Maartje Boer

CROSS-NATIONAL VALIDATION OF THE SMD-SCALE 89 3 For 28 out of 44 countries, the two-factor solution yielded estimation problems. Often, these problems emerged because the two factors showed correlations equal or greater than one. Such model estimates should not be interpreted and warrant re-specification (Bagozzi & Yi, 1988). Hence, for these countries, the two-factor solutions were considered inappropriate. Fromthe 16 countries that showed no estimation problems, the two-factor model showed better model fit than the one-factor model (Table A3.4). Subsequently, we evaluated the quality of the two-factor solutions, whereby each factor should consist of at least three items with significant ( p < 0.05) factor loadings higher than 0.50 without any cross-loadings that differed less than 0.20 (Costello & Osborne, 2005; Howard, 2016). Six out of the 16 countries without estimation problems did not meet this requirement (Denmark, Georgia, Kazakhstan, Latvia, Netherlands, and Turkey), suggesting the quality of the two-factor solution was poor in these countries. Nine out of the 16 countries showed one factor with the items preoccupation, tolerance, and withdrawal and a second factor that was not consistent across countries (Canada, Croatia, England, Ireland, North Macedonia, Poland, Scotland, Serbia, and Wales). More specifically, after removal of items with factor loadings < 0.50 and cross- loadings, the second factor consisted of at least three of the items persistence, displacement, problem, deception, escape, and conflict. One out of the 16 countries showed one factor with items preoccupation, tolerance, withdrawal, and displacement and a second factor with items problems, deception, and conflict (Belgium: Flanders). For 41 out of 44 countries, the three-factor solution yielded estimation problems, also mostly because factor correlations were equal or greater than one. Hence, the three-factor solution was considered inappropriate for these countries (Bagozzi & Yi, 1988). From the three countries that showed no estimation problems, the three-factor model showed better model fit than the one- and two-factor solution (Austria, Belgium: Flanders, and Turkey). However, the quality of the three-factor solutions was poor, because after removal of items with factor loadings < 0.50 and cross-loadings, the second and/or third factor consisted of less than three items (Table A3.4). In sum, in 34 out of 44 countries, results from the eigenvalues and parallel analysis suggested a one-factor solution and, accordingly, the quality of the two- and three-factor solutions was poor. In the 10 other countries, the quality of the two-factor solution was acceptable, however, the eigenvalues