Skip to main content

Table 1 Estimation of the true effect in the simulated datasets using three different methods under the MCAR mechanism

From: Propensity score analysis with missing data using a multi-task neural network

Missing rate

Method

True effect =0

True effect =1

Mean

SD

RMSE

Mean

SD

RMSE

0.2

Missing indicator

0.119

0.059

0.131

1.119

0.059

1.12

Multiple imputation

0.11

0.071

0.128

1.11

0.071

1.112

Multi-task neural network

0.091

0.052

0.103

1.077

0.055

1.079

0.3

Missing indicator

0.146

0.063

0.157

1.146

0.063

1.147

Multiple imputation

0.136

0.076

0.153

1.136

0.076

1.138

Multi-task neural network

0.107

0.057

0.12

1.12

0.052

1.121

0.4

Missing indicator

0.173

0.061

0.182

1.173

0.061

1.174

Multiple imputation

0.192

0.089

0.209

1.192

0.089

1.195

Multi-task neural network

0.138

0.05

0.146

1.139

0.058

1.141

0.5

Missing indicator

0.206

0.069

0.216

1.206

0.069

1.207

Multiple imputation

0.214

0.078

0.226

1.214

0.078

1.216

Multi-task neural network

0.173

0.058

0.181

1.169

0.06

1.17

0.6

Missing indicator

0.234

0.071

0.243

1.234

0.071

1.236

Multiple imputation

0.228

0.076

0.239

1.228

0.076

1.23

Multi-task neural network

0.2

0.064

0.208

1.198

0.06

1.199

0.7

Missing indicator

0.242

0.081

0.254

1.242

0.081

1.244

Multiple imputation

0.248

0.08

0.259

1.248

0.08

1.251

Multi-task neural network

0.207

0.078

0.219

1.204

0.071

1.206

0.8

Missing indicator

0.258

0.061

0.264

1.258

0.061

1.259

Multiple imputation

0.26

0.074

0.269

1.26

0.074

1.262

Multi-task neural network

0.226

0.053

0.232

1.224

0.057

1.225

  1. SD, standard deviation; RMSE, root mean square error; MCAR, missing completely at random; MAR, missing at random; MNAR, missing not at random