Skip to main content
. 2020 Aug 20;10(8):e037744. doi: 10.1136/bmjopen-2020-037744

Table 3.

Pairwise agreement between treatment hierarchies obtained from the different ranking metrics measured by Spearman ρ, Kendall τ, Yilmaz τAP and Average Overlap

pBVvsSUCRAF SUCRAFversus relative treatment effect pBVversus relative treatment effect SUCRAFversusSUCRAB
Spearmanρ 0.9 (0.8 to 0.96) 1 (0.99 to 1) 0.9 (0.8 to 0.97) 1 (0.98 to 1)
Kendallτ 0.8 (0.67 to 0.91) 1 (0.95 to 1) 0.8 (0.69 to 0.91) 1 (0.93 to 1)
YilmazτAP 0.78 (0.6 to 0.9) 1 (0.93 to 1) 0.79 (0.65 to 0.9) 1 (0.93 to 1)
Average Overlap 0.85 (0.72 to 0.96) 1 (0.91 to 1) 0.88 (0.79 to 1) 1 (0.94 to 1)

Medians, first and third quartiles are reported.

Relative treatment effect stands for the relative treatment effect against fictional treatment of average performance.

PBV, probability of producing the best value; SUCRAB, surface under the cumulative ranking curve (calculated in Bayesian setting); SUCRAF, surface under the cumulative ranking curve (calculated in frequentist setting).