© Andrei Udriște, 2025
The discrimination index is calculated as the corrected point-biserial (item-rest) correlation — the correlation between an item’s outcome (0 = incorrect, 1 = correct) and the total score on the remaining items.
Thresholds applied are stricter than in mainstream psychometric practice with multiple-choice tests because open-response items are not subject to guessing noise and can therefore achieve markedly higher correlations, a point emphasized to me by Paul Cooijmans.
| Designation | Discrimination index |
|---|---|
| Excellent | ≥0.7 |
| Very Good | 0.6–0.6999 |
| Good | 0.5–0.5999 |
| Acceptable | 0.4–0.4999 |
| Borderline | 0.3–0.3999 |
| Poor | 0.2–0.2999 |
| Nonfunctional | ≤0.1999 |
| Insufficient data | NaN |