Bias of Maximum-Likelihood estimates in logistic and Cox regression models: A comparative simulation study

Parameter estimates of logistic and Cox regression models are
biased for finite samples. In a simulation study we investigated
for both models the behaviour of the bias in relation to sample
size and further parameters. In the case of a dichotomous
explanatory variable x the magnitude of the bias is strongly
influenced by the baseline risk defined by the constants of the
models and the risk resulting for the high risk group. To conduct a
direct comparison of the bias of the two models analyses were based
on the same simulated data. Overall, the bias of the two models
appear to be similar, however, the Cox model has less bias in
situations where the baseline risk is high.