Maartje Boer

CHAPTER 3 64 Sum-scores were computed and transformed into proportional ranks given their residential country (Elgar et al., 2017), and subsequently divided in three categories (1 lowest 20% , 2 middle 60% , and 3 highest 20% ). Analyses Missing Data Missing data on the study variables were imputed based on multiple imputation with chained equations (Royston & White, 2011). Five imputations were generated using predictivemeanmatchingwith five ‘nearest neighbors’ and logistic regression for the dichotomous items, predicted by the available data on the study measures, demographic characteristics, other wellbeing indicators, and residential country to control for the nested structure of the data. Structural Validity The structural validitydefines theextent towhich the scores on the scale reflect the underlying dimension. The SMD-scalewas developed as a unidimensional scale (Boer, Stevens, Finkenauer, Koning, et al., 2021; Van den Eijnden et al., 2016). Hence, we evaluated the factor structure of the scale based on CFA of a one-factor model, based on the Comparative Fit Index (CFI), Tucker Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA), and Standardized Root Mean Square Residual (SRMR) (CFI/TLI: ≥ 0.9 acceptable, ≥ 0.95 good; RMSEA: ≤ 0.08 acceptable, ≤ 0.06 good; SRMR: ≤ 0.10 acceptable, ≤ 0.08 good) (Hu & Bentler, 1999). We did not rely on the Chi-square statistic given its sensitivity to large sample sizes (Fabrigar et al., 1999). Solid structural validity was established when the model fit was acceptable and at least five items had factor loadings of 0.50 or higher (Costello & Osborne, 2005). Prior to the CFA, we conducted Exploratory Factor Analysis (EFA) for each country to consolidate the proposition that the SMD-scale measures one underlying dimension (Boer, Stevens, Finkenauer, Koning, et al., 2021; Van den Eijnden et al., 2016). The EFA and CFA were conducted on different random subsamples, referring to calibration (EFA) and validation (CFA) subsamples.