Page 168 - Contributed Paper Session (CPS) - Volume 6

CPS1866 Milica Maricic et al.
decide on the indicators which will be used in the composite indicator
framework. One study direction which has been recently developing is the
dimension reduction analysis of composite indicators (Marozzi, 2009). The
goal of such analysis is to exclude indicators used to rank entities and thus
simplify the composite indicator framework. This study aims to propose a
novel hybrid multivariate statistical approach for dimension reduction which
also improves the stability of the metric.
We begin the study with a short literature review on dimension reduction
techniques which have been used in the field of composite indicators. The
Section 3 sees the presentation of the methodologies. We first present in brief
the methodological framework of the Sustainable Society Index (SSI), the
composite index which was here used as a case study. Next, we introduce the
basics of the enhanced Scatter Search – Composite I-Distance Indicator (eSS-
CIDI) approach which we used to reduce the dimensionality of the SSI. The
obtained results are provided in Section 4, while the concluding remarks are
given in the final chapter.

2. Dimension reduction
The issue of dimension reduction is a topic of high interest for researchers,
but also for policy makers. So far different approaches have been suggested.
Namely, Fodor (2002) in his detailed literature review on dimension reduction
listed Principal Component Analysis (PCA), Factor Analysis (FA), Projection
pursuit (PP), Independent component analysis (ICA), Non-linear principal
component analysis, Random projections and other non-linear methods and
extensions as major dimension reduction techniques. Herein, we will place our
attention on the methods which have been used in the field of composite
indicators.
One of the most common dimension reduction techiniques is the Principal
component analysis (PCA) which was initialy proposed by Carl Pearson. The
idea behind the PCA is to find a linear combination of variables which accounts
for as much variation in the original variables as possible (Tabachnick & Fidell,
2013). The benefits of this analysis have been acknowledged by composite
indicator creators. Namely, the OECD Handbook on creating composite
indicators suggests to perfrom PCA to define the dimensionality of the
composite indicator and to define weights (Nardo et al., 2005). Just one of the
examples of researches in the field of composite indicators which employ the
PCA are Kotzee & Reyer (2016).
In his research Marozzi (2009) proposed a four-step algorithm for
dimension reduction based on the Spearman correlation coefficient. The first
step is to create the composite indicator using all indicators and obtain the
rank R . In the following step h indicators are excluded where h 1,2,...,k ,
k

157 | I S I W S C 2 0 1 9