APPENDIX_2_alternative_tests_index

Published: 29-06-2020| Version 1 | DOI: 10.17632/grbty6gwbc.1
Contributor:
George Angelopulo

Description

Alternative tests data – 2014-19 African inclusive urbanisation index This spreadsheet includes the base data and statistical process used to assess the predictive validity of the instrument described in the paper ‘A comparative measure of inclusive urbanisation in the cities of Africa’. The original 2014 data and its resultant cities index is assessed against data for a consolidated three-indicator set of inclusive urbanisation criteria five years following the term of the baseline study with data available in 2019. The original instrument is assessed in three predominant and originally untested areas. The first is in the area of its constituent variables, or indicators. Second, the value of dividing the index into two related sub-indices as opposed to a single index is tested. The third reviewed area is the statistical construction of the instrument. The original index is subjected to a range of tests evaluating the original instrument in all three areas, and a revised instrument is proposed. Composites of the three-indicator 2014-19 data set are created and correlated with the original 2014 indicator set. Geometric and arithmetic means for both sets are tested, with and without separation of the two sub-indices. The highest Correlation R score is identified as the arithmetic composite paired with the arithmetically averaged original indicator set with leading and lagging subsets. A series of tests are compiled to improve the figure with adjusted combinations of the original indicator set. All 24 indicators are tested against both 2014-19 composites, and 13 removed as they display poor or negative correlations. Different combinations of the remaining 11 indicators are tested and despite demonstrating adequate correlations within the composite, three are further discarded as they lower the overall correlation score of the reconfigured instrument against the three-indicator set. The final eight equally weighted, arithmetically averaged indicators are retained in a reconfigured composite that results in the highest correlation with the arithmetically averaged 2014-2019 composite. The resultant model is used as the foundation of a reconstituted instrument that displays reduced complexity and increased accuracy. Sheet 3 of the spreadsheet replicates the regression analyses and Sheet 2 the correlation analysis for the results that are replicated on Sheet 1, with the indicator set of the reconfigured instrument listed in Columns T-AE.

Files