| Continuous full-sample model |
n = 180 |
Adjusted R2 = 0.462
for GLS + GWE model |
0.371, 0.541 |
Maximum VIF = 2.36 |
| Tertile classification |
n = 120 |
AUC = 0.91
for GLS + GWE model |
0.85, 0.96 |
Accuracy = 85.8% |
| Quartile classification |
n = 90 |
AUC = 0.90
for GLS + GWE model |
0.82, 0.96 |
Accuracy = 84.4% |
| Decile classification |
n = 36 |
AUC = 0.89
for GLS + GWE model |
0.76, 0.98 |
Accuracy = 83.3% |
| Bootstrap stability |
1, 000 resamples |
Median βGLS = -0.26;
median βGWE = 0.23 |
βGLS: -0.35, -0.17; βGWE: 0.12, 0.32 |
Direction retained in 94.7% of resamples |
| Normality of model residuals |
n = 180 |
Shapiro-Wilk W = 0.982 |
p = 0.116 |
Assumption retained |
| Variance equality for primary group models |
n = 120 |
Levene’s test range = 0.084-0.612 |
- |
Welch correction used where p < 0.05 |
| Multiple-testing correction |
n = 180 |
Benjamini-Hochberg procedure |
q threshold = 0.05 |
Primary GLS and GWE findings retained |