logistf (Q19163)

Fit a logistic regression model using Firth's bias reduction method, equivalent to penalization of the log-likelihood by the Jeffreys prior. Confidence intervals for regression coefficients can be computed by penalized profile likelihood. Firth's method was proposed as idealsolution to the problem of separation in logistic regression, see Heinze and Schemper (2002) <doi:10.1002/sim.1047>. If needed, the bias reduction can be turned off such that ordinarymaximum likelihood logistic regression is obtained. Two new modifications of Firth's method, FLIC and FLAC, lead to unbiased predictions and are now availablein the package as well, see Puhr et al (2017) <doi:10.1002/sim.7273>.