Table 2.
Psychometric properties of the AAQ 17-items instrument
| Psychometric properties and related research questions | Statistics/indices | Criterion | Software | Results |
|---|---|---|---|---|
| Unidimensionality: Do all items of the measure assess a common construct? | ||||
| Confirmatory factor analysis (CFA)a | lavaan (R package) | |||
| CFI | > 0.95 | 0.95 | ||
| TLI | > 0.95 | 0.94 | ||
| RMSEA | < 0.06 | 0.15 | ||
| SRMSR | < 0.08 | 0.08 | ||
| Exploratory bifactor analysis | psych (R package) | |||
| ECV | > 0.70 | 0.76 | ||
| Omega H | > 0.80 | 0.84 | ||
| Local independence: Do the items relate only to the construct being measured? | ||||
| Residual correlation matrix resulting from CFA | Residual correlations of item pairs (rRes) | 0.20 | lavaan (R package) | rRes ≤ 0.20 in 97% of item pairs |
| Monotonicity: Do the probabilities of affirmative responses to the items increase with increasing levels of the construct? | ||||
| Mokken scale analysis | Scalability coefficient of the total scale (H) | > 0.50 | mokken (R package) | 0.60 |
| Measurement invariance: Is it valid to use the same IRT-model to compare these groups? | ||||
| Differential item functioning by age (median split) | McFadden’s pseudo R2-change | < 2% | lordif (R package) | R2-change < 2% in 100% of items |
| Differential item functioning by gender (female versus male) | R2-change < 2% in 100% of items | |||
| Differential item functioning by country (Brazilian data versus data from all other countriesb) | R2-change < 2% in 100% of items | |||
| IRT model fit: Can the relationship between the items adequately be described by a GRM? | ||||
| GRM fit | S-X2 p-valuec | ≥ 0.003 | mirt (R package) | p ≥ 0.003 in 100% of items |
CFI comparative fit index, ECV explained common variance, IRT item response theory, GRM graded response model, H Loevinger’s Homogeneity coefficient, r correlation coefficient, RMSEA root mean square error of approximation, SRMSR standardized root mean square residual, TLI Tucker-Lewis index
aFit statistics are based on a weighted least squares means and variance adjusted (WLSMV) estimator, which is a robust variant of the diagonally weighted least squares estimator
bMeasurement invariance across all included countries except Brazil has already been demonstrated (Peter et al. [17])
cS-X2 item fit statistics were evaluated after adjusting for multiple testing (p ≥ 0.003)