Table 13.
Measurement invariance across gender.
| Model | χ 2 | df | CFI | TLI | RMSEA | 90% CI | SRMR | ΔCFI | ΔRMSEA | ΔSRMR | Decision |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Configural | 1264.25 | 960 | 0.981 | 0.979 | 0.026 | [0.022, 0.029] | 0.045 | — | — | — | Supported |
| Loading invariance (metric) | 1288.46 | 987 | 0.981 | 0.980 | 0.025 | [0.021, 0.029] | 0.046 | 0.000 | −0.001 | 0.001 | Supported |
| Threshold invariance (scalar) | 1281.54 | 981 | 0.981 | 0.980 | 0.025 | [0.021, 0.029] | 0.046 | 0.000 | 0.000 | 0.000 | Supported |
Models were estimated using DWLS for ordered categorical indicators. ΔCFI, ΔRMSEA, and ΔSRMR compare each model with the immediately preceding model. The reduction from df = 987 in the loading-invariance model to df = 981 in the threshold-invariance model was verified against the original software output. Under threshold invariance, six latent-factor means in the nonreference group became freely estimated while the corresponding reference-group means remained fixed to zero for identification. Releasing these six parameters accounts for the six-degree-of-freedom decrease. Invariance decisions were based primarily on changes in approximate fit indices.