Table 3.
χ2 Difference testing between 3 branches of nested models
| Model | Δχ2 wrt previous model in Branch 1 | df | Δχ2 wrt previous model in Branch 2 | df | Δχ2 wrt previous model in Branch 3 | df |
|---|---|---|---|---|---|---|
| 1. Standard model (14 latents), incl. QL | Root node | Root node | Root node | |||
| 2. Physical health, mental health and QL | 293 | 17 | – | – | – | – |
| 3. Physical burden, mental function and QL | 77 | 2 | – | – | – | – |
| 4. Symptom burden, function and QL | – | – | 377 | 15 | – | – |
| 5. HRQL and QL | 241 | 3 | 47 | 2 | – | – |
| 6. Formative symptom burden (free weights), function, and QL | – | – | – | – | 336 | 12 |
| 7. Formative symptom burden (fixed weights), function, and QL | – | – | – | – | 241 | 5 |
All χ2 difference tests of model comparisons were significant at P < 0.01
χ2 difference testing—when using the WLSMV estimator—is not a simple difference between two model χ2s. In addition, a model can only be directly compared—using χ2 difference testing—with other models in the same branch of (nested) models