Fix 3 bugs found in code review: stacked_did, drdid, jackknife_se

- dgp_did: reserve ~20% units as never-treated for stacked_did compatibility
- drdid: accept `seed` as alias for `random_state` parameter
- jackknife_se: fix dependent variable lookup using `dependent_var` key

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: brycewang-stanford

src/statspai/did/wooldridge_did.py

@@ -363,6 +363,7 @@ def drdid(
     alpha: float = 0.05,
     n_boot: int = 500,
     random_state: Optional[int] = None,
+    seed: Optional[int] = None,
 ) -> CausalResult:
     """
     Doubly Robust Difference-in-Differences (Sant'Anna & Zhao 2020).
@@ -416,7 +417,7 @@ def drdid(
     True
     """
     df = data.copy()
-    rng = np.random.default_rng(random_state)
+    rng = np.random.default_rng(random_state if random_state is not None else seed)
 
     # ── Validate 2×2 design ─────────────────────────────────────────
     g_vals = sorted(df[group].dropna().unique())

src/statspai/inference/jackknife.py

@@ -492,7 +492,9 @@ def _parse_formula(
 
     # Fallback: use result parameter names
     param_names = list(result.params.index)
-    y_var = result.model_info.get('depvar', result.data_info.get('y_var', 'y'))
+    y_var = result.data_info.get('dependent_var',
+                result.model_info.get('depvar',
+                    result.data_info.get('y_var', 'y')))
     x_vars = [p for p in param_names
               if p.lower() not in ('intercept', '_const', 'const', '(intercept)')]
     return y_var, x_vars

src/statspai/utils/dgp.py

@@ -71,13 +71,14 @@ def dgp_did(
     first_treat = np.full(n_units, np.inf)
 
     if staggered:
-        group = rng.integers(0, n_groups, size=n_units)
-        # Spread adoption across the last 60 % of periods
+        # Group 0 = never-treated (~20% of units); groups 1..n_groups = treated
+        group = rng.integers(0, n_groups + 1, size=n_units)
         treat_times = np.linspace(
             int(n_periods * 0.4), n_periods - 1, n_groups, dtype=int
         )
         for g in range(n_groups):
-            first_treat[group == g] = treat_times[g]
+            first_treat[group == g + 1] = treat_times[g]
+        # group 0 stays never-treated (first_treat = inf)
     else:
         # Classic 2x2: half treated at mid-point
         treated_mask = rng.choice(n_units, size=n_units // 2, replace=False)