Addi van Bergen

Social Exclusion Index-f or Health Surveys (SEI-HS) 77 4 Trichotomisation As an important application of the SEI-HS in public health policy will be the comparison of SE rates between population groups and monitoring changes over time, we trichotomised both index and scaling scores. The P85 and P95 have been chosen as cut-off points in consultation with Community Health Service epidemiologists. Scores less than or equal to the 85 th percentile in the weighted population were labelled “little or no” exclusion, scores greater than the 85 th percentile but smaller than or equal to the 95 th percentile “some”, and scores greater than the 95 th percentile were labelled “moderate to strong” exclusion. Measurement properties The final version of the SEI-HS was evaluated on (1) content validity, (2) internal consistency, (3) structure, (4) construct validity, and (5) generalisability. The analyses were carried out in the development sample and replicated in the validation sample. 1. Content validity: We examined whether all dimensions and aspects of SE of the SCP index were measured by the SEI-HS and compared the distributions of the SEI-HS and the SCP index. 2. Internal consistency: The canonical correlation in OVERALSmeasures the degree to which the items contribute to the underlying construct of SE. The internal consistency of the index was considered sufficient if the canonical correlation was 0.30 or higher [33, 45]. The internal consistency of the underlying dimension scales was considered sufficient if Cronbach’s alpha was 0.70 or higher [46]. 3. Internal structure: We computed the intercorrelations between the subscales and the general index. We expected strong positive correlations between the subscales and the general index (r > = 0.60) and sufficient but not strong positive correlations between the subscales (0.20 < = r < 0.40) [47, 48]. If the correlations between the subscales are sufficient, it is assumed that these scales refer to a shared underlying concept [45]. Additionally, we conducted confirmatory factor analysis in AMOS. We considered a root mean square error of approximation (RMSEA) < 0.05 and upper bound of 90% confidence interval (HI90) < 0.06, Tucker-Lewis index (TLI) ≥ 0.95, comparative fit index (CFI) > 0.90 and Hoelter’s .05 Index ≥ 200 to indicate good model fit [49]. 4. Construct validity: We tested a number of hypotheses using linear regression analysis (point biserial correlation). Based on previous research, we expected a positive correlation between the SEI-HS and the following risk factors and correlates: low educational level, non-Western ethnic background, single- parent family with minor children, living alone, low labour market status (and/ or recipient of social security or disability benefits), not having paid work, low household income, health problems and living in a deprived neighbourhood. Household income referred to the standardised disposable household income after payment of income tax and social contributions. Low household income