Selection into treatment means you are going to have a hard time with causal inference unless you have an instrumental variable - ThreadSky

About ThreadSky

stephenjwild.bsky.social • 5 days ago

Selection into treatment means you are going to have a hard time with causal inference unless you have an instrumental variable

Comments

solomonkurz.bsky.social•5 days ago

I was planning on using inverse propensity score weighting with a deep covariate set using {WeightIt} and {marginaleffects}

jonhuang.bsky.social•5 days ago

I think there some subtleties in your question. 1) extent to which duration is built into your tx & target causal contrast (i.e. is it 1 pill / 1 therapy session?) & 2) how you are/n't using repeated outcome scores -- but generally if duration is a true consequence of your treatment, don't adjust

jonhuang.bsky.social•5 days ago

E.g. if a treatment decreases # clinic visits, that is part of the effect.
Regarding 2, there are some decision wrt correlated measurement errors & mediation beautifully explained via augmented DAGs in this paper: https://journals.sagepub.com/doi/abs/10.1177/0049124119826155

jonhuang.bsky.social•5 days ago

Back to (1). If duration is truly not important itself and it is only a matter of repeated exposures (and outcomes), you might consider a pooled structural model with lagged treatments and outcomes (for which IPW-ted marginal structural models were explicitly designed)

jfeltphd.bsky.social•5 days ago

Will shadish had a great paper on this where participants were randomized to choose their condition or to be randomly assigned to condition https://www.statpower.net/Content/MLRM/Readings/Shadish-JASA2008(1334-1356)-17mr09.pdf

Posting Rules

Be respectful to others
No spam or self-promotion
Stay on topic
Follow Bluesky's terms of service

Comments

Posting Rules

Reply