Stop deciding a household's UC health element rate by its position

changelog.d/2250.fixed.md

@@ -0,0 +1 @@
+- Let a household situation state whether its Universal Credit health element is at the new-claimant rate (`uc_receives_new_claimant_health_element`, default false), so a household's award no longer depends on its position among the other benefit units in a simulation; simulations built from data keep the seeded draw.

docs/book/policy/uc-rebalancing.md

@@ -24,7 +24,7 @@ Existing recipients are treated differently. Universal Credit Regulations 2013 r
 
 For a tax year without a legislated amount, the model uses the lowest amount the Universal Credit Act 2025 s. 4 duty allows. For every standard allowance amount, the protected amount plus that allowance must be at least the previous year's sum, increased by the relevant CPI percentage: the CPI 12-month rate in the September before the tax year, never below 0% (s. 4(4)(a)). The model reads that rate from `gov.economic_assumptions.statutory_uprating_inputs.cpi_september`. Section 3 switches off the element's ordinary uprating, so the amount otherwise stays where it was. Run from the 2025-26 amounts at the 3.8% September 2025 CPI rate, this rule gives the legislated £429.80.
 
-The implementation uses transition probabilities based on WPI Economics analysis for the Trussell Trust, derived from administrative Personal Independence Payment data. The probability of being a new claimant varies by year:
+Whether a benefit unit gets the new-claimant rate is the benefit unit's `uc_receives_new_claimant_health_element` for the year. A household situation can set it, for example `{"uc_receives_new_claimant_health_element": {"2026": True}}` on the benefit unit; otherwise it is false, which pays the protected amount. A dataset can supply it too. A simulation built from data that does not supply it assigns it by a seeded draw, using transition probabilities based on WPI Economics analysis for the Trussell Trust, derived from administrative Personal Independence Payment data. The probability of being a new claimant varies by year:
 
 - 2026: 11%
 - 2027: 13%

policyengine_uk/scenarios/uc_reform.py

@@ -95,6 +95,33 @@ def _protected_existing_health_element_monthly(
     )
 
 
+NEW_CLAIMANT_STATUS = "uc_receives_new_claimant_health_element"
+
+
+def _new_claimant_health_element_status(
+    sim: Microsimulation, year: int, drawn: np.ndarray, built_from_data: bool
+) -> np.ndarray:
+    """Which benefit units get the new-claimant LCWRA rate in a year.
+
+    A value the situation or dataset supplies is used as given. Data without
+    one takes the seeded draw, recorded as an input. A household situation
+    without one gets the protected amount (the variable's default): a draw
+    there would make a household's award depend on its position among the
+    other benefit units in the simulation.
+    """
+    # sim.get_array reads the simulation's own branch (and the branches it
+    # inherits from); the holder's get_array reads only the default branch.
+    supplied = sim.get_array(NEW_CLAIMANT_STATUS, year)
+    if supplied is not None:
+        return np.asarray(supplied, dtype=bool)
+    if built_from_data:
+        status = np.asarray(drawn, dtype=bool)
+    else:
+        status = np.zeros(len(drawn), dtype=bool)
+    sim.set_input(NEW_CLAIMANT_STATUS, year, status)
+    return status
+
+
 def add_universal_credit_reform(sim: Microsimulation):
     rebalancing = sim.tax_benefit_system.parameters.gov.dwp.universal_credit.rebalancing
 
@@ -109,10 +136,13 @@ def add_universal_credit_reform(sim: Microsimulation):
         2029: 0.22,
     }  # WPI Economics for Trussell Trust based on admin PIP data, 2025
     new_claimant_health_element = rebalancing.new_claimant_health_element
+    built_from_data = getattr(sim, "built_from_dataset", False)
     for year in range(2026, 2030):
         if not rebalancing.active(year):
             continue
-        is_post_2025_claimant = uc_seed < post_2025_claimant_share[year]
+        is_post_2025_claimant = _new_claimant_health_element_status(
+            sim, year, uc_seed < post_2025_claimant_share[year], built_from_data
+        )
         # Copy: calculate returns the simulation's cached array, and the writes
         # below must reach the cache only through set_input.
         current_health_element = np.array(sim.calculate("uc_LCWRA_element", year))

policyengine_uk/tests/test_uc_rebalancing.py

@@ -261,19 +261,112 @@ def test_section_4_floor_rises_with_the_previous_amount(
     ) >= uc_reform.protected_lcwra_floor(previous, before, now, factor)
 
 
-def test_new_claimants_use_fixed_health_element(monkeypatch):
-    _force_uc_seed(monkeypatch, [0.0])
-    sim = Simulation(situation=_uc_claimant(30))
+def _new_claimant_rate(sim: Simulation, year: int) -> float:
+    return float(
+        sim.tax_benefit_system.parameters(
+            str(year)
+        ).gov.dwp.universal_credit.rebalancing.new_claimant_health_element
+    )
+
+
+def test_new_claimants_use_fixed_health_element():
+    situation = _uc_claimant(30)
+    situation["benunits"]["benunit"]["uc_receives_new_claimant_health_element"] = {
+        year: True for year in YEARS
+    }
+    sim = Simulation(situation=situation)
 
     for year in range(2026, 2030):
         health_element = sim.calculate("uc_LCWRA_element", year)[0] / 12
-        expected_health = float(
-            sim.tax_benefit_system.parameters(
-                str(year)
-            ).gov.dwp.universal_credit.rebalancing.new_claimant_health_element
+        assert health_element == pytest.approx(_new_claimant_rate(sim, year))
+
+
+def _many_uc_claimants(count: int) -> dict:
+    people = {
+        f"person_{i}": {
+            "age": {year: 30 for year in YEARS},
+            "employment_income": {year: 0 for year in YEARS},
+            "uc_limited_capability_for_WRA": {year: True for year in YEARS},
+        }
+        for i in range(count)
+    }
+    return {
+        "people": people,
+        "benunits": {
+            f"benunit_{i}": {"members": [f"person_{i}"]} for i in range(count)
+        },
+        "households": {
+            f"household_{i}": {"members": [f"person_{i}"]} for i in range(count)
+        },
+    }
+
+
+def test_household_award_does_not_depend_on_its_position():
+    # Before the fix, a seeded draw over the benefit units in the simulation
+    # gave about 11% of these identical households the new-claimant rate in
+    # 2026, depending only on their position.
+    sim = Simulation(situation=_many_uc_claimants(40))
+    alone = Simulation(situation=_uc_claimant(30))
+
+    for year in range(2026, 2030):
+        health_elements = sim.calculate("uc_LCWRA_element", year)
+        assert np.all(health_elements == health_elements[0])
+        assert health_elements[0] == pytest.approx(
+            alone.calculate("uc_LCWRA_element", year)[0]
         )
+        assert health_elements[0] / 12 > _new_claimant_rate(sim, year)
+
+
+def _dataset(new_claimant_status=None):
+    from policyengine_uk.data import UKMultiYearDataset, UKSingleYearDataset
+    import pandas as pd
+
+    benunit = {"benunit_id": [1, 2]}
+    if new_claimant_status is not None:
+        benunit["uc_receives_new_claimant_health_element"] = new_claimant_status
+    single_year = UKSingleYearDataset(
+        person=pd.DataFrame(
+            {
+                "person_id": [1, 2],
+                "person_benunit_id": [1, 2],
+                "person_household_id": [1, 2],
+                "age": [30, 30],
+                "uc_limited_capability_for_WRA": [True, True],
+            }
+        ),
+        benunit=pd.DataFrame(benunit),
+        household=pd.DataFrame(
+            {
+                "household_id": [1, 2],
+                "household_weight": [1.0, 1.0],
+                "region": ["NORTH_EAST", "NORTH_EAST"],
+            }
+        ),
+        fiscal_year=2026,
+    )
+    return UKMultiYearDataset(datasets=[single_year])
+
+
+def test_data_without_a_status_keeps_the_seeded_draw(monkeypatch):
+    _force_uc_seed(monkeypatch, [0.0, 0.99])
+    sim = Simulation(dataset=_dataset())
+
+    health_elements = np.asarray(sim.calculate("uc_LCWRA_element", 2026)) / 12
+    assert health_elements[0] == pytest.approx(_new_claimant_rate(sim, 2026))
+    assert health_elements[1] == pytest.approx(429.80)
+    assert list(sim.calculate("uc_receives_new_claimant_health_element", 2026)) == [
+        True,
+        False,
+    ]
+
 
-        assert health_element == pytest.approx(expected_health)
+def test_data_that_supplies_the_status_overrides_the_draw(monkeypatch):
+    _force_uc_seed(monkeypatch, [0.0, 0.99])
+    sim = Simulation(dataset=_dataset(new_claimant_status=[False, True]))
+
+    health_elements = np.asarray(sim.calculate("uc_LCWRA_element", 2026)) / 12
+    assert health_elements[0] == pytest.approx(429.80)
+    assert health_elements[1] == pytest.approx(_new_claimant_rate(sim, 2026))
 
 
 def test_standard_allowance_reforms_still_change_standard_allowance(monkeypatch):
@@ -292,3 +385,26 @@ def test_standard_allowance_reforms_still_change_standard_allowance(monkeypatch)
     reformed_standard_allowance = reformed.calculate("uc_standard_allowance", 2026)[0]
 
     assert reformed_standard_allowance / baseline_standard_allowance > 1.5
+
+
+def test_a_branch_can_state_its_own_status():
+    sim = Simulation(situation=_uc_claimant(30))
+    parent = sim.calculate("uc_LCWRA_element", 2026)[0]
+    branch = sim.get_branch("new_claimant")
+    branch.set_input("uc_receives_new_claimant_health_element", 2026, [True])
+    branch.delete_arrays("uc_LCWRA_element", 2026)
+    uc_reform.universal_credit_july_2025_reform.simulation_modifier(branch)
+
+    assert branch.calculate("uc_LCWRA_element", 2026)[0] / 12 == pytest.approx(
+        _new_claimant_rate(sim, 2026)
+    )
+    assert sim.calculate("uc_LCWRA_element", 2026)[0] == pytest.approx(parent)
+
+    # A branch of that branch inherits its status.
+    nested = branch.get_branch("nested")
+    nested.delete_arrays("uc_LCWRA_element", 2026)
+    uc_reform.universal_credit_july_2025_reform.simulation_modifier(nested)
+    assert nested.calculate("uc_LCWRA_element", 2026)[0] / 12 == pytest.approx(
+        _new_claimant_rate(sim, 2026)
+    )
+    assert sim.calculate("uc_LCWRA_element", 2026)[0] == pytest.approx(parent)

policyengine_uk/variables/gov/dwp/universal_credit/disability_element/limited_work_ability/uc_receives_new_claimant_health_element.py

@@ -0,0 +1,19 @@
+from policyengine_uk.model_api import *
+
+
+class uc_receives_new_claimant_health_element(Variable):
+    value_type = bool
+    entity = BenUnit
+    label = "Universal Credit LCWRA element at the new-claimant rate"
+    documentation = (
+        "Whether the award's LCWRA element is paid at the rate for claimants "
+        "other than a pre-2026 claimant, a severe conditions criteria claimant "
+        "or a claimant who is terminally ill (UC Regs 2013 reg. 36, from 6 "
+        "April 2026). False pays the protected LCWRA amount. A household "
+        "situation can set it; otherwise it is false. Simulations built from "
+        "data that do not supply it assign it by a seeded draw at the share of "
+        "health-element claimants expected to be new claimants each year."
+    )
+    definition_period = YEAR
+    default_value = False
+    reference = "https://www.legislation.gov.uk/uksi/2013/376/regulation/36"