pzivich/zEpid

TMLE improved stability

pzivich opened this issue · 0 comments

My TMLE code currently copies R's tmle and this approach uses as older (but valid) version. A more stable version instead uses HA1W/HA0W as weight in the targeting step (rather than including as terms in the model).

This approach is suppose to increase stability. It is potentially worth looking into in a future update. Maybe allow for both procedures?