Causal Method Picker
Answer a few questions about your situation; get a recommendation from my working causal-inference playbook.
Which causal method fits your situation?
Answer a few questions. I'll recommend one from my working playbook.
How it works
What it computes
A fixed decision tree, not a model. It walks through your design and returns the first method whose preconditions you meet: randomization first; with a clear launch moment it goes to synthetic control (many donor units), DiD (a few comparable groups), or interrupted time series (none); with no launch moment but a sharp score cutoff, RDD; otherwise propensity score matching or weighting.
The ordering encodes a practical priority: run the design you control if you can, then lean on the structure your setting already has (a rollout moment, a cutoff), and match on observables only as a last resort.
Assumptions
- Each recommendation still carries its own assumptions: parallel trends for DiD, no manipulation at the cutoff for RDD, donor quality and pre-period fit for synthetic control. The picker sequences the questions; it does not verify your answers.
Where it breaks
- Real situations often satisfy several branches at once. The tree picks one by priority instead of listing every viable method.
- The fallback, propensity methods, requires no unobserved confounding, which is exactly the assumption you can never test. Treat that recommendation as the start of an argument, not the end.