docs(v4): PR 4b of the 4.0 cut - rebuild tutorials 02/04/05/07/23 on the TWFE event-study mode (#867)

Author: igerber

changelog.d/20261023-v4-pr4b-tutorials-twfe-es.md

@@ -0,0 +1,29 @@
+### Documentation
+- **Tutorials 02/04/05/07/23 are rebuilt on `TwoWayFixedEffects().fit(..., event_study=True)`**
+  (`spec="within"`: unit + period fixed effects, unit-clustered SEs) ahead of the
+  MultiPeriodDiD removal (`M-010`, 4.0 cut PR 4c). Point estimates match the former
+  MultiPeriodDiD fits on these balanced panels; the SEs change because the within spec
+  clusters by unit. Tutorial 23's naive headline is the equal-weight average of the two
+  post-period coefficients computed in the cell (ATT unchanged at -4.29; the SE is now the
+  unit-clustered one), and tutorial 02's period-effects loop reads `to_dataframe()`.
+- The static `TwoWayFixedEffects().fit(..., time=...)` calls in tutorials 01/02/09 (the
+  `M-082` alias, misused there as the post dummy) are replaced: 01 passes `post="post"`;
+  02 and 09 report `TwoWayFixedEffects().decompose(...).twfe_estimate`, the unit + period
+  fixed-effects coefficient their Goodman-Bacon sections decompose (the old cells printed
+  `treated x period` artefacts: 0.6309 and 0.0001). Tutorial 02 cell 36's covariate draw is
+  seeded. Tutorials 02 and 09 are re-executed on the Rust backend, which produced the
+  committed outputs, so the seeded bootstrap cells are unchanged.
+- Tutorials 08 and 16 describe `StaggeredTripleDifference` as the removed class (4.0 cut,
+  PR 4a). `diff_diff/guides/llms-full.txt`'s event-study snippet calls `print(es.summary())`
+  (`EventStudyResults` has no `print_summary`).
+
+### Testing
+- `tests/test_t23_spillover_tva_drift.py`'s `naive_fit` fixture mirrors the notebook's TWFE
+  event-study headline cell, runs the fit with warnings escalated to errors, and pins the
+  post-period-average identity and the unit-clustered vcov-block SE alongside the existing
+  -4.29 endpoint.
+
+### Internal
+- `docs/v4-design.md` cut-sequence item 4b marked shipped; item 4c lists the two remaining
+  markdown-only MultiPeriodDiD mentions (06 cell 34, 16 cell 40). `docs/doc-deps.yaml` maps
+  the TWFE module to tutorials 01/02/04/05/07/09/23. No ledger row flips.

diff_diff/guides/llms-full.txt

@@ -146,7 +146,7 @@ results.print_summary()
 # auto-cluster applies with the static carve-outs)
 es = twfe.fit(data, outcome='y', treatment='treated', unit='unit_id',
               event_study=True, time='period', post_periods=[4, 5, 6, 7])
-es.print_summary()
+print(es.summary())
 ```
 
 **Note:** TWFE can be biased with staggered treatment timing and heterogeneous effects. Consider CallawaySantAnna, SunAbraham, or ImputationDiD for staggered designs.

docs/doc-deps.yaml

@@ -186,6 +186,27 @@ sources:
         note: "TWFE block documents the full fit signature incl. event-study params; the Conley section's panel examples use post=/event_study (rows M-010/M-082)"
       - path: docs/migration-4.0.md
         type: user_guide
+      - path: docs/tutorials/01_basic_did.ipynb
+        type: tutorial
+        note: "Static TWFE example (post= dummy, row M-082)"
+      - path: docs/tutorials/02_staggered_did.ipynb
+        type: tutorial
+        note: "decompose() TWFE estimate + the TWFE event-study comparison section (event_study=True, spec within)"
+      - path: docs/tutorials/04_parallel_trends.ipynb
+        type: tutorial
+        note: "Event-study parallel-trends check uses event_study=True"
+      - path: docs/tutorials/05_honest_did.ipynb
+        type: tutorial
+        note: "HonestDiD inputs come from TWFE event_study=True"
+      - path: docs/tutorials/07_pretrends_power.ipynb
+        type: tutorial
+        note: "PreTrendsPower inputs come from TWFE event_study=True"
+      - path: docs/tutorials/09_real_world_examples.ipynb
+        type: tutorial
+        note: "decompose() TWFE estimate in the castle-doctrine section"
+      - path: docs/tutorials/23_spillover_tva.ipynb
+        type: tutorial
+        note: "Naive headline = TWFE event_study=True post-period average"
 
   # ── CallawaySantAnna (staggered group) ─────��────────────────────────
 

docs/tutorials/01_basic_did.ipynb

@@ -306,7 +306,7 @@
    "source": [
     "## Two-Way Fixed Effects (TWFE)\n",
     "\n",
-    "For panel data, the `TwoWayFixedEffects` estimator automatically includes both unit and time fixed effects using within-transformation."
+    "For panel data, the `TwoWayFixedEffects` estimator absorbs unit fixed effects and the pre/post period effect by within-transformation: the regressor is `treated x post`, and `post=` names the 0/1 post-treatment column. For period-by-period effects with calendar-time fixed effects use the event-study mode, `fit(..., unit=, time=, event_study=True, post_periods=)` (tutorials 04 and 05)."
    ]
   },
   {
@@ -321,7 +321,7 @@
     "    panel_df,\n",
     "    outcome=\"outcome\",\n",
     "    treatment=\"treated\",\n",
-    "    time=\"period\",  # Use actual time periods\n",
+    "    post=\"post\",  # 0/1 post-treatment indicator\n",
     "    unit=\"unit\"\n",
     ")\n",
     "\n",

docs/tutorials/02_staggered_did.ipynb

@@ -25,7 +25,7 @@
    "source": [
     "import numpy as np\n",
     "import pandas as pd\n",
-    "from diff_diff import DifferenceInDifferences, MultiPeriodDiD\n",
+    "from diff_diff import DifferenceInDifferences, TwoWayFixedEffects\n",
     "from diff_diff.utils import (\n",
     "    check_parallel_trends,\n",
     "    check_parallel_trends_robust,\n",
@@ -543,7 +543,7 @@
    "source": [
     "## Event Study as a Parallel Trends Check\n",
     "\n",
-    "An **event study** shows period-by-period effects. Pre-treatment coefficients should be close to zero if parallel trends holds."
+    "An **event study** shows period-by-period effects. Pre-treatment coefficients should be close to zero if parallel trends holds. We fit a TWFE event study (`TwoWayFixedEffects().fit(..., event_study=True)`): unit and period fixed effects, one coefficient per period relative to the reference period 3."
    ]
   },
   {
@@ -552,13 +552,15 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Event study\n",
-    "mp_did = MultiPeriodDiD()\n",
-    "event_results = mp_did.fit(\n",
+    "# Event study: unit + period fixed effects, one coefficient per period\n",
+    "twfe = TwoWayFixedEffects()\n",
+    "event_results = twfe.fit(\n",
     "    df_parallel,\n",
     "    outcome='outcome',\n",
     "    treatment='treated',\n",
+    "    unit='unit',\n",
     "    time='period',\n",
+    "    event_study=True,\n",
     "    post_periods=[4, 5, 6, 7],\n",
     "    reference_period=3  # Use period 3 as reference\n",
     ")\n",

docs/tutorials/04_parallel_trends.ipynb

@@ -32,7 +32,7 @@
    "source": [
     "import numpy as np\n",
     "import pandas as pd\n",
-    "from diff_diff import MultiPeriodDiD\n",
+    "from diff_diff import TwoWayFixedEffects\n",
     "from diff_diff.honest_did import (\n",
     "    HonestDiD,\n",
     "    compute_honest_did,\n",
@@ -148,7 +148,7 @@
    "source": [
     "## Fit Standard Event Study\n",
     "\n",
-    "First, let's estimate a standard event study using `MultiPeriodDiD`."
+    "First, let's estimate a standard event study with `TwoWayFixedEffects().fit(..., event_study=True)` (unit and period fixed effects; the last pre-period, 4, is the reference)."
    ]
   },
   {
@@ -158,13 +158,15 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "# Fit event study\n",
-    "mp_did = MultiPeriodDiD()\n",
-    "event_results = mp_did.fit(\n",
+    "# Fit event study (unit + period fixed effects)\n",
+    "twfe = TwoWayFixedEffects()\n",
+    "event_results = twfe.fit(\n",
     "    df,\n",
     "    outcome='outcome',\n",
     "    treatment='treated',\n",
+    "    unit='unit',\n",
     "    time='period',\n",
+    "    event_study=True,\n",
     "    post_periods=[5, 6, 7, 8, 9]\n",
     ")\n",
     "\n",

docs/tutorials/05_honest_did.ipynb

@@ -33,7 +33,7 @@
     "import pandas as pd\n",
     "\n",
     "from diff_diff import (\n",
-    "    MultiPeriodDiD,\n",
+    "    TwoWayFixedEffects,\n",
     "    PreTrendsPower,\n",
     "    compute_pretrends_power,\n",
     "    compute_mdv,\n",
@@ -131,7 +131,26 @@
    "id": "cell-6",
    "metadata": {},
    "outputs": [],
-   "source": "# Fit event study with ALL periods (pre and post) relative to reference period\n# For pre-trends power analysis, we need coefficients for pre-periods too\nmp_did = MultiPeriodDiD()\n\n# Use period 4 as the reference period (last pre-period, excluded from estimation)\n# Specify post_periods as the actual post-treatment periods; MultiPeriodDiD\n# automatically estimates pre-period coefficients for the event study.\nevent_results = mp_did.fit(\n    df,\n    outcome='outcome',\n    treatment='treated',\n    time='period',\n    post_periods=[5, 6, 7, 8, 9]\n)\n\nprint(event_results.summary())"
+   "source": [
+    "# Fit event study with ALL periods (pre and post) relative to reference period\n",
+    "# For pre-trends power analysis, we need coefficients for pre-periods too\n",
+    "twfe = TwoWayFixedEffects()\n",
+    "\n",
+    "# Use period 4 as the reference period (last pre-period, excluded from estimation)\n",
+    "# Specify post_periods as the actual post-treatment periods; the event-study mode\n",
+    "# estimates one coefficient per other period, pre-periods included.\n",
+    "event_results = twfe.fit(\n",
+    "    df,\n",
+    "    outcome='outcome',\n",
+    "    treatment='treated',\n",
+    "    unit='unit',\n",
+    "    time='period',\n",
+    "    event_study=True,\n",
+    "    post_periods=[5, 6, 7, 8, 9]\n",
+    ")\n",
+    "\n",
+    "print(event_results.summary())"
+   ]
   },
   {
    "cell_type": "code",
@@ -520,7 +539,46 @@
    "id": "cell-30",
    "metadata": {},
    "outputs": [],
-   "source": "# Typical workflow for pre-trends power analysis\n\n# Step 1: Estimate event study with proper pre/post period classification\nmp_did = MultiPeriodDiD()\n\n# Specify actual post-treatment periods; pre-period coefficients are\n# estimated automatically by MultiPeriodDiD for the event study\npre_treatment_periods = [0, 1, 2, 3]  # Define which are pre-treatment\n\nresults = mp_did.fit(\n    df, \n    outcome='outcome',\n    treatment='treated', \n    time='period',\n    post_periods=[5, 6, 7, 8, 9]\n)\n\n# Step 2: Assess power of the pre-trends test  \nprint(\"Step 2: Pre-Trends Power Analysis\")\npt = PreTrendsPower(alpha=0.05, power=0.80, violation_type='linear')\npt_results = pt.fit(results, pre_periods=pre_treatment_periods)\nprint(f\"MDV (80% power): {pt_results.mdv:.3f}\")\nprint(\"\")\n\n# Step 3: Interpret\nprint(\"Step 3: Interpretation\")\nprint(f\"Your pre-trends test could only detect violations >= {pt_results.mdv:.3f}\")\nprint(f\"Violations smaller than this would likely go undetected.\")\nprint(\"\")\n\n# Step 4: Connect to Honest DiD for robust inference\nprint(\"Step 4: Robust Inference with Honest DiD\")\nhonest = HonestDiD(method='smoothness', M=pt_results.mdv)\nhonest_results = honest.fit(results)\nprint(f\"Robust 95% CI (M=MDV): [{honest_results.ci_lb:.3f}, {honest_results.ci_ub:.3f}]\")\nprint(f\"Conclusion: {'Effect is robust' if honest_results.is_significant else 'Effect may not be robust'}\")"
+   "source": [
+    "# Typical workflow for pre-trends power analysis\n",
+    "\n",
+    "# Step 1: Estimate event study with proper pre/post period classification\n",
+    "twfe = TwoWayFixedEffects()\n",
+    "\n",
+    "# Specify actual post-treatment periods; the event-study mode estimates\n",
+    "# the pre-period coefficients automatically\n",
+    "pre_treatment_periods = [0, 1, 2, 3]  # Define which are pre-treatment\n",
+    "\n",
+    "results = twfe.fit(\n",
+    "    df, \n",
+    "    outcome='outcome',\n",
+    "    treatment='treated', \n",
+    "    unit='unit',\n",
+    "    time='period',\n",
+    "    event_study=True,\n",
+    "    post_periods=[5, 6, 7, 8, 9]\n",
+    ")\n",
+    "\n",
+    "# Step 2: Assess power of the pre-trends test  \n",
+    "print(\"Step 2: Pre-Trends Power Analysis\")\n",
+    "pt = PreTrendsPower(alpha=0.05, power=0.80, violation_type='linear')\n",
+    "pt_results = pt.fit(results, pre_periods=pre_treatment_periods)\n",
+    "print(f\"MDV (80% power): {pt_results.mdv:.3f}\")\n",
+    "print(\"\")\n",
+    "\n",
+    "# Step 3: Interpret\n",
+    "print(\"Step 3: Interpretation\")\n",
+    "print(f\"Your pre-trends test could only detect violations >= {pt_results.mdv:.3f}\")\n",
+    "print(f\"Violations smaller than this would likely go undetected.\")\n",
+    "print(\"\")\n",
+    "\n",
+    "# Step 4: Connect to Honest DiD for robust inference\n",
+    "print(\"Step 4: Robust Inference with Honest DiD\")\n",
+    "honest = HonestDiD(method='smoothness', M=pt_results.mdv)\n",
+    "honest_results = honest.fit(results)\n",
+    "print(f\"Robust 95% CI (M=MDV): [{honest_results.ci_lb:.3f}, {honest_results.ci_ub:.3f}]\")\n",
+    "print(f\"Conclusion: {'Effect is robust' if honest_results.is_significant else 'Effect may not be robust'}\")"
+   ]
   },
   {
    "cell_type": "markdown",

docs/tutorials/07_pretrends_power.ipynb

@@ -673,8 +673,8 @@
     "The staggered parameters are keyword-only, and mixing the two designs' parameters\n",
     "raises rather than guessing which one you meant.\n",
     "\n",
-    "> This replaces `StaggeredTripleDifference`, which is deprecated in 3.9 and\n",
-    "> removed in 4.0. It runs the same engine, so the numbers are unchanged; on the\n",
+    "> This replaces the former `StaggeredTripleDifference` class (deprecated in 3.9,\n",
+    "> removed at the 4.0 cut). It runs the same engine, so the numbers are unchanged; on the\n",
     "> merged surface the third dimension is named `partition=` (not `eligibility=`)\n",
     "> and `control_group` takes `\"not_yet_treated\"` / `\"never_treated\"`."
    ]