For me, 2024 was the year of Synthesis Estimators. Synthesis is work that came to fruition from my interest in merging ideas from ‘causal inference’ and ‘mathematical modeling’ throughout my PhD. Here are some highlights to close out the year
Comments
Log in with your Bluesky account to leave a comment
The first paper on using a mathematical model to fill in nonpositive regions was published in Epidemiology in the January issue. This paper lays out the basics in the case of nonpositivity by a binary variable and we proposed IPW and g-computation estimators
This work was recognized at the annual SER meeting, where I won the Lilienfeld Postdoctoral Paper Award Winner for this paper (completed and accepted while I was still a postdoc)
One of the reviewers of the paper asked whether you could use the proposed approach for nonpositivity by continuous covariates. As part of the revisions, I sketched out how it would work in the appendix and wrote up some simulations. My coauthors wisely suggested we make it a separate paper
That led to this paper which was accepted in JRSSA. In it, I make the Epidemiology paper a bit more formal, extend to continuous covariates, proposed two new AIPW estimators, and provide a new illustrative example
I’ve also been fortunate enough to present on this work a few times. I’ve test run different ways to present these ideas and how to express them verbally and pictorially. So, in case you’re looking for guest speakers in 2025…
I have some drafts in-progress using these ideas for other systematic errors or extending these tools for other data structures. I’m looking forward to sharing them with you all in 2025
Comments
https://journals.lww.com/epidem/abstract/2024/01000/transportability_without_positivity__a_synthesis.4.aspx
https://academic.oup.com/jrsssa/advance-article-abstract/doi/10.1093/jrsssa/qnae084/7747432