formative-ds 2.2.1__tar.gz → 2.2.2__tar.gz

{formative_ds-2.2.1 → formative_ds-2.2.2}/CLAUDE.md

@@ -71,6 +71,7 @@ Do not change `_RCC_SEED` or `_PLACEBO_SEED` to 42 — the generated noise colli
 **Package layout:**
 - `formative/causal/dag.py` — `DAG`, `_Node`
 - `formative/causal/_exceptions.py` — `IdentificationError`, `GraphError`
+- `formative/causal/_assumptions.py` — `Assumption` (estimators must not import from `refutations/` at module level — keeping this one-directional is what lets refutation modules import estimator internals without import cycles)
 - `formative/causal/estimators/_base.py` — `_BaseResult`, `_StatsmodelsResult` (shared result plumbing: effect stats via `_param`, `decide()`, assumptions footer)
 - `formative/causal/estimators/_cate.py` — `GroupEffect`, `_fit_cate`, `_CATEResultMixin`, modifier validation (shared by OLS/RCT and the CATE refutations)
 - `formative/causal/estimators/ols.py` — `OLSObservational`, `OLSResult`, `OLSCATEResult`
@@ -80,12 +81,12 @@ Do not change `_RCC_SEED` or `_PLACEBO_SEED` to 42 — the generated noise colli
 - `formative/causal/estimators/rct.py` — `RCT`, `RCTResult`, `RCTCATEResult`
 - `formative/causal/estimators/did.py` — `DiD`, `DiDResult`
 - `formative/causal/estimators/rdd.py` — `RDD`, `RDDResult`
-- `formative/causal/refutations/_check.py` — `Assumption`, `RefutationCheck`, `RefutationReport`
+- `formative/causal/refutations/_check.py` — `RefutationCheck`, `RefutationReport`, shared verdict helpers (`_shift_check`, `_placebo_check`)
 - `formative/causal/refutations/ols.py` — `OLSRefutationReport`, `_check_random_common_cause`
 - `formative/causal/refutations/iv.py` — `IVRefutationReport`, `_check_first_stage_f`, `_check_random_common_cause`
 - `formative/causal/refutations/matching.py` — `MatchingRefutationReport`, `_check_placebo_treatment`, `_check_random_common_cause`
 - `formative/causal/refutations/rct.py` — `RCTRefutationReport`, `_check_random_common_cause`
-- `formative/causal/refutations/policy.py` — `PolicyRefutationReport`, `_check_placebo_modifiers`, `_check_random_modifier` (import `_fit_policy` lazily — a module-level import is circular via `refutations/__init__`)
+- `formative/causal/refutations/policy.py` — `PolicyRefutationReport`, `_check_placebo_modifiers`, `_check_random_modifier`
 - `formative/causal/refutations/cate.py` — `_check_placebo_modifier`, `_check_random_modifier`, `_check_random_common_cause` (used by `OLSCATEResult`/`RCTCATEResult`; reports reuse the OLS/RCT report classes)
 - `formative/causal/refutations/did.py` — `DiDRefutationReport`, `_check_placebo_group`, `_check_random_common_cause`
 - `formative/causal/refutations/rdd.py` — `RDDRefutationReport`, `_check_placebo_cutoff`, `_check_random_common_cause`

{formative_ds-2.2.1 → formative_ds-2.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: formative-ds
-Version: 2.2.1
+Version: 2.2.2
 Summary: A Python library for quantitative reasoning.
 Project-URL: Homepage, https://getformative.dev
 Project-URL: Documentation, https://docs.getformative.dev

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/__init__.py

@@ -1,3 +1,4 @@
+from ._assumptions import Assumption
 from .dag import DAG
 from .decision import DecisionReport
 from .estimators._cate import GroupEffect
@@ -19,7 +20,6 @@ from .refutations import (
     RefutationCheck,
     RefutationReport,
 )
-from .refutations._check import Assumption
 
 __all__ = [
     "DAG",

formative_ds-2.2.2/formative/causal/_assumptions.py

@@ -0,0 +1,25 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True)
+class Assumption:
+    """
+    A single modelling assumption required for causal identification.
+
+    Every estimator exposes its assumptions via ``result.assumptions``,
+    a list of ``Assumption`` objects. Each assumption has a human-readable
+    name and a ``testable`` flag indicating whether it can be empirically
+    checked in the data or must be justified on substantive grounds.
+    """
+
+    name: str
+    """Human-readable description of the assumption."""
+
+    testable: bool
+    """``True`` if the assumption can be empirically checked; ``False`` if it rests on domain knowledge."""
+
+    def fmt_tag(self) -> str:
+        """Return a fixed-width bracketed testability label for use in summary output."""
+        return "[  testable  ]" if self.testable else "[ untestable ]"

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/_base.py

@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-from ..refutations._check import Assumption
+from .._assumptions import Assumption
 
 
 class _BaseResult:

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/_cate.py

@@ -16,8 +16,8 @@ import numpy as np
 import pandas as pd
 import statsmodels.formula.api as smf
 
+from .._assumptions import Assumption
 from ..dag import DAG
-from ..refutations._check import Assumption
 
 CATE_ASSUMPTIONS: list[Assumption] = [
     Assumption("Correct specification of effect modification (linear interaction form)", testable=False),

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/did.py

@@ -3,8 +3,8 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
+from .._assumptions import Assumption
 from ..dag import DAG
-from ..refutations._check import Assumption
 from ._base import _StatsmodelsResult
 
 DID_ASSUMPTIONS: list[Assumption] = [

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/iv.py

@@ -5,8 +5,8 @@ import statsmodels.api as sm
 import statsmodels.formula.api as smf
 from statsmodels.sandbox.regression.gmm import IV2SLS as _IV2SLS
 
+from .._assumptions import Assumption
 from ..dag import DAG
-from ..refutations._check import Assumption
 from ._base import _StatsmodelsResult
 
 IV_ASSUMPTIONS: list[Assumption] = [

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/matching.py

@@ -2,11 +2,12 @@ from __future__ import annotations
 
 import numpy as np
 import pandas as pd
+import statsmodels.api as sm
 import statsmodels.formula.api as smf
 
+from .._assumptions import Assumption
 from .._exceptions import IdentificationError
 from ..dag import DAG
-from ..refutations._check import Assumption
 from ._base import _BaseResult
 
 MATCHING_ASSUMPTIONS: list[Assumption] = [
@@ -23,19 +24,19 @@ _BOOTSTRAP_SEED = 42
 # ── Private helpers (also imported by formative/refutations/matching.py) ──────
 
 
+def _ps_model(data: pd.DataFrame, treatment: str, adjustment_set: set[str]):
+    """Unfitted logit model of treatment on the adjustment set (intercept-only when empty)."""
+    rhs = " + ".join(sorted(adjustment_set)) if adjustment_set else "1"
+    return smf.logit(f"{treatment} ~ {rhs}", data=data)
+
+
 def _propensity_scores(
     data: pd.DataFrame,
     treatment: str,
     adjustment_set: set[str],
 ) -> np.ndarray:
     """Fit logistic regression and return propensity scores as a 1-D array."""
-    if adjustment_set:
-        rhs = " + ".join(sorted(adjustment_set))
-        ps = smf.logit(f"{treatment} ~ {rhs}", data=data).fit(disp=0).predict()
-    else:
-        # Intercept-only: all units get the same PS (treatment base rate).
-        ps = smf.logit(f"{treatment} ~ 1", data=data).fit(disp=0).predict()
-    return np.asarray(ps)
+    return np.asarray(_ps_model(data, treatment, adjustment_set).fit(disp=0).predict())
 
 
 def _att_from_ps(
@@ -54,11 +55,19 @@ def _att_from_ps(
         raise ValueError("Sample must contain both treated and control units.")
 
     ps_treated = ps[treated_idx]
-    ps_control = ps[control_idx]
 
-    # For each treated unit find the nearest control unit by PS distance.
-    dists = np.abs(ps_treated[:, None] - ps_control[None, :])
-    matched = control_idx[np.argmin(dists, axis=1)]
+    # For each treated unit find the nearest control unit by PS distance,
+    # by binary search over the sorted control scores rather than a full
+    # treated × control distance matrix.
+    order = np.argsort(ps[control_idx], kind="stable")
+    sorted_controls = control_idx[order]
+    sorted_ps = ps[sorted_controls]
+
+    pos = np.searchsorted(sorted_ps, ps_treated)
+    left = np.clip(pos - 1, 0, len(sorted_ps) - 1)
+    right = np.clip(pos, 0, len(sorted_ps) - 1)
+    take_right = np.abs(sorted_ps[right] - ps_treated) < np.abs(sorted_ps[left] - ps_treated)
+    matched = sorted_controls[np.where(take_right, right, left)]
 
     return float(np.mean(outcome[treated_idx] - outcome[matched]))
 
@@ -340,22 +349,21 @@ class PropensityScoreMatching:
         Y_arr = data[Y].values.astype(float)
 
         # Point estimate ATT.
-        ps = _propensity_scores(data, T, adjustment_set)
+        model = _ps_model(data, T, adjustment_set)
+        ps = np.asarray(model.fit(disp=0).predict())
         att = _att_from_ps(T_arr, Y_arr, ps)
 
         # Bootstrap SE and CI.
-        # Use a minimal DataFrame (treatment + confounders only) to avoid
-        # copying unneeded columns on every iteration.
+        # Resample rows of the design matrix built for the point estimate —
+        # re-parsing the formula per replicate dominates the runtime otherwise.
+        X = model.exog
         rng = np.random.default_rng(_BOOTSTRAP_SEED)
         n = len(data)
-        ps_cols = sorted({T} | adjustment_set)
-        ps_data = data[ps_cols]
         boot = []
         for _ in range(_BOOTSTRAP_N):
             idx = rng.integers(0, n, size=n)
-            bd = ps_data.iloc[idx].reset_index(drop=True)
             try:
-                bps = _propensity_scores(bd, T, adjustment_set)
+                bps = np.asarray(sm.Logit(T_arr[idx], X[idx]).fit(disp=0).predict())
                 boot.append(_att_from_ps(T_arr[idx], Y_arr[idx], bps))
             except Exception:
                 # Degenerate sample or logit failure — skip this replicate.

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/ols.py

@@ -3,9 +3,9 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
+from .._assumptions import Assumption
 from .._exceptions import IdentificationError
 from ..dag import DAG
-from ..refutations._check import Assumption
 from ._base import _StatsmodelsResult
 from ._cate import (
     CATE_ASSUMPTIONS,

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/policy.py

@@ -22,7 +22,7 @@ from dataclasses import dataclass
 import numpy as np
 import pandas as pd
 
-from ..refutations._check import Assumption
+from .._assumptions import Assumption
 from ._cate import _validate_modifier_dag
 
 _FOLD_SEED = 13579

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/rct.py

@@ -3,8 +3,8 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
+from .._assumptions import Assumption
 from ..dag import DAG
-from ..refutations._check import Assumption
 from ._base import _StatsmodelsResult
 from ._cate import (
     CATE_ASSUMPTIONS,

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/estimators/rdd.py

@@ -3,8 +3,8 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
+from .._assumptions import Assumption
 from ..dag import DAG
-from ..refutations._check import Assumption
 from ._base import _StatsmodelsResult
 
 RDD_ASSUMPTIONS: list[Assumption] = [
@@ -253,9 +253,9 @@ class RDD:
         # Apply bandwidth filter
         if self._bandwidth is not None:
             mask = (data[self._running_var] - self._cutoff).abs() <= self._bandwidth
-            filtered = data.loc[mask].copy()
+            filtered = data.loc[mask]
         else:
-            filtered = data.copy()
+            filtered = data
 
         # Unadjusted effect: naive mean difference above vs below cutoff
         above = filtered.loc[filtered[self._running_var] >= self._cutoff, self._outcome].mean()

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/_check.py

@@ -1,7 +1,5 @@
 from __future__ import annotations
 
-from dataclasses import dataclass
-
 import numpy as np
 import pandas as pd
 
@@ -17,26 +15,32 @@ def _add_random_column(data: pd.DataFrame, seed: int = _RCC_SEED) -> tuple[pd.Da
     return data.assign(**{col: rng.normal(size=len(data))}), col
 
 
-@dataclass(frozen=True)
-class Assumption:
-    """
-    A single modelling assumption required for causal identification.
-
-    Every estimator exposes its assumptions via ``result.assumptions``,
-    a list of ``Assumption`` objects. Each assumption has a human-readable
-    name and a ``testable`` flag indicating whether it can be empirically
-    checked in the data or must be justified on substantive grounds.
-    """
-
-    name: str
-    """Human-readable description of the assumption."""
-
-    testable: bool
-    """``True`` if the assumption can be empirically checked; ``False`` if it rests on domain knowledge."""
-
-    def fmt_tag(self) -> str:
-        """Return a fixed-width bracketed testability label for use in summary output."""
-        return "[  testable  ]" if self.testable else "[ untestable ]"
+def _shift_check(new_effect: float, original_effect: float, original_se: float, fail_suffix: str) -> RefutationCheck:
+    """Random-common-cause verdict: pass iff the refitted estimate moved by at most one standard error."""
+    shift = abs(new_effect - original_effect)
+    passed = shift <= original_se
+    if passed:
+        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
+    else:
+        detail = f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  {fail_suffix}"
+    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+
+
+def _placebo_check(
+    name: str,
+    label: str,
+    placebo_effect: float,
+    original_se: float,
+    pass_suffix: str,
+    fail_suffix: str,
+) -> RefutationCheck:
+    """Placebo verdict: pass iff the placebo estimate is within one standard error of zero."""
+    passed = abs(placebo_effect) <= original_se
+    if passed:
+        detail = f"{label} = {placebo_effect:.4f}  (≤ 1 SE = {original_se:.4f})  {pass_suffix}"
+    else:
+        detail = f"{label} = {placebo_effect:.4f}  (> 1 SE = {original_se:.4f})  {fail_suffix}"
+    return RefutationCheck(name=name, passed=passed, detail=detail)
 
 
 class RefutationCheck:

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/cate.py

@@ -13,7 +13,7 @@ import numpy as np
 import pandas as pd
 
 from ..estimators._cate import _fit_cate
-from ._check import RefutationCheck, _add_random_column
+from ._check import RefutationCheck, _add_random_column, _shift_check
 
 _PLACEBO_MODIFIER_SEED = 77777
 _RANDOM_MODIFIER_SEED = 31313
@@ -42,17 +42,9 @@ def _check_random_common_cause(
 
     fit = _fit_cate(augmented, treatment, outcome, modifier, adjustment_set | {col})
 
-    shift = abs(fit.effect - original_effect)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random common cause destabilised the estimate."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    return _shift_check(
+        fit.effect, original_effect, original_se, "Adding a random common cause destabilised the estimate."
+    )
 
 
 def _check_placebo_modifier(

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/did.py

@@ -4,7 +4,7 @@ import numpy as np
 import pandas as pd
 import statsmodels.formula.api as smf
 
-from ._check import RefutationCheck, RefutationReport, _add_random_column
+from ._check import RefutationCheck, RefutationReport, _add_random_column, _placebo_check, _shift_check
 
 _PLACEBO_SEED = 99999
 _PLACEBO_TIME_SEED = 22222
@@ -25,15 +25,14 @@ def _placebo_permutation(
     rng = np.random.default_rng(seed)
     augmented = data.assign(**{permute_var: rng.permutation(data[permute_var].values)})
     placebo_effect = float(smf.ols(f"{outcome} ~ {group} * {time}", data=augmented).fit().params[f"{group}:{time}"])
-    passed = abs(placebo_effect) <= original_se
-    if passed:
-        detail = (
-            f"placebo estimate = {placebo_effect:.4f}  (≤ 1 SE = {original_se:.4f})  "
-            f"Permuting {axis_label} labels yields near-zero effect, as expected."
-        )
-    else:
-        detail = f"placebo estimate = {placebo_effect:.4f}  (> 1 SE = {original_se:.4f})  {fail_suffix}"
-    return RefutationCheck(name=check_name, passed=passed, detail=detail)
+    return _placebo_check(
+        name=check_name,
+        label="placebo estimate",
+        placebo_effect=placebo_effect,
+        original_se=original_se,
+        pass_suffix=f"Permuting {axis_label} labels yields near-zero effect, as expected.",
+        fail_suffix=fail_suffix,
+    )
 
 
 def _check_placebo_group(
@@ -120,20 +119,11 @@ def _check_random_common_cause(
     augmented, col = _add_random_column(data)
 
     result = smf.ols(f"{outcome} ~ {group} * {time} + {col}", data=augmented).fit()
-    interaction = f"{group}:{time}"
-    new_effect = float(result.params[interaction])
-
-    shift = abs(new_effect - original_effect)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random common cause destabilised the DiD estimate."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    new_effect = float(result.params[f"{group}:{time}"])
+
+    return _shift_check(
+        new_effect, original_effect, original_se, "Adding a random common cause destabilised the DiD estimate."
+    )
 
 
 class DiDRefutationReport(RefutationReport):

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/iv.py

@@ -5,7 +5,7 @@ import statsmodels.api as sm
 import statsmodels.formula.api as smf
 from statsmodels.sandbox.regression.gmm import IV2SLS as _IV2SLS
 
-from ._check import RefutationCheck, RefutationReport, _add_random_column
+from ._check import RefutationCheck, RefutationReport, _add_random_column, _shift_check
 
 _FIRST_STAGE_F_THRESHOLD = 10.0
 
@@ -68,17 +68,9 @@ def _check_random_common_cause(
 
     new_effect = float(_IV2SLS(endog=augmented[outcome], exog=X, instrument=Z_mat).fit().params[treatment])
 
-    shift = abs(new_effect - original_effect)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random common cause destabilised the estimate."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    return _shift_check(
+        new_effect, original_effect, original_se, "Adding a random common cause destabilised the estimate."
+    )
 
 
 class IVRefutationReport(RefutationReport):

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/matching.py

@@ -4,7 +4,7 @@ import numpy as np
 import pandas as pd
 
 from ..estimators.matching import _att_from_ps, _propensity_scores
-from ._check import RefutationCheck, RefutationReport, _add_random_column
+from ._check import RefutationCheck, RefutationReport, _add_random_column, _placebo_check, _shift_check
 
 _PLACEBO_SEED = 99999
 
@@ -42,19 +42,17 @@ def _check_placebo_treatment(
             detail="Matching failed on permuted treatment — check data quality.",
         )
 
-    passed = abs(placebo_att) <= original_se
-    if passed:
-        detail = (
-            f"placebo ATT = {placebo_att:.4f}  (≤ 1 SE = {original_se:.4f})  "
-            f"Permuting treatment labels yields near-zero effect, as expected."
-        )
-    else:
-        detail = (
-            f"placebo ATT = {placebo_att:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"A randomly permuted treatment produced a large effect — the original "
-            f"result may be driven by residual confounding or overfitting."
-        )
-    return RefutationCheck(name="Placebo treatment", passed=passed, detail=detail)
+    return _placebo_check(
+        name="Placebo treatment",
+        label="placebo ATT",
+        placebo_effect=placebo_att,
+        original_se=original_se,
+        pass_suffix="Permuting treatment labels yields near-zero effect, as expected.",
+        fail_suffix=(
+            "A randomly permuted treatment produced a large effect — the original "
+            "result may be driven by residual confounding or overfitting."
+        ),
+    )
 
 
 def _check_random_common_cause(
@@ -91,17 +89,9 @@ def _check_random_common_cause(
             detail="Matching failed after adding random covariate — check data quality.",
         )
 
-    shift = abs(new_att - original_att)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random common cause destabilised the ATT estimate."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    return _shift_check(
+        new_att, original_att, original_se, "Adding a random common cause destabilised the ATT estimate."
+    )
 
 
 class MatchingRefutationReport(RefutationReport):

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/ols.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
-from ._check import RefutationCheck, RefutationReport, _add_random_column
+from ._check import RefutationCheck, RefutationReport, _add_random_column, _shift_check
 
 
 def _check_random_common_cause(
@@ -27,17 +27,9 @@ def _check_random_common_cause(
     rhs = " + ".join([treatment] + controls)
     new_effect = float(smf.ols(f"{outcome} ~ {rhs}", data=augmented).fit().params[treatment])
 
-    shift = abs(new_effect - original_effect)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random common cause destabilised the estimate."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    return _shift_check(
+        new_effect, original_effect, original_se, "Adding a random common cause destabilised the estimate."
+    )
 
 
 class OLSRefutationReport(RefutationReport):

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/policy.py

@@ -13,6 +13,7 @@ from __future__ import annotations
 import numpy as np
 import pandas as pd
 
+from ..estimators.policy import _fit_policy
 from ._check import RefutationCheck, RefutationReport
 
 _PLACEBO_POLICY_SEED = 88888
@@ -44,8 +45,6 @@ def _check_placebo_modifiers(
     (z ≤ 2). A significantly positive placebo value means the learner is
     manufacturing value out of noise.
     """
-    from ..estimators.policy import _fit_policy
-
     rng = np.random.default_rng(_PLACEBO_POLICY_SEED)
     augmented = data.assign(**{m: rng.permutation(data[m].to_numpy()) for m in modifiers})
 
@@ -87,8 +86,6 @@ def _check_random_modifier(
     Noise carries no targeting information, so the honest value should not
     improve by more than one standard error of the original estimate.
     """
-    from ..estimators.policy import _fit_policy
-
     rng = np.random.default_rng(_RANDOM_POLICY_MODIFIER_SEED)
     col = "_rpol"
     while col in data.columns:

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/rct.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
-from ._check import RefutationCheck, RefutationReport, _add_random_column
+from ._check import RefutationCheck, RefutationReport, _add_random_column, _shift_check
 
 
 def _check_random_common_cause(
@@ -24,18 +24,12 @@ def _check_random_common_cause(
 
     new_effect = float(smf.ols(f"{outcome} ~ {treatment} + {col}", data=augmented).fit().params[treatment])
 
-    shift = abs(new_effect - original_effect)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (≤ 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random covariate destabilised the estimate — "
-            f"verify that treatment was truly randomised."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    return _shift_check(
+        new_effect,
+        original_effect,
+        original_se,
+        "Adding a random covariate destabilised the estimate — verify that treatment was truly randomised.",
+    )
 
 
 class RCTRefutationReport(RefutationReport):

{formative_ds-2.2.1 → formative_ds-2.2.2}/formative/causal/refutations/rdd.py

@@ -3,7 +3,7 @@ from __future__ import annotations
 import pandas as pd
 import statsmodels.formula.api as smf
 
-from ._check import RefutationCheck, RefutationReport, _add_random_column
+from ._check import RefutationCheck, RefutationReport, _add_random_column, _shift_check
 
 _PLACEBO_SEED = 99999
 
@@ -86,17 +86,9 @@ def _check_random_common_cause(
     ).fit()
     new_effect = float(result.params[treatment])
 
-    shift = abs(new_effect - original_effect)
-    passed = shift <= original_se
-
-    if passed:
-        detail = f"estimate shifted by {shift:.4f}  (\u2264 1 SE = {original_se:.4f})"
-    else:
-        detail = (
-            f"estimate shifted by {shift:.4f}  (> 1 SE = {original_se:.4f})  "
-            f"Adding a random common cause destabilised the RDD estimate."
-        )
-    return RefutationCheck(name="Random common cause", passed=passed, detail=detail)
+    return _shift_check(
+        new_effect, original_effect, original_se, "Adding a random common cause destabilised the RDD estimate."
+    )
 
 
 class RDDRefutationReport(RefutationReport):

{formative_ds-2.2.1 → formative_ds-2.2.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "formative-ds"
-version = "2.2.1"
+version = "2.2.2"
 description = "A Python library for quantitative reasoning."
 readme = "README.md"
 requires-python = ">=3.10"
@@ -48,7 +48,7 @@ docs = ["sphinx>=7,<10", "sphinx-rtd-theme>=2,<4", "sphinx-copybutton>=0.5,<0.6"
 packages = ["formative"]
 
 [tool.bumpversion]
-current_version = "2.2.1"
+current_version = "2.2.2"
 commit = true
 tag = true
 tag_name = "v{new_version}"