What Percentage of Car Accidents Are Caused by Human Error?

Ninety-four percent. That is the number behind most claims that human error causes nearly all crashes, and it comes from the National Highway Traffic Safety Administration. In a national survey of crashes investigated at the scene from 2005 to 2007, NHTSA researchers assigned the “critical reason” for the crash to a driver in an estimated 94 percent of cases, with a 95 percent confidence margin of plus or minus 2.2 percentage points.¹

The number answers a narrower question than the one it is usually asked to answer. The “critical reason” is the immediate reason for the last event before impact, which NHTSA describes as often the last failure in the causal chain. NHTSA states that it “is not intended to be interpreted as the cause of the crash nor as the assignment of the fault to the driver, vehicle, or environment.”¹ So the accurate short answer is this: in about 94 percent of the crashes NHTSA sampled, researchers traced the final failure before impact to a driver. Whether human error caused those crashes, in the everyday sense of the word, is a question the survey was not built to settle.

Where the number comes from

Congress authorized the National Motor Vehicle Crash Causation Survey in the 2005 SAFETEA-LU highway law, and NHTSA’s National Center for Statistics and Analysis ran it from July 3, 2005, to December 31, 2007.¹ Unlike databases built from police reports after the fact, the survey sent researchers to crash scenes while police and vehicles were still there. They interviewed drivers, passengers and witnesses, assessed vehicles and roadways, and recorded at least 600 data elements per crash.²

Not every crash qualified. NHTSA’s report to Congress lists the rules:²

• The crash happened between 6 a.m. and midnight.
• Emergency medical services were dispatched to the scene.
• Police were at the scene when the researcher arrived, and a completed police report was available.
• At least one of the first three vehicles involved was a light passenger vehicle that was towed, or would be towed, because of damage.

Those rules shaped the sample. It leans toward crashes bad enough to bring an ambulance and a tow truck. It leaves out the hours between midnight and 6 a.m. And because it requires a light passenger vehicle, crashes involving only motorcycles or heavy trucks fall outside it.

What “critical reason” means

NHTSA coded each crash in layers.¹ First came the vehicle’s movement just before trouble began. Next came the decisive event before the crash, which NHTSA defines as the circumstance that led to the vehicle’s first impact: a car drifting from its lane, say, or turning across another car’s path. The “critical reason” is the immediate reason for that event. Last came associated factors: conditions present that were likely to add to the probability of a crash.

Researchers normally assigned one “critical reason” per crash, to a driver, a vehicle or the environment.¹ That design puts a single label on events that often have several contributors. Picture a driver who glances at a phone and drifts across the center line of a narrow, unlit two-lane road. The drift is the event. The glance is a recognition error, so the “critical reason” goes to the driver. The road’s width, the missing rumble strip and the posted speed may be noted elsewhere in the file. They do not change the label.

The driver errors, broken down

Among the estimated 2,046,000 crashes in which the “critical reason” was assigned to a driver, NHTSA sorted the errors into five groups.¹

Margins are 95 percent confidence limits in percentage points. Shares are of driver-assigned crashes, not of all crashes, and NHTSA computed them from unrounded estimates. Source: NHTSA, DOT HS 812 115, Table 2.

Two details trip up careful readers. The note’s summary paragraph gives the recognition-error margin as ±2.1, while Table 2 gives ±2.2. And NHTSA’s 2008 report to Congress, drawing on the same survey, put decision errors at about 34 percent and performance errors at 10 percent.² The 2015 note’s figures, 33 and 11 percent, are the ones in circulation today.

Recognition errors lead the list, and that is where distraction sits. A driver who looks toward oncoming traffic without registering a motorcycle, or who is reading a screen when the car ahead stops, belongs in this group. We look at what that means for riders in Distracted Driving and Motorcycle Crashes.

What the number does not mean

NHTSA’s own note is blunt. Of the crashes assigned to drivers, it says “in none of these cases was the assignment intended to blame the driver for causing the crash.”¹ Four limits follow from the survey’s design.

It is not a cause. The label marks the last link in a chain. Lane width, sight lines, speed limits and vehicle design sit earlier in that chain and are not what the 94 percent measures.

It likely undercounts vehicle problems. NMVCCS researchers did not perform detailed vehicle inspections. NHTSA says vehicle-related reasons were mainly inferred from external visual checks of parts such as tires and brakes, so the 2 percent vehicle share may miss problems that cannot be seen from outside.¹

It describes a sample, not all crashes. The survey covered tow-away crashes involving light passenger vehicles, with EMS dispatched, between 6 a.m. and midnight. It was not designed as a study of fatal crashes, and it does not define a category of “serious crashes.”²

It is old. The survey ended in December 2007, the year the first iPhone went on sale. Phones, vehicle technology and traffic deaths have all changed since.

The NTSB chair’s objection

In January 2022, with traffic deaths rising, National Transportation Safety Board Chair Jennifer Homendy told the Associated Press the statistic was misleading and that the Department of Transportation should stop using it.³ At the time, according to the AP, NHTSA’s website said “94% of serious crashes are due to human error” in a section about automated vehicles.

“That has to change,” Homendy said. “It’s dangerous.” She added: “You can’t simultaneously say we’re focused on a ‘safe system’ approach … and have a 94% number out there, which is not accurate.”³ NHTSA responded that it would update the wording on its website “to address that characterization of the data as well as provide additional information.”³

How truck crash research compares

The Federal Motor Carrier Safety Administration used the same “critical reason” concept in its Large Truck Crash Causation Study. That study examined 967 crashes from 2001 to 2003 at 24 sites in 17 states, each involving a large truck and at least one death or injury. Its 1,127 trucks were weighted to represent about 141,000 nationally.⁴

The results read differently from the passenger-vehicle survey:

• The truck was assigned the “critical reason” in 55 percent of the crashes studied.
• In two-vehicle crashes between a large truck and a passenger vehicle, the passenger vehicle was assigned the “critical reason” in 56 percent and the truck in 44 percent.
• When the truck was assigned the “critical reason,” it went to the truck driver in 87.2 percent of cases, the truck itself in 10.1 percent, the environment in 2.3 percent and unknown in 0.3 percent.⁴

Two cautions apply before setting these beside the passenger-vehicle survey. The 10.1 percent vehicle share is a share of truck-assigned crashes, while NMVCCS’s 2 percent is a share of all crashes in its sample. And the two studies drew different samples in different years with different methods. Neither gap is a measured difference between cars and trucks.

FMCSA was explicit about the label’s limits: “Identification of a single critical reason merely begins the process of explaining why a crash occurred.”⁴ The study also coded associated factors, such as brake problems and fatigue, as present or absent, without judging whether each one contributed. Our breakdown of the leading causes of large truck crashes walks through those factors and what they can and cannot show.

The systems view

In January 2022, USDOT published its National Roadway Safety Strategy and adopted the Safe System Approach as the frame for its road safety work.⁵ The approach starts from the same observation as the 94 percent figure, that people make mistakes, and draws a different conclusion from the one the number is often used to support. Instead of treating error as the thing to eliminate, it treats error as a given and asks how roads, vehicles and speeds can be arranged so that mistakes are survivable.

The strategy lists six principles. Two carry most of the weight here:

• Humans make mistakes. “People will inevitably make mistakes and decisions that can lead or contribute to crashes, but the transportation system can be designed and operated to accommodate certain types and levels of human mistakes, and avoid death and serious injuries when a crash occurs.”⁶
• Humans are vulnerable. “Human bodies have physical limits for tolerating crash forces before death or serious injury occurs.”⁶

The others hold that deaths and serious injuries are unacceptable, that responsibility is shared, that safety work should be proactive, and that the system needs redundancy so a single failure does not decide the outcome. The strategy organizes its work around five objectives: safer people, safer roads, safer vehicles, safer speeds and post-crash care.⁵

Read this way, the 94 percent figure describes a design problem more than a verdict on drivers. If a driver is the last link in nearly every crash chain, the earlier links (lane width, crossing distance, speed, vehicle structure, emergency response) are where a crash can still be made survivable. That logic runs through the countermeasures covered in Pedestrians, Cyclists and the Shape of the Road.

One note for readers following the links: the department’s NRSS web page now summarizes the strategy’s priorities as safer people, safer roads and safer vehicles.⁷ The January 2022 strategy document remains the source for the full Safe System framework.⁵

How to cite this figure accurately

For reporters, researchers and safety writers, a short checklist:

1. Name the source. NHTSA, Traffic Safety Facts Crash•Stats, DOT HS 812 115, February 2015, using National Motor Vehicle Crash Causation Survey data.
2. Use NHTSA’s term. Write “critical reason,” not “cause,” “fault” or “blame.”
3. Give the period and scope. Crashes from 2005 to 2007, investigated at the scene, involving a towed light passenger vehicle with EMS dispatched.
4. Include the margin when space allows: 94 percent, plus or minus 2.2 percentage points.
5. Avoid three common rewrites. “94% of crashes are caused by human error,” “94% of serious crashes are due to human error” and “drivers are at fault in 94% of crashes” all claim more than NHTSA measured.
6. Use current data for current questions. The survey describes crashes from nearly two decades ago. For today’s fatality counts and trends, start with FARS and the other federal data sets in our field guide to U.S. crash data.