Figure 3. Box plots of error rates for dichotomizing at cut-point of mRS 1 (mRS 0–1, 2–6), trichotomizing at cut-points 1 and 4 (mRS 0–1, 2–4, 5–6), dichotomizing at mRS 2 (mRS 0–2, 3–6), and trichotomizing at cut-points 2 and 4 (mRS 0–2, 3–4, 5–6).

van Swieten’s inter-rater reliability matrix used as confusion matrix. Post-hoc testing shows that both trichotomization errors are higher than dichotomization (p<.05).