Methodology

In plain English: we take every official MOT test from 2024 (52 million of them), count how often each car family failed at its first attempt, and then ask one fair question: is that more or less than you'd expect, given how old those cars are? Old cars fail more — that's normal. Our index is 100 when a family fails exactly as much as its age predicts; 118 means 18% more than expected; 76 means 24% less. Everything below explains the exact maths behind that one idea.

This site computes vehicle-family statistics from the DVSA "Anonymised MOT tests and results" bulk dataset (Open Government Licence v3.0). Current data release: dvsa-2024-v1— calendar-year 2024, 66,857,355 test records. Every published figure carries aclaim_id that maps to a specific script in our versioned build pipeline; nothing is hand-edited and no generative model writes or alters any number on this site.

1. What counts as a test, and what counts as a failure

We use initial tests only (DVSA test type NT) for class-4 vehicles (cars), outcomes P (pass), F (fail) and PRS. PRS means the vehicle did not pass its initial examination: a failing defect was found and rectified at the test station within an hour. Our headline initial-failure rate therefore counts F + PRS — the share of cars that arrived at the test with at least one failing defect. The stricter "walked away with a fail certificate" rate (F only) is shown alongside. Retests are excluded from all headline rates and analysed separately.

2. The index: age-adjusted MOT initial-failure ratio

Raw failure rates punish models with older fleets. The index corrects for this by comparing a family's observed initial failures with the number expected for its exact age profile:

Index = 100 × observed initial failures ÷ Σ (tests in age stratum × national stratum rate)

Age strata: under-3, single years 3–19, 20-plus, unknown. No clamping of extreme ages.
The reference population is all class-4 initial tests in the same window (2024), frozen per data release. When a new release is published, previous baselines are preserved — time-comparisons are only made against frozen baselines.
100 = exactly the failure level expected for that fleet's age profile. Below 100 = fewer failures than age predicts; above = more.
Confidence intervals are computed on the index itself (binomial interval on the observed count scaled by the frozen expected value). Where the 95% CI includes 100, the verdict is labelled directional and no colour rating is assigned.

The index is a test-outcome statistic, not a general "reliability" score: the MOT does not examine engine internals, gearboxes, air conditioning or infotainment. Residual confounders not yet adjusted for: annual mileage, fleet/private use mix, fuel mix within a family. Mileage stratification is planned; until then, per-fuel splits are shown wherever the spread between powertrains exceeds 5 percentage points.

3. National baseline (release dvsa-2024-v1)

Age	Initial-failure rate
3	10.9%
4	11.6%
5	13.5%
6	15.6%
7	18.2%
8	21.3%
9	24.4%
10	27.7%
11	30.6%
12	33.4%
13	36.0%
14	38.6%
15	39.9%
16	41.5%
17	41.5%
18	42.4%
19	42.2%
20+	37.5%

4. Grain: what a "family" is

Pages aggregate at model-family level (e.g. all Fiesta variants), built by an explicit published mapping from DVSA's free-text make/model fields. We do not split by generation yet — the word "generation" is deliberately absent from data pages until a validated generation-resolver exists. Where powertrains within a family diverge materially, a per-fuel table is shown.

5. Suppression and small samples

Families with fewer than 1,000 initial tests are not published. Cells (any published claim's numerator or denominator) below threshold are suppressed, with secondary suppression applied to downloadable data so suppressed cells cannot be reconstructed by subtraction. Strata with 1,000–5,000 tests carry a low-confidence marker.

6. Known caveats of the source data

Defect severity fields (dangerous/major/minor) are absent from the 2018–2021 releases; severity analysis uses releases where present.
Unknown first-use dates are encoded by DVSA as 01/01/1971 — we treat these as "unknown" and exclude them from age curves (343 of 52,245,925 class-4 initial tests in 2024, i.e. 0.0007%; the issue is more material in older releases).
Hybrid fuel coding was corrected by DVSA only from the 2022 dataset onward.
Make/model are free-text fields with typos and trim tails; our mapping is conservative and published.
DVSA's own compliance survey (2021–22, repeated 2023–24) estimated that roughly 1 in 10 vehicles receiving a pass should have failed — and that errors run in both directions. This bounds the precision of any MOT-derived statistic, ours included.

Bounds of applicability

These statistics describe MOT test outcomes of vehicle populations in Great Britain, calendar 2024. They are not: predictions for any individual vehicle; measures of breakdown risk (a different construct — see ADAC-style data); or comparable across countries without harmonisation. Review date for this release's claims: 2026-10-30 (quarterly).

7. Reproducibility

Each page shows its dataset_release_id. The full computation chain — cleaning rules (embedded header rows, quote escaping), family mapping, index formula, CI method — is versioned in git and pinned per release. A claim registry mapping every published figure class to its script ships with each release.