Causal inference in R

[emilyriederer]

Opening a general issue on this idea as an area for discussion. I believe numerous participants expressed interest or had experience in this field. Clearly, projects here could take a lot of different shapes, but let's use this thread to brainstorm further!

[benlistyg]

One idea I have for this is to contribute to the DeclareDesign package from Blair et al. They could use a vignette for a Two-Arm Covariate Design (see here).

[emilyriederer]

Neat idea! They have such rich documentation in most places; anything to make it even better would be a real win

[weiyangtham]

A general thought I've had is that it would be nice to shepherd some causal inference packages into the tidyverse, mainly by making the output extract-able from broom (alternatively contributing a method to DeclareDesign). A few off the top of my head:

• rdrobust for regression discontinuity
• cem for Coarsened Exact Matching
• gsynth and synth for synthetic control

[alexpghayes]

As someone just getting more into causal inference I'd also love to see an overview of packages for causal inference in R. I feel like many methods are siloed in packages maintained by individual research groups that don't talk to each other, so it'd be good to know:

• what methods are implemented
• where do they live
• which could possibly be unified / made consistent with each other

I'm especially curious about the approach in DoWhy, where identification and estimation are separate steps in a modeling workflow.

[emilyriederer]

@weiyangtham This is a great idea! I know the cobalt package tries to do something sort of similar to broom by just providing standardized outputs from many different packages (mostly in the more "classical" weighting / matching space).

It's not in the "tidy" paradigm, so we could either use it for inspiration for making new additions to broom or also consider making a PR to cobalt (authors willing) to include techniques beyond weighting / matching?

[emilyriederer]

@alexpghayes Wow - I had ne ver heard of DoWhy before at all! Thanks for sharing! This looks really interesting 😎

[weiyangtham]

@alexpghayes Yes! Are you thinking of something like a causal inference CRAN task view? DoWhy looks super cool and in the long-term a unifying framework would be ideal

[weiyangtham]

@emilyriederer I hadn't known of cobalt but what you said definitely seems like the right direction! Will play around with it a bit