Skip to main content
. Author manuscript; available in PMC: 2017 Oct 23.
Published in final edited form as: Stat Methods Med Res. 2017 Feb 2;26(5):2455–2480. doi: 10.1177/0962280216688034

Table 2.

The estimated total integrated absolute bias and RMSE values for each of the models and each of the measures.

Baseline
Hazard rate
Treatment
Hazard rate
Treatment (NPH)
Log-hazard ratio
PSA (PH)
Log-hazard ratio
Model Bias (RMSE) Bias (RMSE) Bias (RMSE) Bias (RMSE)
Aalen’s additive 0.273 (0.097) 0.755 (0.258) 14.950 (5.646) −0.129 (0.336)
HARE
 (All) 2.2 × 1036 4.0 × 1037 (∞) (∞) −0.015 (0.357)
 (Subset) 0.127 (0.0624) 0.186 (0.076)   3.679 (1.543)   0.012 (0.126)
NPMRH
 (0) 0.199 (0.089) 0.569 (0.234)   8.977 (3.779) −0.008 (0.106)
 (2) 0.177 (0.079) 0.376 (0.172)   6.880 (3.041) −0.008 (0.107)
 (5, k sampled) 0.193 (0.086) 0.289 (0.158)   6.949 (3.108) −0.007 (0.106)
 (5, k = 0.5) 0.192 (0.085) 0.278 (0.149)   6.709 (2.931) −0.007 (0.106)
 (5, k = 8) 0.397 (0.141) 0.105 (0.043) 10.008 (3.758) −0.018 (0.109)
 (Lowest AIC) 0.193 (0.085) 0.281 (0.151)   6.763 (2.973) −0.007 (0.106)
Time-varying Cox 0.376 (0.131) 0.215 (0.115) 10.571 (4.164)   0.005 (0.106)

Note: Among the NPMRH models, the models based on the lowest AIC tend to have the lowest values, although there are exceptions, such as the estimated bias and RMSE for the NPMRH-5 model with k = 8: The other conclusions match those observed in Figure 9.