Table 12:

Gaussian and binary data with n = 200 and p = 1002 under MNAR. Approximately 40% features and 88% cases contain missing values. Here Linear RR and MICE are not included due to running out of RAM.

Models	Style	Time(s)	Imp MSE	Imp accu	Bias(${\widehat{\beta }}_1$)	CR(${\widehat{\beta }}_1$)	SE(${\widehat{\beta }}_1$)	SD(${\widehat{\beta }}_1$)
SoftImpute	SI	24.7	0.0432	0.3328	−0.2771	0.64	0.2154	0.2263
GAIN	SI	95.3	0.8517	0.3328	0.9779	0.12	0.1605	0.7701
MIWAE	SI	57.7	0.1258	0.6706	0.6665	0.15	0.1604	0.2274
not-MIWAE	SI	42.7	1.9305	0.3317	0.6158	0.06	0.0851	0.1876
Sinkhorn	MI	43.6	0.1080	0.3273	0.4175	0.72	0.3033	0.2882
MI-NNGP1	MI	3.6	0.0117	0.6477	−0.0137	0.94	0.1492	0.1514
MI-NNGP1-BS	MI	3.7	0.0146	0.6201	−0.0154	0.98	0.1785	0.1423
MI-NNGP2	MI	4.2	0.0089	0.7277	0.0125	0.99	0.1469	0.1238
MI-NNGP2-BS	MI	10.0	0.0093	0.6971	−0.0044	0.98	0.1558	0.1243
Complete data	-	-	-	-	0.0156	0.96	0.1119	0.1041
Complete case	-	-	-	0.3909	0.70	-	0.2854	0.3001
ColMean Imp	SI	-	0.1259	0.3273	0.4620	0.71	0.2996	0.2367