Table 3:

Gaussian data with n = 200 and p = 1001 under MAR. Approximately 40% features and 90% cases contain missing values. Linear RR and MICE are not included due to running out of RAM. Detailed simulation setup information is in appendix.

Models	Style	Time(s)	Imp MSE	Bias(${\widehat{\beta }}_1$)	CR(${\widehat{\beta }}_1$)	SE(${\widehat{\beta }}_1$)	SD(${\widehat{\beta }}_1$)
SoftImpute	SI	30.1	0.0442	−0.2862	0.50	0.1583	0.2019
GAIN	SI	111.1	0.7383	0.6897	0.18	0.1697	0.5693
MIWAE	SI	52.5	0.1228	0.5885	0.15	0.1793	0.2162
Sinkhorn	MI	39.3	0.1031	0.6647	0.26	0.2643	0.2195
MI-NNGP1	MI	4.9	0.0119	0.0351	0.89	0.1194	0.1422
MI-NNGP1-BS	MI	4.9	0.0168	0.0383	0.94	0.1424	0.1416
MI-NNGP2	MI	5.8	0.0086	0.0487	0.90	0.1212	0.1343
MI-NNGP2-BS	MI	13.9	0.0092	0.0347	0.93	0.1257	0.1289
Complete data	-	-	-	0.0350	0.94	0.1122	0.1173
Complete case	-	-	-	0.2804	0.76	0.3466	0.4211
ColMean Imp	SI	-	0.1130	0.7024	0.13	0.2574	0.1957