The Harm Ledger: Which Parts of a Car Actually Hurt People, by the Numbers

The Airbag Biography ended with a harm ledger for one component, and this article owes its existence to a promise made there: run the same ledger for everything. Every complaint in the register carries four harm fields, a crash flag, a fire flag, a count of injuries and a count of deaths, and across 2,166,651 complaints those fields have never been read together. Read them and the register splits into two files. One is enormous and harmless: the squeaks, the stalls, the warning lights, the file the ladder measures. The other is small and terrible, 209,502 complaints, 9.7 percent of everything ever filed, that allege a crash, a fire, an injury or a death. This is the file on the second file, with the standing caveat stated up front and repeated wherever it matters: every figure here is a complainant's claim typed into a federal form, not an adjudicated finding. The register records what people say happened. Some of what they say, as the first section shows, did not happen to them at all.

The audit, or the trailer with ninety-five deaths

The raw register reports 6,862 deaths, and the first thing this article did was refuse to print that number. Ten complaints report exactly 99 deaths each, which is not a toll, it is a form field pinned at two digits. Twenty-one more sit in a gray band between 11 and 98, and reading them one by one settles what they are. A teardrop camper trailer carries two complaints reporting 95 deaths each, filed under power train and engine, components a teardrop trailer does not have. One 2010 Civic reports 45 deaths three times across three different component families. A 1998 Grand Cherokee complaint reports 17 deaths, which is not an incident, it is a number from the news: complainants citing a defect's famous running total as if it happened in their driveway. So the ledger applies a cap, counting only complaints that report ten or fewer deaths, and the cap removes 1,670 reported deaths, 24 percent of the raw total. One reported death in four, dissolved by reading the rows. The filtered figure, 5,192, is the honest one, and the lesson is the article's first doctrine: a database column is testimony, and testimony gets audited before it gets quoted.

What hurts and what merely annoys

Now the centerpiece, the number nobody computes: reported deaths per thousand complaints, by component family, on the filtered ledger. The register's loudest families are its least lethal. Electrical systems, the biggest shelf in the modern file, run 1.4 deaths per thousand; power train 0.9; plain engine complaints 0.6. At the other end sit the families that exist only for the worst moment of a car's life: seat belts lead the entire register at 9.4 deaths per thousand complaints, tires follow at 9.0, air bags at 6.0, speed control at 5.1, seats at 3.7. The pattern is almost too clean. Crashworthiness components, belts, bags and seats, have no everyday failure mode worth a federal complaint, so when someone does file, it is disproportionately because the crash already happened and the part is accused of failing its one job; the injury column says the same thing louder, with air-bag complaints reporting injuries at 257 per thousand, one filing in four, and belts at 190. Tires are the exception that proves the register remembers: a tire complaint is routine, but the tire file carries the Firestone era inside it, and the deaths stayed in the column after the headlines moved on. Complaint volume measures annoyance. The harm rate measures the opposite thing, and the two rankings are close to inverted.

The crash board and the fire board

The two flags split the same way, on the unfiltered register since a flag cannot inflate. The crash flag belongs to the crashworthiness cluster: 25.9 percent of all air-bag complaints are crash-flagged, speed control runs 16.9, seat belts 13.8, the brake families around 10. The fire flag belongs to the energized parts: electrical systems lead at 7.5 percent, 18,675 fire-flagged complaints on that family alone, with engine-and-cooling at 5.7 and gasoline fuel systems at 4.5. The quiet surprise sits third on the fire board: seats, at 5.4 percent, which reads strange for furniture until you remember the modern seat is furniture wrapped around a heating element. The Parts Bin showed the car becoming an electrical appliance; the fire board shows the appliance's failure mode arriving in the upholstery.

The vehicles the ledger remembers

Rank vehicles by filtered reported deaths and the board reads like a shelf of case files, which is the best evidence the audit worked: strip the placeholder junk and what remains is the catastrophes the country actually had.

Three Explorers from the Firestone years carry 155 reported deaths among them. The 2008 Cobalt, the ignition-switch car, sits at 21. The one row the file flags rather than trusts is the top one: a 1987 Bronco carrying 120 deaths on only 61 complaints, two per filing, a density that reads as coordinated litigation-era submissions rather than owners reporting their own incidents, so the board prints it with its asterisk and moves on. Even the audit's survivors are testimony, not verdicts. The difference is that this testimony is at least arithmetically possible.

The register is getting less deadly

Cut the filtered ledger by failure decade and the death rate falls off a cliff: 5.6 reported deaths per thousand complaints for 1990s failures, 4.6 for the 2000s, 1.5 for the 2010s, 1.1 for the 2020s. Injuries trace the same slope, 97 per thousand down to 36, and the share of complaints carrying any harm flag at all has halved since the 1990s, 16.9 percent then, 7.0 now. The pre-1990 rows are too thin and back-filed to read, per the usual stratum rule. Two explanations fit, and the honest file keeps both. Cars really did get safer, the crashworthiness revolution showing up as fewer tragedies per grievance. And the register's denominator exploded, because filing got easy enough that the file now absorbs millions of low-stakes annoyances that would never have earned a stamp in 1995. The truth is some mix, and the ledger cannot apportion it. What it can say is that a complaint drawn at random from the 1990s file was five times more likely to carry a death than one drawn from the 2020s file, and that the register's whole shape has been drifting from safety log toward suggestion box for thirty years.

Reading the second file

The doctrine, in three parts. First, volume and harm are different instruments: the complaint count on a vehicle page measures how often owners were moved to write, and this ledger measures what they wrote about, and the loudest families are the least lethal, so never read a big number as a body count. Second, every harm figure is an allegation, and a quarter of the raw death column did not survive one afternoon of reading, so treat any uncured quote of these fields, including by this site, as a defect in its own right. Third, the audit is the article: the register is a public instrument maintained by the public, headlines and placeholder digits included, and a site built on it earns its keep exactly here, by reading the rows before repeating them. The second file is real. People crash, cars burn, and some of the saddest rows in the federal record sit under the component families built to prevent exactly that. The ledger's job is to count them without inflating them, and now, for the first time, it has.