Significantly improved the speed and memory usage of analytic clustered standard errors by aggregating score contributions at the cluster level before constructing the influence function.
First-stage fixed effects nested within the clustering variable are now detected and omitted from the variance correction. The correction solve is skipped entirely when all first-stage fixed effects are nested.
Added a matrix_free covariance backend for models with high-cardinality crossed fixed effects, with vcov_method, vcov_tol, and vcov_maxiter arguments to control the solver.
Added support for estimation weights in event_study() for supported estimators.
Improved the speed and robustness of sparse model-matrix construction, including better handling of collinear fixed effects.
Updated did2s() for compatibility with newer versions of fixest.
honest_did function to compute a sensitivity analysis using the approach of Rambachan and Roth (2021)ARMA_64_BIT for larger matriciesestimator argument to event_study and added vignette discussing the different estimators' assumptionsplot_event_study, which may break some previous code (https://github.com/kylebutts/did2s/pull/15/)Made code more robust by improving sparse model matrix creation
Rewrote covariance matrix creation to be faster and more robust
Improved block bootstrapping code using boot::boot
Fixed Sun and Abraham bug in event_study, h/t Frederik Noack
Add event_study and plot_event_study functions
Seperated did_imputation to new package didimputation
Add did_imputation following Borusyak, Jaravel, and Spiess (2021)
Speed up did2s
Use sparse matrices to use much less memory
Fix analytic standard errors when using weights