Research index · Open Government Licence
UK MOT data
PriceMyRepair’s UK MOT data studies are built from the full national testing record — 829.7 million tests, 697.8 million recorded defects, every car and van in Great Britain. Nine studies, one dataset, one method, published here in full.
DVSA bulk datasetOpen Government Licence v3.0Free to cite
Every figure in PriceMyRepair’s MOT studies comes from one source: the DVSA MOT testing bulk dataset, covering 829.7 million tests on 137.8 million vehicles in Great Britain, analysed by PriceMyRepair. We publish the method behind each number, the sample size under every table, and — unusually — the four analyses we ran and decided not to publish.
829.7m
MOT tests in the dataset, on 137.8 million individual vehicles
166.2m
advisories followed to the same vehicle’s next test, to see what happened to each one
9
studies published from it, each answering a different question
The nine studies
PriceMyRepair has published nine studies from the DVSA MOT record. Each takes a different cut of the same data. None repeats another’s table, and each names the sample behind every figure it publishes.
The most reliable used cars
How often a car fails and what each failure costs turn out to be independent of each other. Four kinds of car rather than one ranked list.
Correlation between the two axes: 0.045 Fuel and mileageDiesel vs petrol reliability
The same failure rate at the same age, reached on very different odometers. The only study here with a comparison that needs no adjustment at all.
Both fail 33.2% at fourteen years — diesel on 28% more miles By modelMOT failure rates by car
144 models compared at exactly ten years old, because rankings that mix ages mostly measure how old each nameplate’s fleet is.
37.8% at worst against 17.7% at best First testFirst MOT failure rates
The one comparison age cannot distort, because every vehicle taking a first test is three years old by law. Cars and vans counted separately.
15.6% fail — vans 23.0%, cars 14.6% What drivers doMOT advisory statistics
What happens to a warning after it is written. The categories drivers skip are not the ones that turn out to matter least.
33.4% of advisories are never repaired By categoryWhat ignoring an advisory costs
Eight defect categories ranked by how often each returns as a failure, multiplied by what the repair costs when it does. Category averages; the wording-level figures are in the ninth study below.
Single wordings range from 0.3% to 18.6% inside those averages By costThe most expensive MOT failures
Every failure item priced against the cost of the test itself — including the one we refuse to put a number on.
The dearest costs 27 times the test By ageCar repair costs by age
Risk does not climb forever. Everything that wears out peaks and then falls away; the one thing that rusts never does.
Wear peaks at fifteen years, corrosion never turnsThe ninth study: every advisory wording, followed one by one. The eight above work by defect category. The ninth takes the exact phrases testers write — “suspension arm ball joint has slight play”, “brake fluid close to minimum mark”, seventy of them — and follows each to the same vehicle’s next test, with the odds of failing on that same part, the tester notes, the age curve and the models most affected. It is published as the MOT advisory decoder and as one page per wording. The odds run from 0.3% to 18.6%, which is why a category average is not a figure for any one line on a certificate.
Four things we ran and did not publish
A dataset this large will produce a striking result for almost any question, and most of those results are artefacts. These four looked publishable and did not survive testing.
Ranking models by what their failures cost produced a 2.0× spread — until we noticed that our own price matrix applies multipliers from 0.85 to 1.30 by vehicle class, which is 1.53× of that built in before any data is touched. The ranking was largely our pricing quoting itself back. Zero the multipliers and small hatchbacks appear above premium saloons, which nobody would believe and nor should they.
One badge topped the table among cars over twelve years old and sat sixth among cars under seven, where a different badge led. An ordering that does not reproduce across age bands is measuring which manufacturers’ cars survive long enough to appear in the older group. We publish the shape that does survive — every manufacturer’s rate rises with age, and the gaps between them narrow — and not the league table.
The share of each model still on the road looked like a durability ranking. It is a ranking of how recently each model stopped being sold: the top is filled with nameplates barely a decade old and the bottom with cars discontinued forty years ago. The dataset carries no year of first registration, so there is no way to normalise it. Dropped entirely rather than published with a caveat.
Failures per ten thousand miles falls steadily as cars age, which reads like a finding and is arithmetic. The mileage figure is lifetime rather than annual, and it rises faster than the failure rate does, so the ratio must fall whatever the vehicles are doing. We use the metric to compare fuels within a single age, where the denominator is shared, and never to describe a change over time.
Each of these took between twenty minutes and several hours to produce and about the same again to disprove. We publish them because a study that shows only what worked gives a reader no way to judge how hard the numbers were tested.
The dataset and the method
Every study shares this. Where one departs from it, the departure is stated on that page.
Source and coverage
The DVSA MOT testing bulk dataset, published by the Driver and Vehicle Standards Agency under the Open Government Licence v3.0 and analysed by PriceMyRepair. Our processed extract covers 829,654,015 tests and 697,809,455 recorded defects across 137,840,356 vehicles, filtered to tests recorded by DVSA with a pass or fail result. Northern Ireland operates a separate system under the DVA and is not included anywhere.
Defect records start later than test records. The current MOT inspection manual came into use partway through the period covered, and defects recorded under the previous scheme are excluded — around 1.18 billion of them. Test counts cover the full period. Any figure dividing defects by tests is therefore understated, and understated more for models with longer histories, which is why several studies here deliberately avoid that form.
Make and model are canonicalised. The raw record contains 247,624 distinct make and model strings, many of them variant spellings of the same vehicle. These resolve to 97,349 canonical models. Vehicles whose make could not be resolved are excluded from model-level tables, which is why per-model advisory totals are about 4.5% below the national figure.
Age is the controlling variable. Comparing vehicles of different ages mostly measures how old each group’s fleet is, so almost every comparison here holds age fixed. Where a study reports several ages, they are reported separately and never averaged into one number.
Advisory wordings are tracked one by one. The advisory pages and the decoder use a second processing of the same record, restricted to tests with a pass or fail result under the current inspection manual: 350,943,111 tests and 442,101,206 advisories carrying a DVSA defect code, matched to the same vehicle’s next test between 180 and 730 days later, giving 166,165,845 advisory-to-next-test pairs. “Fails on the same part” means a major or dangerous defect at that next test under the same defect code and naming the same component, on either side of the vehicle; because testers do not always name a component identically twice, it is a floor. “Fails under the same code” counts any defect under the same three-level code. “Advised again” means the same wording recorded at the next test. We do not publish a “fixed” rate, because the remainder includes vehicles that failed on something else or left the record, and we do not publish a median time to failure, because on an annual test it is the length of a year for every wording and says nothing.
Repair prices are ours, not the DVSA’s. Where a study puts a pound figure against a defect, that figure comes from our own independent-garage price ranges, not from the dataset. The two sources are combined on our pages and nowhere in the official record. Prices are estimates and final quotes come from garages.
Sample thresholds. No model appears in a table without a stated minimum number of tests, and the threshold is given on each page because it differs with the age band — a three-year-old fleet is necessarily smaller than a ten-year-old one. Advisory pages rank models only above 60,000 tests.
Contains public sector information licensed under the Open Government Licence v3.0.
Using and citing this work
Anyone may quote these figures. Journalists, researchers and other sites are welcome to use any number published here, with a link to the study it came from. We do not ask for approval and we do not require anything beyond attribution.
How to attribute. The simplest accurate form is “PriceMyRepair analysis of DVSA MOT data”, with a link to the specific study. The numbers are the government’s record; the age control, the canonicalisation and the price layer are ours, so naming both sources describes the work correctly.
The underlying data is public and free. The DVSA bulk dataset is available to anyone under the Open Government Licence — nothing here is proprietary. What we add is the processing, the age control and the price layer, all of which are described above so the work can be checked or repeated.
If a figure looks wrong, we would like to know. Several numbers on this site have already been corrected after checking, and two whole studies were withdrawn before publication for the reasons set out further up this page. The largest correction so far: our advisory pages originally quoted defect-category averages as if they applied to a single wording, which overstated the risk for most wordings by two to three times. They now carry the figure for that wording alone. Corrections are made on the page rather than quietly.
These studies describe millions of vehicles. Enter a registration and we’ll read the full test history for that one — every advisory, every failure, priced for that model.
About this data — FAQ
Where does this MOT data come from?
The DVSA MOT testing bulk dataset, published by the Driver and Vehicle Standards Agency under the Open Government Licence v3.0. It is free for anyone to download and use. Our extract covers 829.7 million tests on 137.8 million vehicles in Great Britain; Northern Ireland runs a separate system and is not included.
Can I quote these figures in an article?
Yes, with a link to the study the figure came from. No permission is needed and we ask for nothing else. The underlying dataset is public, so anyone can check or reproduce the processing — the method behind each number is set out on this page and on each study.
Does this cover Northern Ireland?
No. Vehicle testing in Northern Ireland is run by the Driver and Vehicle Agency under a separate system, and those records are not part of the DVSA dataset. Every figure on this site describes Great Britain only.
Why do some totals differ between your studies?
Because different cuts of the same record include different things. Tables built per make and model exclude vehicles whose make could not be resolved to one canonical name, which is about 4.5% of advisories. The advisory pages use a second processing restricted to tests under the current inspection manual, which is why they cite 350.9 million tests rather than 829.7 million. Where a study uses a smaller total than the national one, it says so and explains the gap.
Are the repair prices part of the government data?
No. The DVSA dataset records what failed, never what it cost. Every pound figure on this site is our own independent-garage estimate, combined with the official data on our pages and nowhere in the official record. Prices are ranges rather than quotes, and a garage’s figure is the one that counts.
Why publish the analyses you rejected?
Because a dataset this size will produce a striking result for almost any question asked of it, and most of those results are artefacts of how the data was collected. Showing which ones we threw away, and why, is the only way a reader can judge how hard the published numbers were tested.
Tools built on the same data
The studies describe the fleet. These read one vehicle.