Addi van Bergen

Chapter 3 50 Construction of measurement instrument Following the SCP procedures, we applied non-linear canonical correlation analysis (OVERALS) to the different sets of survey data. OVERALS is an optimal scaling technique developed by the University of Leiden, which is available in the SPSS software package. Canonical correlation analysis is often used to explore relationships between two sets of variables, an independent and dependent set, and to reduce the dimensionality to a few linear combinations of the measures under study [13]. In the context of the current study, we used canonical correlation analysis to construct a composite index based on selected sets of variables, each measuring one of the four dimensions of social exclusion (Figure 1). OVERALS differs in three ways from standard linear canonical correlation analysis: variables can be nominal, ordinal or interval; there can be more than two sets of variables; and instead of maximizing correlations between the variable sets, the sets are compared to an unknown compromise set that is defined by the object scores [13]. If the correlation between the sets is sufficient, it is assumed that these sets refer to an underlying concept. [9,12]. Figure 1. Measurement model for social exclusion. The model illustrates the construction of a composite index based on selected sets of variables, that each measures one of the four dimensions of social exclusion. From each dataset we selected items matching one of the four dimensions of social exclusion as operationalised by the SCP. All items were coded in the same direction, so that a high score refers to more exclusion. Records with one or more missing values on all dimensions were removed from the analyses. As the items in The Hague and Rotterdam datasets matched exactly, these were merged. The analysis thus resulted in three indices: Amsterdam (Index1), Rotterdam / The Hague (Index2) and Utrecht (Index3).