researchzy-causal-inferencelisted
Install: claude install-skill Researchzy/Researchzy-academic-skills
# Researchzy Causal Inference
This skill estimates causal effects with quasi-experimental and econometric designs: difference-in-differences (2×2, staggered adoption, event studies), instrumental variables, regression discontinuity, panel fixed effects, basic synthetic control, and propensity-score weighting. For each design it names the estimand and the identifying assumption first, reads the shape of the data before any result, runs a modern estimator as the headline with the traditional one beside it, runs the diagnostics the design needs, clusters at the level where treatment was assigned, records every specification and diagnostic in `causal/results.json`, and checks the text against that file.
Every request ends in one outcome: `matched` (a design fits), `ambiguous` (ask the user), `not_identifiable_yet` (say what is missing) or `not_attempted` (outside this skill; say why). A failed diagnostic is a result too: report it, do not work around it.
It does not: call an effect causal without its identifying assumption written down; run double machine learning, causal forests, mediation, shift-share or judge-leniency designs, nonlinear or continuous-treatment DiD, treatments that switch on and off, synthetic DiD or augmented synthetic control, RD kinks or several cutoffs, Bayesian or structural models; use MCP computation engines; run Stata or R (it can list the equivalent commands); design or register a trial. Sample size and power go to `researchzy-statistics`, figure st