Compared to the highway, the sky is supposed to be a friendly place.

The risk of being killed on the average plane trip, statistics traditionally have told us, is about one-twentieth that of being killed on the highway. As flight attendants are fond of telling passengers, the most dangerous part of their trip is the drive to the airport.

But according to a new study in the journal Risk Analysis, those statistics tell only part of the story. Under certain conditions, the study asserts, many Americans are not more likely to survive in the air at all. In fact, for trips up to 600 miles long, they may be better off in a car.

The findings carry a daunting list of caveats and assumptions. For one thing, the figures apply only to dying, and do not count the risk of nonfatal injury, which all concede is much higher in a car. Moreover, the results pertain only to certain types of drivers traveling in certain types of cars on only the safest kind of highways.

But the study -- conducted by three statisticians at General Motors led by Leonard Evans -- flies so resolutely in the face of conventional wisdom that it has generated a wave of controversy among risk assessment experts. It also serves as a compelling illustration of why the statistics that shape human behavior can sometimes be highly misleading.

Evans's analysis is based on a simple observation about the differences between the risks of airline travel and highway travel.

The risk of being killed in a plane crash -- 0.55 fatalities per billion passenger miles flown -- is borne almost equally by everyone on the aircraft. There are few ways for a passenger to improve the chance of living.

On the other hand, the comparable fatality rate for drivers -- 12.56, nearly 23 times as high -- is typically calculated as an average, lumping high-risk travelers -- teenagers, madcap lane-changers and drunks -- with middle-aged accountants with four children in a large station wagon who drive so carefully they never so much as nick a fender.

The drunk's risk of dying is much higher than the average and the accountant's is much less.

When Cars May Be Safer

What Evans argues is that for safe drivers the chances of being killed in a car accident are less than the chance of dying in a plane crash -- even if the assessments of average risk show the opposite.

Consider the effect various factors have on the risks of driving:

Age: Eighteen-year-olds are about five times as likely to die in a car accident as is the average driver. A 40-year-old, by contrast, has a death rate about half the average.

Alcohol: People legally drunk behind the wheel have a fatality risk 7.66 times the norm. Drivers who don't drink have a risk only about 60 percent of the norm.

Seat belts: Wearing a seat belt cuts a driver's risk of dying on the road in half.

Car size: Generally speaking, the bigger the car, the safer its occupants. Drive a large, heavy car -- a Chevrolet Caprice, for example, or a Ford LTD -- and your chances of dying in an accident are 26 percent less than average. Drive a compact and your death rate is about 35 percent higher than average.

Road type: Traveling on a rural interstate highway is about half as dangerous as driving on smaller roads and highways.

According to Evans, people who violate every one of those safety factors -- young, drunk, un-seat-belted drivers in small cars on a winding country road -- risk a fatality rate of 930 deaths per billion miles driven. But someone at the other end of the scale -- experienced, sober, seat-belted, driving a big car on interstate routes -- would risk 0.804 fatalities per billion miles -- a rate 1/1,157th as great.

How do the safer group's odds compare with the chances of dying in a plane crash? Very favorably, Evans says.

The risk-per-passenger-mile, the usual way airline safety is compared with that of car travel, actually decreases with the length of trip. This is because almost all plane crashes occur on takeoff or landing. Short flights and long flights, in other words, have the same chance of crashing. But longer flights spread those risks over many more miles, lowering the overall fatality rate for the passengers.

Cars, the study concludes, can be safer than planes taking the statistically riskier shorter trips. For a 200-mile trip, the ideal driver is four times as safe in his or her car as on a plane. For a 300-mile trip, the car is twice as safe. Only after 600 miles -- when airplane fatality rates fall below 0.8 deaths per billion miles -- does flying become a better alternative.

That's not all. Evans's analysis includes only the driver of a car -- the one at greatest risk. But passengers -- particularly those sitting in the relative safety of the back seat -- have much lower fatality rates than do drivers. If car occupants are included, the best-case highway fatality risk sinks to 0.59 per billion miles driven, making sitting in the back seat preferable to flying for trips up to 800 miles.

Numerous Caveats Apply

The assumptions do, however, come with a number of caveats. Not everyone choosing between a plane and a car can meet all the ideal driving characteristics. Not everyone, for example, wants to stay on interstates or has a big car available.

Driving an average-size car -- say a Ford Taurus -- under otherwise ideal circumstances increases driving risk to 1.09 per billion miles, which makes driving preferable only for trips under 500 miles. Traveling in a compact -- which many Americans do -- pushes the risk to 1.48 per billion miles, wiping out much of the driving advantage.

The analysis is also very selective because it compares only fatalities. But on the highway, injuries are 70 times as likely as are fatalities. If a traveler is unwilling to accept a greatly increased chance of serious injury -- even if driving brings a lower fatality rate -- then flying still makes sense.

Then there's the question of balance.

"It sounds as if the article is much more interested in decomposing the average auto trip than the average airplane trip," said Arnold Barnett, an operations research specialist at the Massachusetts Institute of Technology.

Barnett said that if a similar effort were made to separate out all the factors that affect airline safety, the picture would change. The risks of flying aren't nearly as variable as those of driving, but there are some differences. Some airports -- like Atlanta and Dulles -- are safer than others such as Los Angeles and San Diego. Newer, bigger planes are less likely to crash than are smaller, older planes. An airline passenger sitting by an emergency exit may have a better chance of survival than someone trapped in the middle of a row.

On a trip from Washington to Atlanta, is a sober 40-year-old sitting by the exit in a brand new jumbo jet still better off than behind the wheel of a Mercedes? Perhaps the airline industry will put forth an analysis to compare with GM's.