v2.4.5 release: dev → master (bundles v2.4.2–v2.4.5, live on CRAN 2026-05-30)

.Rbuildignore

@@ -18,4 +18,5 @@
 ^data-raw$
 ^cell-D1\.rds$
 ^fect_.*\.tar\.gz$
-^fect\.Rcheck$
+^fect\.Rcheck$
+^tests/coverage-study$

.gitignore

@@ -29,6 +29,9 @@ vignettes/*.html
 vignettes/*.pdf
 vignettes/*_cache/
 vignettes/*_files/
+vignettes/*.rmarkdown
+vignettes/site_libs/
+vignettes/.quarto/
 vignettes/_back/**
 vignettes/_book/**
 !vignettes/_book/rscript/

ARCHITECTURE.md

@@ -1,10 +1,14 @@
 # Architecture — fect
 
 > Generated by scriber for run `2026-04-30-architecture-regen-v241` on 2026-04-30.
+> Manual v2.4.2 patches applied 2026-05-01 (warm-start C++ infrastructure;
+> ci.method extensions) and 2026-05-02 (CFE `fit_init` plumbing on
+> `complex_fe_ub` + `cfe_iter`, dormant). Full regen via scriber
+> deferred until next material module change.
 
 ## Overview
 
-fect is an R package for estimating causal effects in panel data using counterfactual imputation methods (Fixed Effects Counterfactual Estimators). It targets causal panel analysis with binary treatments under the parallel trends assumption, supporting treatment switching and limited carryover effects. The core abstraction is counterfactual imputation: impute missing potential outcomes Y(0) for treated units using control units, then compute the Average Treatment Effect on the Treated (ATT) as the gap between observed and imputed outcomes. The package is an R/C++ hybrid using Rcpp and RcppArmadillo for numerically intensive linear algebra (SVD, EM iterations, matrix factorization). Key external dependencies include fixest (initial FE regression), ggplot2 (visualization), doParallel/doFuture/future.apply (parallel bootstrap), MASS (generalized inverse), and mvtnorm (multivariate normal draws). Estimation methods include FE (fixed effects), IFE (interactive fixed effects / factor model), MC (matrix completion via nuclear norm regularization), CFE (complex fixed effects with structured covariates), and wrappers for modern DID estimators. Version 2.4.1. References: Liu, Wang, and Xu (2024); Chiu et al. (2026).
+fect is an R package for estimating causal effects in panel data using counterfactual imputation methods (Fixed Effects Counterfactual Estimators). It targets causal panel analysis with binary treatments under the parallel trends assumption, supporting treatment switching and limited carryover effects. The core abstraction is counterfactual imputation: impute missing potential outcomes Y(0) for treated units using control units, then compute the Average Treatment Effect on the Treated (ATT) as the gap between observed and imputed outcomes. The package is an R/C++ hybrid using Rcpp and RcppArmadillo for numerically intensive linear algebra (SVD, EM iterations, matrix factorization). Key external dependencies include fixest (initial FE regression), ggplot2 (visualization), doParallel/doFuture/future.apply (parallel bootstrap), MASS (generalized inverse), and mvtnorm (multivariate normal draws). Estimation methods include FE (fixed effects), IFE (interactive fixed effects / factor model), MC (matrix completion via nuclear norm regularization), CFE (complex fixed effects with structured covariates), and wrappers for modern DID estimators. Version 2.4.2. References: Liu, Wang, and Xu (2024); Chiu et al. (2026).
 
 ---
 
@@ -121,7 +125,7 @@ graph TD
 | `R/interFE.R` (515 lines) | API | Standalone interactive fixed effects estimator | `interFE()` | no |
 | `R/did_wrapper.R` (656 lines) | API | Modern DID estimator wrappers (did, DIDmultiplegtDYN) | `did_wrapper()` | no |
 | `R/fect_mspe.R` (370 lines) | API | MSPE computation for model comparison | `fect_mspe()` | no |
-| `R/po-estimands.R` (1,123 lines) | API | Two-tier post-hoc estimand surface: long-form imputed PO accessor and typed dispatcher; `vartype` enum extended to `"parametric"` in v2.4.1 | `estimand()`, `imputed_outcomes()` | yes (new v2.4.0) |
+| `R/po-estimands.R` (1,236 lines) | API | Two-tier post-hoc estimand surface: long-form imputed PO accessor and typed dispatcher; `vartype` enum extended to `"parametric"` in v2.4.1; `ci.method` enum extended to `c("basic","percentile","bc","normal")` with per-type defaults via NULL trigger in v2.4.2; hard-error on bootstrap cell-drop pathology in `log.att`/`aptt` | `estimand()`, `imputed_outcomes()` | yes |
 | `R/fe.R` (955 lines) | Estimation | Interactive Fixed Effects / factor model estimation | `fect_fe()` | no |
 | `R/mc.R` (807 lines) | Estimation | Matrix Completion via nuclear norm regularization | `fect_mc()` | no |
 | `R/cfe.R` (1,173 lines) | Estimation | Complex Fixed Effects with structured covariates | `fect_cfe()` | no |
@@ -155,8 +159,8 @@ graph TD
 | `src/ife.cpp` (534 lines) | C++ Core | IFE algorithm: `inter_fe()`, `inter_fe_ub()`, `inter_fe_d()` | (Rcpp exports) | no |
 | `src/ife_sub.cpp` (577 lines) | C++ Core | IFE sub-routines: SVD factor estimation, EM iterations, alternating minimization | (internal) | no |
 | `src/mc.cpp` (223 lines) | C++ Core | Matrix completion: `inter_fe_mc()`, nuclear norm penalization | (Rcpp exports) | no |
-| `src/cfe.cpp` (203 lines) | C++ Core | Complex FE: `complex_fe_ub()` | (Rcpp exports) | no |
-| `src/cfe_sub.cpp` (564 lines) | C++ Core | Complex FE sub-routines: `cfe_iter()`, structured covariate handling | (internal) | no |
+| `src/cfe.cpp` (203 lines) | C++ Core | Complex FE: `complex_fe_ub()`. v2.4.2: optional `fit_init` matrix to seed EM (dormant; mirrors IFE/MC infra) | (Rcpp exports) | no |
+| `src/cfe_sub.cpp` (564 lines) | C++ Core | Complex FE sub-routines: `cfe_iter()`, structured covariate handling. v2.4.2: optional `fit_init` matrix replaces default `fit = Y0` initialization (dormant) | (internal) | no |
 | `src/fe_sub.cpp` (291 lines) | C++ Core | Shared FE utilities: `Y_demean()`, `panel_beta()`, `panel_factor()`, `panel_FE()`, `XXinv()` | (internal) | no |
 | `src/binary_sub.cpp` (539 lines) | C++ Core | Probit model sub-routines for binary outcomes | (internal) | no |
 | `src/binary_qr.cpp` (347 lines) | C++ Core | QR-based probit estimation | (internal) | no |
@@ -268,8 +272,8 @@ graph TD
 | `fect.default()` | `R/default.R` | `fect.formula()`, user | `fect_cv()`, `fect_fe()`, `fect_mc()`, `fect_cfe()`, `fect_boot()`, `diagtest()` | yes | Workhorse: validation, preprocessing, method routing, inference, diagnostics; added `W.est`/`W.agg`/`carryover.rm` |
 | `fect_fe()` | `R/fe.R` | `fect.default()`, `fect_cv()`, `fect_boot()`, `r.cv.rolling()` | `inter_fe_ub()`, `inter_fe_d_qr_ub()` (C++) | no | IFE estimation (factor model with r latent factors) |
 | `fect_mc()` | `R/mc.R` | `fect.default()`, `fect_cv()`, `fect_boot()` | `inter_fe_mc()` (C++) | no | Matrix completion estimation (nuclear norm regularization) |
-| `fect_cfe()` | `R/cfe.R` | `fect.default()`, `fect_boot()`, `r.cv.rolling()` | `complex_fe_ub()` (C++) | no | Complex FE with structured covariates (Z, Q, gamma, kappa) |
-| `fect_nevertreated()` | `R/fect_nevertreated.R` | `fect.default()` | `fect_fe()`, `fect_mc()`, `fect_cfe()`, `.fect_make_future_cluster()` | yes | Wrapper for never-treated-only estimation sample; PSOCK cluster via shared helper (v2.3.3) |
+| `fect_cfe()` | `R/cfe.R` | `fect.default()`, `fect_boot()`, `r.cv.rolling()` | `complex_fe_ub()` (C++) | no | Complex FE with structured covariates (Z, Q, gamma, kappa); v2.4.2 dormant `fit.init = NULL` parameter threaded through to C++ |
+| `fect_nevertreated()` | `R/fect_nevertreated.R` | `fect.default()` | `fect_fe()`, `fect_mc()`, `fect_cfe()`, `.fect_make_future_cluster()` | yes | Wrapper for never-treated-only estimation sample; PSOCK cluster via shared helper (v2.3.3); v2.4.2 dormant `fit.init = NULL` parameter, sliced to control units |
 | `fect_cv()` | `R/cv.R` | `fect.default()` | `fect_fe()`, `fect_mc()`, `.fect_apply_cv_rule()`, `.fect_cv_aggregate_folds()` | yes | Cross-validation to select r (IFE) or lambda (MC); rolling default (v2.3.0), 1-SE rule (v2.3.0) |
 | `r.cv.rolling()` | `R/cv-rolling.R` | user / exported | `fect_fe()`, `fect_cfe()`, `.fect_apply_cv_rule()`, `.fect_make_future_cluster()` | yes (new v2.3.0) | Standalone rolling-window CV for rank selection; per-fold unit sampling; closes AR forward-leakage |
 | `fect_boot()` | `R/boot.R` | `fect.default()` | `fect_fe()`, `fect_mc()`, `fect_cfe()`, `impute_Y0()`, `valid_controls()`, `.fect_make_future_cluster()` | yes | Bootstrap/jackknife/parametric inference with PSOCK retry-with-backoff (v2.3.3) |
@@ -279,7 +283,7 @@ graph TD
 | `did_wrapper()` | `R/did_wrapper.R` | user / exported | `fixest::feols()`, `did::att_gt()` | no | Modern DID estimator wrappers |
 | `plot.fect()` | `R/plot.R` | user / exported | ggplot2, `.modern_theme()` | yes | 14 plot types; modernized defaults, `legacy.style`/`highlight`/`highlight.fill` (v2.3.2) |
 | `esplot()` | `R/esplot.R` | user / exported | ggplot2, `.modern_theme()` | yes | Standalone event-study plot; modern theme integration (v2.3.2) |
-| `estimand()` | `R/po-estimands.R` | user / exported (new v2.4.0) | `imputed_outcomes()`, internal aggregators | yes (new v2.4.0) | Typed dispatcher: `att`, `att.cumu`, `aptt`, `log.att` across `event.time`/`cohort`/`calendar.time`/`overall`; `vartype` enum extended to `"parametric"` in v2.4.1 |
+| `estimand()` | `R/po-estimands.R` | user / exported (new v2.4.0) | `imputed_outcomes()`, internal aggregators, `.compute_ci()` | yes | Typed dispatcher: `att`, `att.cumu`, `aptt`, `log.att` across `event.time`/`cohort`/`calendar.time`/`overall`; `vartype` enum extended to `"parametric"` in v2.4.1; `ci.method` enum extended to `c("basic","percentile","bc","normal")` with per-type defaults via NULL trigger in v2.4.2 |
 | `imputed_outcomes()` | `R/po-estimands.R` | user / exported (new v2.4.0); `estimand()` | internal helpers | yes (new v2.4.0) | Long-form accessor for cell-level imputed PO surface; `cells =` filter (logical / formula); `direction = c("on", "off")`; `eff_debias` slot support |
 | `effect()` | `R/effect.R` | user / exported | (internal helpers) | yes (soft-deprecated v2.4.0) | Treatment effect decomposition; emits one-time-per-session deprecation message: *"`effect()` is soft-deprecated as of fect 2.4.0; prefer `estimand(fit, "att.cumu", ...)` API."* Removal not before v3.0.0 |
 | `att.cumu()` | `R/cumu.R` | user / exported | (internal helpers) | yes (soft-deprecated v2.4.0, `n_cells` fix v2.4.1) | Cumulative ATT; emits same one-time deprecation; `n_cells` column fixed in v2.4.1 |
@@ -290,7 +294,7 @@ graph TD
 | `diagtest()` | `R/diagtest.R` | `fect.default()` | — | yes | Pre-trend, placebo, carryover, equivalence tests; `drop=FALSE` guard (v2.3.3) |
 | `inter_fe_ub()` | `src/ife.cpp` | `fect_fe()` | `panel_factor()`, `fe_ub()`, `Y_demean()` | no | C++ IFE with unbalanced panels (EM algorithm) |
 | `inter_fe_mc()` | `src/mc.cpp` | `fect_mc()` | `panel_FE()`, `Y_demean()` | no | C++ matrix completion with nuclear norm |
-| `complex_fe_ub()` | `src/cfe.cpp` | `fect_cfe()` | `cfe_iter()`, `Y_demean()` | no | C++ complex FE estimation |
+| `complex_fe_ub()` | `src/cfe.cpp` | `fect_cfe()` | `cfe_iter()`, `Y_demean()` | no | C++ complex FE estimation; v2.4.2 optional `fit_init` parameter (NULL default = byte-identical to pre-v2.4.2) |
 | `panel_factor()` | `src/fe_sub.cpp` | `inter_fe_ub()`, others | SVD routines | no | Extract latent factors via SVD |
 | `panel_FE()` | `src/fe_sub.cpp` | `inter_fe_mc()`, others | soft-thresholding | no | Nuclear norm regularization / soft-thresholding |
 | `Y_demean()` | `src/fe_sub.cpp` | most C++ estimators | (arma operations) | no | Remove unit and/or time fixed effects |

DESCRIPTION

@@ -1,18 +1,17 @@
 Package: fect
 Type: Package
 Title: Fixed Effects Counterfactual Estimators
-Version: 2.4.1
-Date: 2026-04-29
+Version: 2.4.5
+Date: 2026-05-29
 Authors@R: 
-    c(person("Licheng", "Liu", , "lichengl@stanford.edu", role = c("aut")), 
+    c(person("Yiqing", "Xu", , "yiqingxu@stanford.edu", role = c("aut", "cre")),
+      person("Licheng", "Liu", , "lichengl@stanford.edu", role = c("aut")), 
       person("Ziyi", "Liu", , "zyliu2023@berkeley.edu", role = c("aut")),
-      person("Ye", "Wang", , "yezhehuzhi@gmail.com", role = c("aut")), 
-      person("Yiqing", "Xu", , "yiqingxu@stanford.edu", role = c("aut", "cre")),
+      person("Ye", "Wang", , "yezhehuzhi@gmail.com", role = c("aut")),       
       person("Tianzhu", "Qin", , "tianzhu@stanford.edu", role = c("aut")),
       person("Shiyun", "Hu", , "hushiyun@pku.edu.cn", role = c("aut")),
       person("Rivka", "Lipkovitz", , "rivkal@mit.edu", role = c("aut")))
-Author: Licheng Liu [aut], Ziyi Liu [aut], Ye Wang [aut], Yiqing Xu [aut, cre],
-    Tianzhu Qin [aut], Shiyun Hu [aut], Rivka Lipkovitz [aut]
+Author: Yiqing Xu [aut, cre], Licheng Liu [aut], Ziyi Liu [aut], Ye Wang [aut], Tianzhu Qin [aut], Shiyun Hu [aut], Rivka Lipkovitz [aut]
 Maintainer: Yiqing Xu <yiqingxu@stanford.edu>
 Description: Provides tools for estimating causal effects in panel data using counterfactual methods, as well as other modern DID estimators. It is designed for causal panel analysis with binary treatments under the parallel trends assumption. The package supports scenarios where treatments can switch on and off and allows for limited carryover effects. It includes several imputation estimators, such as Gsynth (Xu 2017), linear factor models, and the matrix completion method. Detailed methodology is described in Liu, Wang, and Xu (2024) <doi:10.48550/arXiv.2107.00856> and Chiu et al. (2025) <doi:10.48550/arXiv.2309.15983>. Optionally integrates with the "HonestDiDFEct" package for sensitivity analyses compatible with imputation estimators. "HonestDiDFEct" is not on CRAN but can be obtained from <https://github.com/lzy318/HonestDiDFEct>.
 URL: https://yiqingxu.org/packages/fect/, https://github.com/xuyiqing/fect
@@ -46,9 +45,10 @@ Suggests:
     panelView,
     testthat (>= 3.0.0),
     did,
-    DIDmultiplegtDYN,    
+    DIDmultiplegtDYN,
     ggrepel,
-    HonestDiDFEct
+    HonestDiDFEct,
+    withr
 Depends: R (>= 4.1.0)
 LinkingTo: Rcpp, RcppArmadillo
 RoxygenNote: 7.3.3