Complaints per Recall: The Ratio, and Why the Obvious Reading Is Wrong

Every model page on this site carries two federal numbers: how many owner complaints a nameplate has attracted, and how many recall campaigns cover it. Readers write in asking the obvious question, which is what happens if you divide one by the other.

It is a good instinct. A complaint is an unverified report from a driver; a recall is a defect the manufacturer accepted and paid to fix. The ratio between them looks like a measure of how much trouble went unacknowledged — and that would be genuinely useful, because nothing else in the public record gets near it.

This page works the calculation properly. It takes longer than expected, because the two obvious ways of doing it produce two different wrong answers.

What each number is actually counting

Start with the halves, because the difference between them is the whole point.

An owner complaint is a report filed with the federal safety regulator by somebody who owns the car. Nobody verifies it. It costs the filer nothing but time, it can describe anything from a catastrophic failure to a rattle, and it arrives only if the owner knows the database exists. Our guide to reading a vehicle’s federal record sets out why the shape of a complaint file matters more than its size.

A recall campaign is the opposite kind of object. The manufacturer — or the regulator, over its objection — has concluded that a defect exists, defined the population, and committed to fixing it free of charge. It is an admission with a cost attached, and our guide to checking a car for open recalls covers how to see whether a specific vehicle’s fix was ever done.

So the ratio is asking: for every fault this manufacturer accepted, how many things did owners complain about? A high number sounds like under-acknowledgement. Hold that thought.

The first trap: the denominator counts the wrong thing

Recall data is published per model year. A campaign that covers, say, a transmission fault in vehicles built across six model years appears in six of those years’ records — correctly, because a buyer looking at one year needs to know it applies to theirs.

Add those per-year figures together across a nameplate’s whole life and you are no longer counting campaigns. You are counting campaign-years. One recall becomes six.

The effect is not small and it is not even. Deduplicating on the campaign identifier across our tracked set turns 3,062 apparent campaigns into 1,227 actual ones. Per model the inflation runs from 1.5 times at the low end to 3.2 times at the high end — so it does not cancel out when you compare two models, it distorts the comparison.

The chart above uses distinct campaign identifiers. Anything computed from summed per-year totals — including some figures elsewhere on this site — is measuring campaign-years.

The second trap: dividing does not cancel fleet size

The more seductive error, because it sounds like arithmetic.

The reasoning goes: complaint counts are inflated by how many cars were sold, and so are recall counts, so dividing one by the other cancels the sales term and leaves something comparable. It is the argument I would have made, and it is wrong.

Complaints scale with fleet size fairly directly — more cars on the road, more owners with something to report. Recall campaigns do not. A campaign is a decision about a defect, and one campaign covers however many vehicles are affected, whether that is eight thousand or eight million. Doubling the fleet roughly doubles the complaints and leaves the campaign count almost unchanged.

The two halves have different elasticities, so the division leaves a residue of fleet size rather than removing it. Big sellers drift up the ratio for reasons that have nothing to do with how they were built.

The measure that works better, and why nobody uses it

There is a cross-model comparison in this data that does not have a denominator problem, and it is not a ratio of counts. It is severity expressed as a share of a model’s own complaint file.

Every complaint record carries flags for whether a crash, a fire, an injury or a death was involved. Expressed per thousand complaints on the same nameplate, those shares are internally normalised — the fleet-size term appears in the numerator and the denominator of the same file, and genuinely does cancel, because both halves are counts of the same kind of object.

Run it and the results are more interesting than the ratio ever gets. The Toyota Camry carries 158.0 crash-involved complaints per thousand against the Honda Accord’s 73.7, and 4.0 deaths per thousand against 1.5 — on a file that is smaller in absolute terms. The Ford F-150’s complaint file is 3.9 per cent fire-involved; the Toyota Tacoma’s is 1.3 per cent.

Two cautions before anybody quotes those. Severity shares reflect what owners chose to report as much as what happened, and a model whose problems attracted press coverage draws a different mix of reports. And a share of a file says nothing about your odds of anything — there is still no denominator of vehicles on the road anywhere in this dataset.

It is a better instrument than the ratio at the top of this page, and it appears almost nowhere, because it produces uncomfortable comparisons between cars with settled reputations rather than a tidy ranking.

The test the ratio fails

There is a straightforward way to check whether the ratio measures manufacturer behaviour, and it does not survive it.

If a high score meant a company that under-acknowledges faults, models from one manufacturer should cluster. They do not. Toyota holds the third-highest score in our set with the Camry at 144.8 complaints per campaign-year, and the lowest with the 4Runner at 36.6 — a fourfold spread inside a single company, on the uncorrected figures.

Whatever the ratio is tracking, it is not a corporate disposition toward recalling things. It is at least as much about which model, which decades, and how many were sold.

The one recall signal that does not need arithmetic

Worth a section, because it is the counter-example to everything above: a recall category where the count is tiny, the meaning is unambiguous, and no ratio is required.

A handful of campaigns carry a do-not-drive or do-not-park-indoors instruction. These are issued when a defect can cause a fire or a failure severe enough that the regulator is not willing to let the vehicle wait for a workshop appointment. Across the six compact SUVs this site tracks, the Toyota RAV4 carries two such campaigns and the Ford Escape one; the CR-V, Forester, Rogue and Equinox carry none.

Those single digits tell a buyer more than any three-digit total on this page. A campaign of that class is a defect the manufacturer considered dangerous enough to ground the car, and whether the specific vehicle in front of you had that work done is checkable, free, in about a minute.

The general lesson: severity beats volume every time in this dataset, and severity is recorded in fields most summaries throw away.

What the corrected figures actually say

Recomputed on distinct campaigns, the leaderboard changes materially.

The Ford F-150 rises to the top at 351.1 complaints per campaign, followed by the Ford Escape at 305.0. The Chevrolet Equinox, which leads on the uncorrected numbers at 195.5, drops to third at 299.7. At the other end the Toyota 4Runner sits at 108 and the Toyota Highlander at 113.3.

Notice what happened to the Equinox. It has by far the fewest campaigns in the set — 30 distinct, against 122 for the Explorer — and a middling complaint file at 8,991. A small denominator makes a large quotient, and a small denominator is exactly what a short model history and a narrow recall record produce. The uncorrected figure flattered a statistical artefact into first place.

What the number is still good for

Having taken it apart, here is what remains, because something does.

Within a single body class, over comparable years, the ratio is a reasonable indicator of how far a model’s reported trouble ran ahead of its acknowledged trouble. It will not tell you the car is unreliable. It will tell you that a lot of owners had something to say relative to what the maker conceded, and that is a prompt to go and read the complaint file rather than a conclusion in itself.

The right use is diagnostic. A high ratio means: open the actual complaints, look at which system dominates, check whether crashes, fires or injuries are attached. Our per-model pages do that year by year — the F-150 and Equinox pages both work their files that way.

The wrong use is a league table, which is what every version of this calculation on the internet turns into.

Comparing within a class, which is the only fair version

If the ratio is to be used at all, it has to be used inside a body class over comparable years, because that is the only way to hold roughly constant the things it fails to correct for.

Two of the eighteen models here cannot join a lifetime comparison honestly: the Chevrolet Equinox is on record from 2005 and the Nissan Rogue from 2007, against 2000 for the other sixteen. Comparing lifetime totals across that gap rewards the newer nameplate for the years it simply was not on sale — a nineteen-year file will be smaller than a twenty-six-year one for reasons that have nothing to do with the car.

Within a class and within matched years, the exercise becomes defensible. Our comparisons of the midsize sedans take that approach at the level of a single model year, which removes the exposure problem entirely at the cost of a smaller sample. Both methods are legitimate; neither produces a league table.

None of this describes the car you are looking at

The limitation that outranks all the others.

Everything above is about nameplates across decades. The vehicle in front of you is one specific car, with one specific history: how it was driven, whether the recalls that applied to it were actually performed, whether it was crashed, flooded or written off. A nameplate’s aggregate record cannot tell you any of that, and a good example of a badly-scoring model will outlast a neglected example of a good one.

The checks that do describe your car are cheap and specific. The free federal recall lookup says whether that vehicle’s outstanding campaigns were completed. A full title and brand check on the number shows what insurers and titling agencies recorded about it. And an inspection reads the metal, which no database does.

What the files are about, which the totals hide

One more reason a single number per model is a poor summary: two models can carry similar totals about entirely different things.

Sort the F-150’s complaint citations by component and the top of the list is powertrain and engine — together a quarter and a seventh of its cited components. Do the same for the Tacoma and the leaders are structure and suspension, at roughly a quarter and a sixth. Those are not two versions of the same problem. One file is dominated by things that drive the truck, the other by things that hold it together, and a buyer inspecting each should be looking in different places.

The sedans differ the same way. The Camry’s largest single category is vehicle speed control; the Accord’s are electrical system and air bags, the latter reflecting an inflator recall that ran across the industry rather than anything peculiar to the model.

This is why the per-model pages exist and why they work year by year. A nameplate total answers no question a buyer actually has; a component profile for a specific generation answers several.

Reading any version of this statistic elsewhere

Four questions, since this calculation circulates widely and is almost never shown working.

Does the denominator count campaigns or campaign-years? If the source does not say, assume campaign-years, because that is what the raw data hands you.

Does it compare like with like? Models with different years on record are not comparable on lifetime totals — two of the eighteen here begin in 2005 and 2007 rather than 2000.

Does it claim to correct for sales? If so, ask where the sales figures came from. The federal complaint and recall databases do not contain any, which is why this page makes no per-vehicle claim of any kind.

Does it present the result as a reliability ranking? Then it has misunderstood both halves.

The denominator is also a corporate decision

There is a third distortion in this ratio that neither of the traps above accounts for, and it is arguably the most awkward of the three, because it cannot be corrected by better arithmetic.

A recall campaign is not a natural event. It is a decision, made by a manufacturer, about whether a pattern of failures crosses the threshold that warrants a formal campaign. Different manufacturers make that call differently, and the same manufacturer makes it differently in different decades, under different regulatory pressure and different legal advice.

Consider what that does to the quotient. A maker that acts early and issues a campaign at the first credible signal grows its own denominator, and the ratio falls. A maker that waits, investigates longer, or settles matters through service bulletins and goodwill repairs instead keeps its denominator small, and the ratio climbs.

Read naively, the cautious manufacturer therefore looks better on this statistic than the proactive one. That is close to the opposite of what a buyer wants the number to tell them.

This is not a hypothetical concern. Recall behaviour across the industry has changed markedly over the twenty-six years covered here, generally in the direction of more campaigns issued sooner, which means the ratio is not even stable within a single nameplate’s own history. A model on sale across that whole span carries early years scored under one regime and recent years under another.

None of which makes the figure worthless. It does mean the denominator carries a policy signal mixed in with the engineering one, and no amount of dividing will separate them.

The eighteen files this page is built from

Everything above is an aggregate of eighteen per-model records, and the aggregate is the least useful view of them. Each nameplate has its own page working its complaint file year by year, flagging the model years that stand out against their neighbours and the ones that are unusually quiet.

Compact SUVs: CR-V, RAV4, Forester, Escape, Rogue and Equinox. Sedans and compacts: Camry, Accord, Civic, Corolla. Larger SUVs: Highlander, Explorer, Grand Cherokee, 4Runner, Wrangler, Outback. Pickups: F-150 and Tacoma.

If you have arrived here from a search about which cars are most complained about, one of those eighteen is almost certainly the page you actually wanted — and once you have a specific car in mind, its own history record outranks every aggregate on this page.

Common questions

Does a high complaints-per-recall ratio mean a car is unreliable?

No. It means owners reported a lot relative to what the manufacturer conceded, which can happen because a model sold in huge numbers, because a problem got publicity, or because the complaints were about things no recall would ever cover.

Why is the Equinox first on some figures and third on others?

Because the uncorrected denominator counts campaign-years. The Equinox has the fewest distinct campaigns in the set, so deduplicating changes its quotient more than most, moving it from 195.5 to 299.7 and from first to third.

Is a manufacturer with more recalls worse?

Not straightforwardly. A recall is a conceded defect and a free fix — a maker acting on what it found. A long campaign list can describe a company that investigates thoroughly as easily as one that builds badly.

Why not just divide complaints by cars sold?

Because the federal databases hold no sales figures, and this site will not import a number it cannot source. Any page showing complaints per thousand vehicles has taken sales data from somewhere it should be naming.

Which is the more reliable number, complaints or recalls?

Recalls, for factual weight — each is a documented decision with a remedy attached. Complaints are richer but unverified, and useful mainly for their shape: which system dominates and whether harm is attached.

Does this tell me which used car to buy?

No, and that is the honest answer rather than a modest one. It tells you which complaint files are worth opening. The buying decision comes from that specific vehicle’s history and an inspection.

Where do these numbers come from?

The federal recall and complaint archives, pulled per model year and stored in this site’s own data files, which is why every figure here can be traced rather than taken on trust.

Why do so few sites show this calculation?

Because done carelessly it produces a satisfying league table, and done carefully it produces a caveat. The careful version is less shareable and considerably more accurate.