Your Favorite Brand Doesn’t Care if You Live

I ran a variance decomposition on FARS fatality rates across every brand with five or more models in the dataset. 28 brands, 313 models, a decade of death data. Within-brand variance is 5.9 times higher than between-brand variance. The brand on your steering wheel explains roughly one-sixth of the safety variation that the specific model explains.

Take Chevrolet. Thirty-two models in the dataset. The Prizm (a rebadged Toyota Corolla, ironically) clocks a death rate of 0.02 per 100 million VMT. The Tracker? 7.83. That’s a 391-fold spread under the same bowtie. You could walk into a Chevy dealer in 2003 and drive home in the safest car in the FARS database or one of the deadliest—same salesperson, same financing desk, same “Like a Rock” jingle playing overhead.

Toyota does it too. The Matrix sits at 0.02 deaths per 100M VMT. The Land Cruiser? 6.27. A 313-fold difference. Honda’s spread is narrower but still absurd: the Passport at 0.10 versus the Accord at 3.07—a 31x gap between two vehicles that share a dealership lot and, occasionally, a platform.^[1]

The Math

Between-brand variance in death rate: 0.1280. This measures how much brands differ from each other on average. Average within-brand variance: 0.7517. This measures how much models within the same brand differ from their brand’s own average. The ratio—0.7517 ÷ 0.1280 = 5.87—means the model you pick within a brand swamps the brand itself as a predictor of whether you die in a crash.^[2]

Put differently: if you ran a one-way ANOVA with brand as the factor, the between-groups sum of squares would be dwarfed by the within-groups residuals. Brand loyalty, as a safety strategy, doesn’t survive the F-test.

Who’s Consistent? Who’s Roulette?

Volvo’s spread is 2.6x—the tightest in the dataset. XC70 at 0.17, S60 at 0.44. Their worst car is still safer than most brands’ average. Audi is nearly as tight at 2.9x (Q5 at 0.11, A6 at 0.32). These are the brands where “I trust the brand” actually holds up in the morgue data.

Hyundai? 142x spread. The Palisade at 0.06 deaths per 100M VMT is one of the safest vehicles in America. The Veloster at 8.54 is the single deadliest car in the entire FARS dataset, rate-wise. One brand. One dealership network. One set of Super Bowl ads. Two completely different odds of survival.

But Isn’t This Just Vehicle Class?

Partially. SUVs are heavier, sit higher, and protect occupants better in multi-vehicle crashes. Sports cars attract aggressive driving. Some of the within-brand spread is sedans vs. SUVs vs. coupes—physics, not branding.

But the class explanation collapses when you look within a class. Honda sells two mainstream sedans: the Accord (3.07) and the Civic (2.25). Same brand, same class, same general buyer profile, yet the Accord kills at a 36% higher rate per VMT.^[3] Chevrolet sells the Malibu (2.03) and the Impala (5.00)—both sedans, 146% difference. The Cobalt (5.10) versus the Cruze (0.63)—both compact Chevys, successor and predecessor, an 8.1× gap. Same bowtie, same slot in the lineup, completely different odds.

Vehicle class matters. But it doesn’t explain a 5.9× within-brand variance surplus on its own. Demographics, fleet age, design generation, ESC fitment rates, and structural engineering all contribute. The point stands: the badge predicts far less than the vehicle.

Limitations

FARS only records fatal crashes—roughly 40,000 per year out of ~6.7 million total crashes. Low-rate vehicles with small fleets have wide confidence intervals; the Chevy Prizm’s rate of 0.02 is based on just 2 deaths. Fleet and VMT estimates carry ±15% uncertainty for low-volume models. The dataset covers 2014–2023, meaning newer models like the Palisade have fewer exposure years, which can depress or inflate rates. The full FARS dataset has 337 models; the variance decomposition uses the 313 that fall within 28 brands carrying 5 or more models.^[1]

The variance decomposition is unweighted—each brand contributes equally regardless of fleet size or model count. A sales-weighted analysis might tighten the ratio. It wouldn’t eliminate it.

Strongest Counter

A sophisticated buyer doesn’t choose “a Toyota.” They choose a RAV4 or a Camry or a Tacoma, each of which has model-specific IIHS ratings and NHTSA star scores. Brand-level thinking, the argument goes, is a strawman—nobody actually buys that way.

Except they do. IIHS publishes model-specific ratings, but consumers still filter by brand first—walking into a Toyota dealership “because Toyota is safe” and then choosing from the lot. The IIHS itself demonstrates the gap: in any given year, the same brand can earn Top Safety Pick+ on one model and fail updated side-impact criteria on another.^[4] When someone says “I buy Toyota because they’re safe,” they’re applying a brand heuristic to a model-level problem. The FARS data quantifies what that costs: between-brand variance accounts for just 0.128 out of 0.880 total (between + within), or roughly 14.5% of the variation in fatality rate. The other 85.5% is within-brand—the specific model, not the badge.

Next time someone tells you they trust a brand, ask them which model.