Clustered & analytical cluster-robust inference (multi-period) + BMisc deprecation fix - #35
Merged
Merged
Conversation
`BMisc::getListElement` is deprecated in BMisc (>= 1.4.9). Replaced the remaining calls with `BMisc::get_list_element` in `process_attgt()` and the aggregation routines. Follow-up to #34; eliminates deprecation warnings. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Calling `ddd(cluster = <var>, boot = FALSE)` in the multiple-period path now returns analytic cluster-robust standard errors (cluster-sum CRVE on the influence function) in `att_gt()` instead of forcing the bootstrap. The `ddd` object now carries `cluster_vector` and `cluster_var` so downstream aggregation can reuse the aligned cluster assignment. Removed the force-bootstrap guard in `run_preprocess_multPeriods()` that previously made the analytic path unreachable. Two-period paths are unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…mark 10) `mboot()` now applies one multiplier per cluster to the influence function aggregated to cluster *sums* (rather than cluster *means*), with SE scaling `bSigma * sqrt(n_clusters) / n`, following Callaway & Sant'Anna (2021, Remark 10). Equal-sized clusters are unaffected; clustered bootstrap standard errors change for unbalanced clusters and repeated cross-sections. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
`compute_se_agg()` gains a cluster-sum branch that computes analytic cluster-robust standard errors on the aggregated influence function (matching the cluster-sum bootstrap). `compute_aggregation()` resolves the aligned per-unit cluster vector stored on the `ddd` object and, on a cluster-variable mismatch (or an object built without clustering), warns and falls back to i.i.d. standard errors for both the analytic and bootstrap paths instead of silently mis-reporting. `agg_ddd()` gains a `cluster` argument. Regenerated man/agg_ddd.Rd and man/compute_se_agg.Rd. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds tests covering the new clustered-inference paths: - test-cluster-analytic.R: analytic cluster SEs equal cluster-sum bootstrap, manual cluster-sum CRVE agreement, equal-size cluster invariance, byte-identical results when no cluster is supplied, and NYT/GMM cells. - test-mboot-cluster.R: cluster-sum multiplier bootstrap behavior. - test-aggte-clustervars-override.R: warn/fallback on cluster-var mismatch and row-permutation alignment of the cluster vector. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
… alignment Add two test files that close mutation-testing gaps in the existing clustered-inference suite: - test-cluster-unbalanced-bootstrap.R: exercises mboot() on UNBALANCED clusters (10 tiny + 10 large), where cluster sums != cluster means. Pins the SE to the cluster-SUM definition (Remark 10) and asserts it diverges materially from the old cluster-MEAN form. The existing equal-sized test cannot catch a full Feature 2 revert (means + old scaling are algebraically identical when clusters are equal-sized); this file kills that surviving mutant. Also checks analytic ~ bootstrap end-to-end on unbalanced data. - test-cluster-alignment-correctness.R: pins att_gt()'s cluster_vector and the reported SE against an INDEPENDENT ground-truth partition (the known id->cl map joined onto first_period_dta$id), rather than the object's own cluster_vector. Catches a wrong-but-stable cluster mapping that the prior shuffle-INVARIANCE check cannot distinguish. Runs in R CMD check. Verified by mutation testing: reverting cluster sums->means, reverting the SE scaling, and applying a stable cluster-vector permutation each fail these tests while passing on the current implementation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…emoved The clustered-inference tests still described run_preprocess_multPeriods()'s force-boot guard as a present "known bug" with checks "SKIPPED". That override was removed, so the analytic cluster path is reached directly and those checks run as regression guards. Update the comments to match; no assertion changes. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The analytic cluster-robust SE, agg_ddd() cluster argument, and Remark 10 bootstrap behavior change are a distinct feature batch from the 0.2.2 BMisc deprecation cleanup, so they move under a new 0.2.3 NEWS section and the package version is bumped to 0.2.3. (0.2.2 was unreleased; CRAN has 0.2.0.) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
The pkgdown site renders the committed README.md (it does not re-knit README.Rmd), so the devel-version badge has to be updated directly to track the 0.2.3 version bump. README.Rmd's badge is generated by badger::badge_devel() and will produce the same value once it is re-knit against the 0.2.3 DESCRIPTION. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Ports three clustered / cluster-robust inference features from
bcallaway11/did(PRs #261/#262) into the multi-period path, plus a BMisc deprecation cleanup. Bumps the dev version to 0.2.3.Point estimates are unchanged everywhere, and with
cluster = NULLstandard errors are byte-identical to before (the new branches are gated on a non-null, row-aligned cluster vector).What's included
Feature 1 — analytic cluster-robust SEs without the bootstrap (multi-period).
ddd(..., cluster = <var>, boot = FALSE)now returns analytical cluster-robust SEs (cluster-sum CRVE on the influence function) inatt_gt.R, instead of silently requiring the bootstrap. Thedddobject gainscluster_vector(per-unit clusters aligned toinf_func_matrows) andcluster_var. Not-yet-treated / GMM cells take their cluster-robust SE from the same cluster-sum V (the per-cellgmm_seoverride is skipped under clustering). Includes removal of the multi-periodrun_preprocess_multPeriods()guard that forcedboot = TRUEunder clustering (which otherwise made this path dead code). Two-period paths (att_dr.R,att_dr_rc.R) are intentionally unchanged and still requireboot = TRUEfor clustered inference.Feature 2 — Remark 10 cluster-sum multiplier bootstrap.
mboot()now applies one multiplier per cluster to the influence function aggregated to cluster sums (not means), with SE scalingbSigma * sqrt(n_clusters) / n. Equal-sized clusters are unaffected; clustered bootstrap SEs change for unbalanced clusters and repeated cross-sections (flagged in NEWS).Feature 3 — clustered SEs in
agg_ddd()aggregations.compute_se_agg()gains a cluster-sum branch;compute_aggregation()resolves the aligned cluster vector and warns + falls back to i.i.d. when clustering is requested on a variableddd()did not cluster on (or on an unclustered object).agg_ddd()gains aclusterargument.Deprecation cleanup.
Replaces the remaining deprecated
BMisc::getListElementwithBMisc::get_list_element(follow-up to #34). Combined with the already-presentrhs_vars/make_balanced_panel, the package is now free of all BMisc deprecation warnings — clearing the CRANexamplesWARN for the next submission.Tests & validation
cluster = NULLbyte-identical, NYT/GMM cells, warn/fallback, and row-permutation alignment.R CMD check: 0 errors / 0 warnings / 0 notes. Test suite: 175 passing, 0 fail, 0 skip.Versioning / docs
0.2.2 -> 0.2.3; clustered-inference NEWS grouped under# triplediff 0.2.3(BMisc deprecation bullets remain under 0.2.2).README.mddevel-version badge updated to 0.2.3 (the pkgdown site renders the committed README; navbar version tracks DESCRIPTION). The live site refreshes when this merges tomain.