Addi van Bergen

Construction of a multidimensional measure for social exclusion 51 3 Initially all items were entered in the OVERALS analysis. Using category quantifications, the most appropriate measurement level of the items was chosen. Similar to the SCP method [9,12], items with component loadings less than 0.300 were removed one by one, starting with lowest correlations. Subsequently, items with weights less than 0.100 were removed, as well as items that scored in the opposite direction. Finally, scores on the subscales were computed using category quantifications and weights (for formulas see [12]). Measurement properties We used a series of methods to evaluate the measurement properties of the constructed indices, i.e. content validity, internal consistency, internal structure and construct validity. To assess the content validity, we examined whether the constructed indices encompassed all dimensions of social exclusion and whether the included items were representative for the dimensions they were expected to measure. In addition, we inspected the distributions of the index scores and compared these with the SCP index. To assess the internal consistency of the indices we calculated the canonical correlation, which measures the degree to which the items contribute to the underlying latent variable. A canonical correlation of 0.300 was defined as the lower limit to ensure reliability of the indices [9,13]. Cronbach’s alpha was used to measure the internal consistency of the subscales, where we considered α ≥ 0.70 to be good [14]. For the assessment of the internal structure of the instruments, we computed the intercorrelations of the subscales and the general indices. Construct validity was assessed by testing predefined hypotheses [14]. For this purpose we selected a number of items that measure risk factors and correlates of social exclusion, derived from previous SCP research [7,10]. None of these were selected for the construction of the indices. The factors and correlates included were: - Sociodemographic variables : low educational level; non-Western ethnic background; single-parent; living alone; unemployed and/or recipient of social security or disability benefits; no paid job; income below modal (1,700 Euros net per month); and living in a deprived neighbourhood ; - Health related factors : fair or poor self-rated health (versus good or very good); being diagnosed with at least one of eighteen chronic conditions; impaired in daily activities at home, at school, at work or in their leisure time due to chronic conditions (light to strong) and high risk for anxiety and depression disorder (score 30 or higher on Kessler psychological distress scale); - Variables on self-reliance : low perceived life control (Pearlin & Schooler Mastery Scale, score < =19); and need of help to complete the health questionnaire. We expected higher levels of social exclusion in these groups. The construct validity was considered satisfactory if at least 75% of the associations were in correspondence with these expectations. [14].