Fig. 2

Imputation methods ranked from lowest (1) to highest (9 or 10) metrics’ values when introducing missing data not at random in A. LEMS at outcome considered at week 52. B LEMS at outcome considered at week 26. C LEMS at baseline. For each subset (n = 500), missing data is introduced and imputed using all methods. Within each subset, imputation performance is compared between imputation methods and ranked from best performance (i.e., closest to 0 and ranked 1) to lowest performance (i.e., highest metric value and ranked 9 or 10). We display the proportion of subsets (out of 500) per rank and imputation method; LEMS: lower extremity motor score, MAE: mean absolute error, RMSE: root mean squared error, SVM: support vector machines, RBF: radial basis function