Figure 1
Discrimination reflected in proportions of random allocations vs. allocations to the favored and disfavored patient by patient feature. The right label corresponds to the patient who we predicted to be favored, and the left label to the patient who we predicted to be disfavored. All four feature clusters indicate that allocations varied between patient features. Asterisks on random allocations indicate that random allocations statistically deviated from 50%, reflecting a lack of impartiality, and asterisks on favored allocations indicate a higher likelihood for favored than for disfavored allocations among non-random allocations, reflecting systematic bias towards one patient (binomial tests against 50%, first asterisk uncorrected, second asterisk Bonferroni corrected for 19 tests). The offset of bars from the center indicate the degree of bias as the absolute difference between favored and disfavored allocations (in percent) and odds ratios (to the right of the bars) provide the effect size of each bias.
