Lead Pipe Risk Model — Methodology v1.0

Model version: v1.0 · Computed: 2026-07-11 · Grade: postcode-level screening

This model estimates the likelihood that properties in a postcode are served by

lead water pipes. It is a screening indicator: it identifies areas where lead

pipes are more likely, using property age and water-quality evidence. It does

not identify individual properties with lead pipes ("dig-here" grade requires

company asset data, which we are pursuing separately via environmental

information requests).

1. Data sources

2. Property preparation (EPC)

• All 15 lodgement-year certificate files streamed from the bulk zip.
• Dedupe: one row per property, keeping the latest lodgement_date.

Property key = uprn where present (~99% of certificates); otherwise a

composite of normalised postcode + address string.

23,579,971 certificates → 18,204,593 unique properties.

• Postcodes normalised (uppercase, single space) and validated against ONSPD.

4,093 of 1,381,195 EPC postcodes (0.3%) did not match ONSPD and are excluded.

3. Construction age mapping

construction_age_band values are mapped to two weights: the share of the band

pre-dating 1970 (the practical end of lead pipe installation; lead was banned

for this use in 1969) and pre-dating 1945 (the highest-prevalence era).

Unmapped values were negligible: 19 distinct garbage values covering **92

properties** out of 18.2M (excluded and logged).

Per postcode we compute: cert_count, pre1970_pct, pre1945_pct, and

flat_pct (flats/maisonettes share — recorded for future use; **not weighted

in v1**, as communication-pipe dynamics for flats are not yet modelled).

4. Zone lead evidence

Each postcode maps to its water supply zone (exact postcode match, postcode-

sector fallback). Join coverage: 99.6% of EPC-evidenced postcodes.

zone_lead_signal (0–1) per zone:

zone_lead_signal = 0.7 × min(1, lead_max / 10) + 0.3 × min(1, lead_mean / 2.5)
if the zone has an active lead enforcement notice: + 0.2 (capped at 1.0)

• 10 µg/l is the UK regulatory standard for lead at the tap; 2.5 µg/l scales

the mean to flag zones with consistently elevated background lead.

• Calibration note (documented deliberately): lead enforcement notices are

recorded at company level and propagate to 75% of zones (1,217 of 1,614).

An earlier draft floored the signal at 0.75 for noticed zones; because the

flag is so widespread this made the zone component near-constant nationally

and inflated the High band to 45% of postcodes. v1.0 instead treats a notice

as a modest +0.2 additive bump so that measured lead concentrations remain

the primary zone evidence.

• 1,596 of 1,614 zones (98.9%) provide a usable signal.

5. Score (v1.0 formula)

age_likelihood  = 0.7 × pre1970_pct + 0.3 × pre1945_pct          (0–1)
lead_risk_score = 70 × age_likelihood + 30 × zone_lead_signal    (0–100)

Additive-only rule: the zone evidence component can only add to the

score. The property-age components alone set the score floor; missing or low

zone data never reduces an age-based score. (A postcode with wholly pre-1945

housing scores at least 70 regardless of zone data.)

Postcodes with no EPC evidence receive a zone-only fallback score

(30 × zone_lead_signal) and are flagged confidence = "None".

Bands

The documented thresholds (65/35) produced a sane national distribution and

were retained unchanged. The High band is deliberately sensitive — it flags

postcodes dominated by pre-1970 (especially pre-1945) housing, where lead

communication or supply pipes are plausible for many properties, not confirmed

for any specific one.

Confidence

6. Validation (run 2026-07-11)

• Distribution: High 34.3% / Medium 28.6% / Low 37.1% (EPC-evidenced);

below the 50%-High sanity threshold.

• Face validity: Victorian/Edwardian districts vs new-build districts —

average scores: Sheffield S6 50.5, York YO1 52.4, Leeds LS6 57.4, Birmingham

B13 58.6, Battersea SW11 59.5 vs Milton Keynes MK4 11.2, MK8 13.0, Cambourne

CB23 22.4, Olympic Park E20 16.6. Example postcode contrasts (verified in

D1): LS6 1AG scores 80.4 (High; 41 certs, 100% pre-1945) and SW11 2AG scores

100 (High; zone at the 10 µg/l standard) vs MK10 7AA at 11.4 (Low; 28 certs,

0% pre-1970) and CB23 1ND at 5.1 (Low).

• Magnitude cross-check (Anglian Water): Anglian discloses ~514,000

households (~9% of connections) with lead supply pipes. In Anglian zones our

screen puts 24.3% of postcodes in High (54.4% High+Medium). These are

different units (postcodes flagged as elevated-likelihood areas vs households

actually having lead), and lead pipes concentrate precisely in old-stock

postcodes, so a screening rate ~2.5× the household prevalence is the expected

relationship for a sensitivity-first screen. It is not evidence that a

quarter of properties have lead.

• Coverage: 1,498,175 postcodes total; 202,368 (13.5%) have no EPC evidence

and carry zone-only fallback scores. 98.6% of all rows carry a zone signal.

7. Known limitations

• 1. Screening grade. Postcode-level likelihood, not property-level fact.
• 2. EPC coverage ≈ 60–65% of the housing stock, biased toward properties

transacted, let, or built since 2008. Older owner-occupied homes — the very

properties most likely to retain lead pipes — are under-represented, so

pre1970_pct in old areas is more likely under- than over-stated.

• 3. England & Wales only. EPC does not cover Scotland or Northern Ireland.
• 4. Zone evidence is zone-wide, not property-specific; zone lead means/maxes

come from regulatory sampling which is partly random-daytime and partly

targeted.

• 5. Construction age describes the building, not the pipework: renovated

properties may have replaced lead; some post-1970 infill shares old

communication pipes.

• 6. Enforcement-notice flags are company-level (see §4 calibration note).

8. Planned v2 inputs (schema join points already in place)

• phosphate_dosed — orthophosphate dosing by works/zone (reduces plumbosolvency;

a dosed zone with old stock is lower acute risk than an undosed one).

• company_lead_estimate_pct — water companies' own lead-pipe population

estimates from our live EIR requests (e.g. Anglian's 514k disclosure).

• DWI lead compliance failure records at sample-point level.
• Flat/communication-pipe weighting once evidence supports a coefficient.

*Produced by MyTapWater. EPC data © Crown copyright (OGL v3). Contains OS,

Royal Mail and ONS data (OGL v3). Zone water-quality data compiled from water

company annual compliance publications.*

← Back to the Lead Pipe Risk Checker