Causal Econometrics
Purpose
Estimate design-based associations and causal estimands when the researcher supplies a coherent design. Software does not prove parallel trends, exclusion restrictions, or unconfoundedness.
Supported method families (implemented)
- OLS / fixed effects / robust inference paths in Causal Lab.
- 2SLS (IV), difference-in-differences, event study, sharp RDD.
- Propensity-score matching with observed-covariate balance diagnostics.
- Partial-linear double machine learning with cross-fitting for the implemented learners.
- Experimental HTE spline and experimental causal forest (not general certified CATE).
Not implemented as a general catalogue
- Meta-learners (T/S/X/R) as a runnable catalogue
- Synthetic control
Inputs and outputs
Provide outcome, treatment/exposure, covariates, and design-specific fields (instruments, cutoff, timing). Outputs include estimates, design-specific sandwiches/gates, and claim labeling such as causal_under_assumptions when applicable.
Reproducibility
Notebook attachments are typically stored summaries. Native EXACT MATCH rerun is not claimed for general Causal Lab links — expect UNSUPPORTED METHOD unless a specific native path exists.
Common mistakes
- Treating DML or matching as automatic identification.
- Omitting design assumptions in the notebook narrative.
- Confusing reduced-form coefficients with structural parameters.