,
10 tweets,
7 min read
Read on Twitter
Hey everyone! Here’s a #tweetorial on our new paper on why we often can’t make #causalinference using #distancetocare as an exposure or instrument! cc: @epiellie
There are 3 problems with #distancetocare as an exposure for estimating causal effects. Let’s walk through them.
The first problem that probably comes to mind is #confounding 🙀Choices about where to live and where to locate care facilities are complicated and depend on a lot of things we might not be able to measure (e.g. socioeconomic status).
The 2nd problem is more sneaky! Because people who live farther from care are more difficult to recruit into your study, #distancetocare likely has a #selectionbias problem. Even worse, those you do recruit could have poorer outcomes & that’s why they show up. Here’s the DAG! 
The third problem is identifying a sufficiently well-defined causal question, aka #consistency. There are multiple possible ways someone could increase or decrease their #distancetocare – building more roads, building more hospitals, and moving are just three.
Even if you don’t define and specify these interventions, you end up taking a weighted average over many interventions and this may not be generalizable. Here’s the DAG!
Even worse, when your causal question is not well-defined, your confounding is not well-defined. If we don’t know how to intervene on distance, we don’t know how to control for confounding for those interventions!
Now that we’ve discussed #distancetocare as an exposure, what are the problems with #distancetocare as an instrument? It turns out they’re basically the same!
#Distancetocare fails on exogeneity b/c it’s probably confounded & on the exclusion restriction b/c it probably has selection bias. Plus, an ill-defined intervention on #distancetocare can make it difficult to interpret the local average treatment effect. Here are some DAGs!
So is it hopeless? No! Even if we can’t confidently answer a clear question, we can use #sensitivityanalyses to estimate plausible values & help us make decisions about #distancetocare.
Check out our #rstats shinyapp to play around with them yourself! emurray.shinyapps.io/distanceApp/
Check out our #rstats shinyapp to play around with them yourself! emurray.shinyapps.io/distanceApp/
