Additional robustness checks
| Robustness check | Statistic | df | p-value | Result |
|---|---|---|---|---|
| Brant test: omnibus | χ2 = 75.31 | 48 | 0.01 | The proportional-odds assumption was rejected globally |
| Brant test: KNOW | χ2 = 10.60 | 4 | <0.05 | KNOW was the only predictor that violated the proportional-odds assumption |
| Brant test: remaining predictors | 1.13 ≤ χ2 ≤ 7.67 | 4 each | >0.05 | The proportional-odds assumption was satisfied for all remaining predictors |
| RPS: ordered logit model | 0.1463 | – | – | – |
| RPS: partial proportional-odds model | 0.146 | – | – | The PPO model showed a slightly lower RPS |
| Paired bootstrap comparison of RPS | Difference = 0.0003 | – | 0.577 | The predictive difference was not statistically significant; 95% CI [−0.0008, 0.0013] |
| Robustness check | Statistic | df | Result | |
|---|---|---|---|---|
| Brant test: omnibus | 48 | 0.01 | The proportional-odds assumption was rejected globally | |
| Brant test: KNOW | 4 | <0.05 | KNOW was the only predictor that violated the proportional-odds assumption | |
| Brant test: remaining predictors | 1.13 ≤ | 4 each | >0.05 | The proportional-odds assumption was satisfied for all remaining predictors |
| RPS: ordered logit model | 0.1463 | – | – | – |
| RPS: partial proportional-odds model | 0.146 | – | – | The PPO model showed a slightly lower RPS |
| Paired bootstrap comparison of RPS | Difference = 0.0003 | – | 0.577 | The predictive difference was not statistically significant; 95% CI [−0.0008, 0.0013] |
Note(s): RPS = normalised Ranked Probability Score; lower values indicate better ordinal probabilistic prediction. The significant omnibus Brant test was attributable exclusively to KNOW. A partial proportional-odds model was therefore estimated allowing KNOW to vary across thresholds. However, the two models displayed virtually identical predictive performance
Sharing content requires targeting cookies to be enabled. Please update your cookie preferences to use this feature.