@joshcohen @BRENDABLUEWAVE @NitekryDPaul @TheChiefNerd 4) Uses Firth penalized regression models which is HIGHLY inaccurate, especially in small sample sizes: https://t.co/MX11i8bQ5J 5) They literally excluded 1/5 of originally-eligible patient who were f
437 followers
1,267 followers
RT @Epi_D_Nique: #StatsTwitter - In small/sparse data it seems Firth’s logistic is a good choice to help handle sparse data bias/quasi/full…
913 followers
RT @Epi_D_Nique: #StatsTwitter - In small/sparse data it seems Firth’s logistic is a good choice to help handle sparse data bias/quasi/full…
6,056 followers
#StatsTwitter - In small/sparse data it seems Firth’s logistic is a good choice to help handle sparse data bias/quasi/full separation. But this paper suggests in some cases Firth’s can *increase* bias. Does anyone have a feel for when that is the case?