Drunk or Sober, the Car Doesn't Care

Every safety campaign in America sells the same story: stay sober, stay alive. And they're right about the first part. Impaired drivers are roughly four times more likely to end up in a fatal crash in the first place.^[3] No argument there.

But what happens after metal meets metal? We cross-tabulated FARS impairment rates with crash lethality ratios across 189 vehicle models. The overall correlation looked real: r = 0.23. Then we controlled for vehicle class. Classic Simpson's Paradox. The number evaporated.

Sedans: r = 0.07. Pickups: r = 0.009. SUVs: r = 0.14.^[1] Within any vehicle class, the percentage of impaired drivers behind the wheel of a given model tells you nothing about whether its occupants survive a fatal crash. Nothing. Steel and airbags determine who walks away. Blood chemistry does not.

The Toyota Solara posts 4.1% impairment and 0.685 lethality. Overwhelmingly sober drivers. The Infiniti G37 runs 25% impaired and posts 0.681. Six times more intoxicated drivers behind the wheel; functionally identical survivability. Cadillac CTS at 25.9% impaired? Lethality of 0.688. Sober Solara owners die at the same rate as the Infiniti's impaired ones.^[1]

That weak overall signal (r = 0.23) is a textbook confound. Both impairment and lethality are higher in sedans than in trucks. Sedans average 20.4% impaired and 0.644 lethality. Pickups: 20.1% and 0.489.^[2] Lump them together and you get a phantom correlation. Separate them and it vanishes. The real variable was vehicle mass all along.

Methodology: we divided own-vehicle deaths by total fatal crashes for each model (the lethality ratio), then correlated that against FARS toxicology impairment percentages. Filtered to 200+ crashes and 50+ toxicology-tested drivers. Pearson r computed overall and within each of five vehicle classes.^[1]

What this does not prove: that sobriety is irrelevant to crash survival. FARS only captures fatal crashes. Impaired drivers crash at higher speeds and in different configurations (single-vehicle rollovers vs. intersection T-bones), so crash kinematics differ.^[4] We are measuring survivability conditional on already being in a fatal-severity crash, not overall risk. FARS toxicology testing is also incomplete, meaning some impairment goes unrecorded.

The strongest case against our finding: drunk drivers may crash hard enough to overwhelm any structural advantage. A 70 mph tree strike and a 35 mph broadside produce wildly different forces. If impaired crashes cluster at higher speeds, vehicle engineering gains cancel against the kinetic energy penalty. The within-class zero correlation might reflect those cancellations, not true independence between sobriety and survivability.

Possibly. But IIHS research has consistently identified vehicle size and structural design as the dominant predictors of real-world occupant survival.^[2] Our data puts a number on what their work implies. Once the crash starts, physics selects for crumple zones. Not for what you drank.