new est_method inputs & outputs

[thomaswiemann]

Summary

This PR extends the custom est_method interface with three opt-in features. No breaking changes---all defaults are NULL and built-in methods ("dr", "ipw", "reg") are unaffected.

• est_method_vars. Character vector of column names from data to pass through to est_method as a data argument, subsetted to each (g,t) cell.

• Cell identity forwarding. If a custom est_method has formal arguments g and t, the current group and time period are automatically passed, allowing the estimator to identify which (g,t) cell it belongs to.

• extra_gt. Any extra fields returned by a custom est_method (beyond ATT and att.inf.func) are captured per cell and accessible via result$extra_gt. Useful for storing model fits for diagnostics or re-estimation.

These additions facilitate broader support of custom ATT estimators including---but not exclusively---using double/debiased machine learning as in Chang (2020). (See also Ahrens et al. (2026) for a staggered DiD example with DML.) In particular, est_method_vars + passing along g and t enables use of pre-assigned fold IDs for cross-fitting across group-time cells, correcting errors in existing DiD cross-fitting implementations (see, e.g., here). extra_gt allows nuisance-function estimation diagnostics and passing through pre-fitted machine learning estimators.

The new features are illustrated in a DML-based DiD vignette. The vignette is not included in this PR.

AI Disclosure

Portions of this code, documentation, and tests were developed with the assistance of AI tools. I reviewed, tested, and edited all AI-generated content. All errors are my own.

References

Ahrens A, Chernozhukov V, Hansen C B, Kozbur D, Schaffer M E, Wiemann T (2026). "An Introduction to Double/Debiased Machine Learning." Journal of Economic Literature, forthcoming.

Chang NC (2020). "Double/debiased machine learning for difference-in-differences models." The Econometrics Journal, 177–191.