Addi van Bergen

Construction of a multidimensional measure for social exclusion 59 3 Table 5. Pearson correlations coefficients between general indices and dimension subscales, SCP and the four cities. Correlation between: SCP Index a Index1 Amsterdam Index2 Rotterdam & The Hague Index 3 Utrecht General index x dimension 1 0.76 0.78* 0.76* 0.82* General index x dimension 2 0.70 0.79* 0.72* 0.68* General index x dimension 3 0.77 0.73* 0.81* Dimension 1 x dimension 2 0.35 0.30* 0.34* 0.33* Dimension 1 x dimension 3 0.43 0.39* 0.55* Dimension 2 x dimension 3 0.44 0.34* 0.38* * p<.01 a Vrooman and Hoff [10]. Construct validity As can be seen from Table 6 all predefined hypotheses were confirmed. Without exception, the indices were positively associated with the selected risk factors and correlates. Regression coefficients showed the expected direction and were statistically significant (p < 0.01). Persons with lower income were more often socially excluded than people with a higher income. People in poor health, persons of non- Western origin and those with low perceived self- control were also at higher risk. The same holds for lower educated persons, people living in deprived neighbourhoods, jobless adults, single persons and single parents. In contrast to the SCP, we also found significant associations with low labour market position and need of assistance in filling in the questionnaire. In general, the associations found in the current research were stronger than in the SCP study. Index selection and generalisabilty When compared with the other indices, Index3 performed best on content validity and performed equally well with regard to internal consistency, internal structure and construct validity. For that reason we continued our analysis with Index3. Generalisability of the items from Index3 was tested in the datasets of Rotterdam/The Hague and Amsterdam, where available. We performed analyses with 2 and 3 sets of variables. In all cases, the OVERALS analysis yielded indices with comparable measurement properties i.e. a distribution of index scores, internal validity, internal structure and construct validity that was similar to Index3.