Page 406 - Special Topic Session (STS) - Volume 4

STS2320 Mohamed A I.
to a selected list of variables and aggregations. Based on this feedback, and
that of the core team who prepared the report, a final list of variables was
produced.
Principal components analysis was used to confirm the consistency of the
selected variables and the structure of their classification into the various sub-
indices, further supporting the consistency of the broader conceptual context
across the variables and their classification in the sub groups—for which the
explained variance ratio in most cases exceeded 50 percent, see Hair et al.
(2015).
The results of the in-depth correlation analysis and Cronbach’s Alpha
coefficient (exceeding 0.70 in most cases) confirmed the validity of the
selection and classification of the variables. Furthermore, the correlation
matrix for normalised variables was analysed to ensure that they followed the
same direction as the composite index, confirming the need to include
variables with high correlation coefficients (above 0.9) with other variables.

Data collection
The 133 variables employed in the 2018 GKI were obtained from sources
including the United Nations Educational, Scientific and Cultural Organization
(UNESCO); the World Bank; the International Telecommunication Union (ITU);
the World Economic Forum (WEF); the International Monetary Fund (IMF); the
Organisation for Economic Co-operation and Development (OECD); the
International Labour Organization (ILO) and other UN and international
agencies. The team reviewed the data multiple times to ensure no errors had
occurred during data entry; consequently, data was processed on the
assumption that it was error-free. In the cases where those variables were
linked to other size-dependent variables – such as population or GDP – results
were recalculated after adjusting for the effect of the size. Variables included
are in the form of hard data, composite indicators and survey
questions/responses. The most recent data for each variable within the period
2007–2018 was used.
As a prerequisite, data employed in the construction of the composite
indices have met certain statistical criteria. For example, each country was
required to have at least 50 percent of the figures for variables in each sectoral
index for it to be included in the general index (GKI). The team had to ensure
these criteria were met before calculating the composite index. The methods
used to identify and treat outliers, severe skewness and severe kurtosis are
outlined below.

Data treatment
A variable was considered to have severe skewness if its absolute skewness
coefficient was above 2.25, while an absolute kurtosis coefficient above 3.5
indicated that the variable had severe kurtosis. Conditions were relaxed due

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