Addi van Bergen

Chapter 4 76 of the four dimensions of SE (Table 1 column 1). Eight of these 16 items are also part of the mandatory national questionnaire (PHM1 to PHM7 and PHM9). These items are included routinely in the health surveys. The other eight PHM items are optional, meaning that cities could choose not to include these items. After comparison with the SCP index, five of these eight items were considered redundant and were not included in the health surveys. The three remaining optional PHM items were PHM8, PHM10 and PHM14 (Table 1 column 1). From the SCP social exclusion index nine items were added to the surveys to enhance the content validity of the SEI-HS (Table 1 column 2). These items were selected in previous research from an item pool of 232 items covering the broad spectrum of SE [34]. Four SCP items (SCP12 to SCP15) were added to measure Normative Integration, four items (SCP5 to SCP8) to measure Material Deprivation and one item (SCP11) on not receiving medical or dental treatment was added in the dimension Social Rights. In total, 20 items were available for the construction of the SEI-HS. Construction of the SEI-HS Nonlinear canonical correlation analysis (OVERALS module in SPSS 19.0) was used to construct a multidimensional index and four underlying dimension scales. OVERALS is a suitable method for the construction of a composite measure as it allows multiple sets of variables (here dimensions of SE), different measurement levels (nominal, ordinal or interval) and distributions [44, 45]. The OVERALS algorithm compares the variable sets to an unknown comprise set that is defined by the object scores [44]. If the correlation between the sets is sufficient, it is assumed that these sets refer to a shared underlying concept [45]. In order to test the generalisability of the extended measure, the dataset was randomly split with SPSS “Select Cases” into a development sample (N = 129,464) and a validation sample (N = 129,464). All analyses were carried out in the development sample and replicated in the validation sample. The 20 items were coded in the same direction (low score = little or no exclusion). Based on the OVERALS category quantifications, their measurement level was set as ordinal. Initially all items were entered in the OVERALS analysis, after which items with low component loadings or low weights were removed one by one, until a workable set of items remained. OVERALS weights are considered low at a value of less than 0.100, component loadings at a value of less than 0.300 [44]. Partial cases with maximum three missing values in total and maximum one per dimension were included in the OVERALS analyses. a Since OVERALS does not calculate scores on the subscales, we calculated these by the formula: scale score = Σ transformed item score * item weight. Maximum one missing value was allowed.