Skip to main content

Table 6 Simulation results for scenario G, n = 500, 25%missing

From: Comparison of methods for handling covariate missingness in propensity score estimation with a binary exposure

Missingness MechanismMethodTrue confoundersLeave ×1 outAdd ×5Leave ×1 out + Add ×5
Bias (SD)SERMSEBias (SD)SERMSEBias (SD)SERMSEBias (SD)SERMSE
 complete−0.014 (0.044)0.0710.0460.037 (0.043)0.0680.057−0.016 (0.045)0.0730.0480.040 (0.043)0.0690.059
comGBM0.029 (0.044)0.0670.053
MCARSI + pe + pu−0.011 (0.052)0.0710.0530.040 (0.049)0.0680.063−0.012 (0.054)0.0720.0550.043 (0.050)0.0680.066
SI + pe−0.013 (0.051)0.0710.0530.039 (0.049)0.0680.063−0.014 (0.053)0.0720.0550.042 (0.049)0.0680.065
TMI0.096 (0.064)0.0760.1150.131 (0.061)0.0730.1450.096 (0.067)0.0780.1170.134 (0.063)0.0740.148
MI−0.011 (0.048)0.0710.0490.039 (0.046)0.0680.060−0.013 (0.050)0.0730.0520.042 (0.047)0.0690.063
MIMP−0.011 (0.048)0.0720.0490.039 (0.046)0.0680.060−0.013 (0.050)0.0730.0520.042 (0.047)0.0690.063
GBM0.060 (0.051)0.0670.079
GBM + SI + pe0.033 (0.049)0.0670.059
MAR1SI + pe + pu0.000 (0.051)0.0700.0510.048 (0.048)0.0670.068−0.002 (0.053)0.0720.0530.052 (0.049)0.0680.071
SI + pe−0.001 (0.050)0.0700.0500.048 (0.047)0.0670.067−0.003 (0.052)0.0720.0520.051 (0.048)0.0680.070
TMI0.113 (0.057)0.0730.1270.148 (0.055)0.0700.1580.116 (0.059)0.0740.1300.152 (0.056)0.0710.162
MI−0.001 (0.047)0.0710.0470.048 (0.045)0.0680.066−0.002 (0.048)0.0720.0480.051 (0.046)0.0680.069
MIMP−0.001 (0.047)0.0710.0470.048 (0.045)0.0680.066−0.002 (0.048)0.0720.0480.052 (0.046)0.0690.069
GBM0.066 (0.050)0.0670.083
GBM + SI + pe0.039 (0.047)0.0670.061
MAR2SI + pe + pu− 0.009 (0.051)0.0700.0520.040 (0.050)0.0680.064−0.011 (0.053)0.0720.0540.043 (0.050)0.0680.066
SI + pe−0.009 (0.051)0.0700.0520.041 (0.049)0.0680.064−0.010 (0.053)0.0720.0540.045 (0.050)0.0680.067
TMI0.090 (0.069)0.0780.1130.125 (0.066)0.0740.1410.093 (0.071)0.0800.1170.128 (0.067)0.0760.144
MI−0.008 (0.046)0.0710.0470.041 (0.046)0.0680.062−0.010 (0.048)0.0730.0490.044 (0.046)0.0690.064
MIMP−0.008 (0.047)0.0720.0480.042 (0.046)0.0680.062−0.010 (0.049)0.0730.0500.045 (0.047)0.0690.065
GBM0.058 (0.051)0.0670.077
GBM + SI + pe0.036 (0.048)0.0670.060
MAR SinisterSI + pe + pu−0.006 (0.050)0.0700.0500.040 (0.048)0.0670.062−0.007 (0.052)0.0710.0520.042 (0.048)0.0680.064
SI + pe−0.012 (0.051)0.0710.0520.039 (0.049)0.0680.063−0.014 (0.053)0.0720.0550.042 (0.050)0.0690.065
TMI0.113 (0.060)0.0750.1280.147 (0.057)0.0720.1580.113 (0.062)0.0770.1290.151 (0.059)0.0730.162
MI− 0.007 (0.047)0.0710.0480.043 (0.045)0.0680.062−0.009 (0.048)0.0720.0490.046 (0.046)0.0690.065
MIMP−0.007 (0.047)0.0710.0480.043 (0.046)0.0680.063−0.009 (0.049)0.0730.0500.046 (0.046)0.0690.065
GBM0.068 (0.050)0.0670.084
GBM + SI + pe0.035 (0.048)0.0670.059
  1. Note. Complete: logistic regression with complete data before introducing missingness; comGBM GBM with complete data before introducing missingness; SI + pe + pu single imputation + prediction error + parameter uncertainty; SI + pe single imputation + prediction error; TMI treatment mean imputation; MI multiple imputation (m = 20); MIMP multiple imputation missingness pattern (m = 20); GBM GBM with incomplete data; GBM + SI + pe GBM after single imputation + prediction error; SD standard deviation; SE standard error; RMSE root mean squared error