
If this non
sponse is related, for example, to health status then it is important to understand the effects
of this ‘attrition’ in any analysis of health production. It is this

which provides the
main focus of this paper.

We are not aware of any other
study examining
health dynamics in Russia in th
context of models that look closely at

the nature of

and non

attrition corrected health dynamics using
data from the
Russian Longitudinal
Monitoring Survey

igher School of Economics (hereon RLMS)

from 2001
inclusive. We choose 2001 as our starting point because the RLMS was substantially
replenished for its 2001 edition and the years from 2001 through 2009 represent a relatively
stable political period a
nd a growing economy for Russia, leading through to the global
financial crisis from 2008. Our main contributions are threefold:
(i) we provide the first
examination of health dynamics in the Russian Federation that explores the nature
and impact
of survey attrition and non
esponse; (ii) we
move beyond Contoyannis

et al

and the restrictive assumption of monotonic attrition and
explore the effects of both
permanent and temporary attrition; (iii) we examine the socioeconomic determinants of
ealth in Russia in the 2000s, using the most up
date and nationally representative data.

These contributions and the findings related to them will serve to inform policy discussion in
the health arena, will provide new profiles of Russian health outcom
es and will further
promote and publicise the RLMS data at this important time in its lifecycle.

We proceed as follows. In section 2 we briefly survey the relevant literature in this field. In
the next two
sections we explain and discuss the theoretical
and empirical foundations of
incorporating an analysis of sample non
response into
estimates of health determinants. In
section 5 we present

results. Section 6 concludes.