Drivers cause or contribute to more than 90 percent of collisions, according to transportation safety experts. But what we don't know, exactly, is what happens in the moments leading up to a crash. Questions like what drivers do just before a collision, and how they react to cues from their vehicle and the surrounding environment, are largely a mystery.

While accident investigators might look at things like skid marks and debris patterns to figure out how cars collided, the problem, says traffic safety expert James Hedlund, is "that's all after the fact. You don't really know what that person was doing. Was he picking his nose? Was he talking on his cellphone?"

A new study could help answer questions like that, revolutionizing our understanding of how motorists interact with their vehicles and the road. It could make driving vastly safer. And it couldn't come at a better time: Last year, the number of auto fatalities crept back up again -- by more than 5 percent -- after having declined for seven straight years by more than 25 percent to 32,367.

This appears in our free, monthly Infrastructure e-newsletter. Not already a subscriber? Click here.

Researchers have equipped more than 3,100 cars nationwide with a slew of sensors and tools designed to capture a bounty of data about driving and behavior. In exchange for $500, volunteers' cars were outfitted with accelerometers, GPS, radar and other devices that track drivers' use of brakes, horns, turn signals and lights. Alcohol sensors determine whether there's booze in the vehicle. Cameras capture volunteers' faces as they drive, as well as their front and rear views and their interactions with the dashboard. The data will capture everything happening inside a car prior to a collision and help explain how and why people get into accidents -- and just as important, how they can be avoided.

Researchers worked to recruit a mix of volunteers whose genders and ages proportionately reflected the population of U.S. drivers. Though drivers are aware they're being studied, they only briefly become more conservative behind the wheel and return to old habits within days. "People have completely forgotten this thing's in their car," says Hedlund, a former deputy administrator of the National Highway Traffic Safety Administration and a consultant on the study. "They do the craziest things."

The more serious issue, which is harder for the researchers to deal with, is the self-selection bias among participants. Someone who enjoys reckless behavior like drinking and driving probably won't volunteer for the study. "We don't get the real crazies, but that's not altogether bad," Hedlund says. "We're trying to do stuff for normal people."

Researchers are also collecting detailed roadway data in the six communities where driver data is being gathered: Bloomington, Ind.; Buffalo, N.Y.; Raleigh-Durham, N.C.; Tampa, Fla.; Central Pennsylvania; and Seattle. The idea is to marry the two sets of data to get information not just on how humans interact with vehicles, but on how the condition and shape of roadways influence their behavior.

This type of research isn't unprecedented, but its scope is. The only other study of its kind was conducted about a decade ago by Virginia Tech University. Data was collected on only about 100 cars. The latest study -- also being led largely by Virginia Tech -- dwarfs that one. As of August, researchers have collected data on 34 million miles of driving, and they expect to generate 4 petabytes (that's 4,000 gigabytes) of data. Researchers have also already collected data on about 500 crashes of varying severity. They expect to have data collected on a total of 700 crashes -- and nearly 7,000 close calls -- by the time data collection concludes in November.

The Naturalistic Driving Study is being coordinated by the Strategic Highway Research Program, known as SHRP 2, which was authorized by Congress to conduct the work, and administered by the Transportation Research Board. The total SHRP 2 budget is around $70.4 million.

Hedlund says it's not designed to answer a single, specific question. Rather, it will collect data that researchers will be able to use to answer their own questions, like how do people really use cellphones in the car, or how do drivers respond to speed limit signs. In fact, three organizations are already under contract to analyze part of the data. Iowa State University is looking at how driver behavior and roadway characteristics influence traffic accidents on two-lane, rural roads.

Not-for-profit researcher MRIGlobal is looking at whether offset left-turn lanes -- a style of turn lanes designed to be safer but that takes up more space -- are really effective. Chalmers University in Sweden is trying to figure out how a driver's glance away from the road impacts the likelihood of a rear-end collision. Hedlund says that study could determine best practices for in-vehicle technology by answering questions like what is safer, one long-glance at an instrument, or several shorter ones. The research could lead to new design standards for dashboard instrument panels, car radios or GPS devices.

Meanwhile, the American Association of State Highway and Transportation Officials is putting together a task force to recommend what sort of initial analysis will be most useful for state DOTs.

By December 2014, the team will have compiled the data into a format that will be easy for researchers to look at and request specific data sets based on narrow topics they're studying. For example, someone could request every record of every rear-end collision in which a driver failed to use a horn and had adjusted the radio in the moments before the crash. A record, which researchers call an "epoch," will contain all the relevant data about what's happening in and around the car for 20 seconds before and 10 seconds after the incident.

Hedlund says he's personally interested in seeing the role that cellphone use plays in crashes. Previous studied have been ineffective, he says, because they've either relied on observations -- researchers standing on corners counting how many drivers are using cellphones while driving -- or questions asked by crash investigators after the fact. Cellphone records of people in crashes aren't very helpful because they're not accurate to the second, Hedlund says.

The research could reveal some uncomfortable truths. For example, maybe there's a safe way to use a cellphone in a car, despite conventional knowledge. "I've read a whole mess of cellphone studies, some of which are absolutely certain cellpones are the worst thing [ever]," says Hedlund. "Others say no, they don't make a darned bit of difference."

Hedlund says the information gleamed from the study could help civil engineers design roadways to promote safer conditions, lead to better types of training for drivers and help lawmakers create traffic safety laws that are more likely to save lives.

Shauna Hallmark of Iowa State University says her team hopes to examine about 1,000 trips. The work will be based on a regression analysis. Researchers will find the crashes they're interested in and then figure out the likelihood that certain conditions preceded those crashes. "When a police officer gets to a crash, they infer a lot of stuff," Hallmark says. "You'll hear officers say, 'They were traveling too fast for the conditions.' Well, we don't know how fast they were traveling. Maybe they figured they were on the cellphone, but you don't really know. It's giving us insight into exactly what's going on with the driver."

Must Reads

Links to planning and infrastructure stories you might have missed

South Florida officials plan to use an optical illusion to trick motorists into driving more slowly, the Sun-Sentinel reports.

Find out why the states are lining up to oppose the American Airlines-US Airways merger.

Amazingly, it took the Washington State Department of Transportation just 66 days to erect a new bridge after the infamous collapse earlier this year.

The Southeastern Pennsylvania Transportation Authority (SEPTA) could close 9 of its 13 regional rail lines because it's running out of money. That's caused a lot of panic in Pennsylvania.

Houston's KUHF reports on a new Texas Department of Transportation publicity campaign that targets young Hispanic men, who are involved in half the state's drunk driving crashes.

A proposed cross-border bridge connecting Tijuana's airport directly to the U.S. could open next year.

St. Louis becomes the latest city to take a look at bus rapid transit, the St. Louis Post-Dispatch reports.

Robert Bruegmann -- a professor of art history and urban planning -- explores the aesthetic beauty of highways.

Amtrak could soon do away with several of its shorter routes if states don't reach cost-sharing agreements with the rail provider.