RAFFIC fatalities in the United States fell steadily from 54,600 in 1972 to 34,900 in 1992. But then they started to rise again, and by 2002 there were 38,300 traffic deaths a year.

Our performance compared with other countries has also deteriorated. America's ranking has fallen from first to ninth over the last 30 years, with Australia, Britain and Canada all having better records.

A big part of the difference between the United States and other countries seems to be the prevalence of sport utility vehicles and pickups on American highways. Sales of light trucks - S.U.V.'s, pickups and minivans - were about a fifth of total automobile sales 30 years ago. Now they account for more than half.

But aren't large vehicles supposed to be safer than small cars? Yes, they are safer for their occupants in collisions, but their design makes them all the more dangerous for anyone they hit.

Michelle White, an economist at the University of California, San Diego, estimates that for each fatality that light-truck drivers avoid for themselves and their passengers, they cause four fatalities involving car occupants, pedestrians, bicyclists and motorcyclists. "Safety gains for those driving light trucks," Ms. White said, "come at an extremely high cost to others."

Being larger than ordinary vehicles, S.U.V.'s and light trucks cause more damage to upper bodies and heads in collisions. Furthermore, their bumpers do not always align with automobile bumpers, and their body structure is stiffer, transferring more force to other vehicles during impact.

A few weeks ago, the auto industry announced a voluntary plan to deal with some of these design problems. They intend to make side-impact airbags standard to help protect heads and upper bodies better in collisions, and they intend to standardize bumper heights.

These will no doubt be helpful improvements, but do they go far enough?

Recently Ms. White examined the econometrics of traffic accidents in an attempt to measure the benefits and costs of changing the number of light trucks on the road. Professor White's paper can be downloaded from http://econ.ucsd.edu/~miwhite/ suv-revision.pdf.

Ms. White notes that changing average vehicle size could, in principle, increase or decrease the cost of accidents.

Suppose the cost of a small vehicle-large vehicle collision is $50, the cost of a small vehicle/small vehicle collision is $45, and the cost of a large vehicle-large vehicle collision is $40.

If all vehicles are small, and there are 10 accidents a year, the total cost of the accidents is $450. But if 10 percent of the small vehicles are replaced by large ones, the average cost of collisions becomes $458.50, since more collisions will be between large and small vehicles. On the other hand, if 60 percent of the vehicles on the road are large, the average cost of a collision is only $456, since more collisions are between large vehicles.

Think about a safety-conscious soccer mom choosing a vehicle. If there are mostly small cars in her town, she can reduce the risk to her and her family in the event of a collision by buying an S.U.V.

The unfortunate side effect is that the large S.U.V. would cause significant damage to smaller cars if she was involved in an accident.

The laudable private incentive to choose a safe vehicle could, perversely, reduce overall safety.

In addition, Ms. White finds that people involved in single-vehicle crashes are more likely to be killed or seriously injured if they are in S.U.V.'s or light trucks rather than cars. This may be a result of the increased likelihood of rollovers.

On the other hand, suppose everybody in town drives an S.U.V. Then the soccer mom will definitely want to purchase one for herself, since it would both increase her family's safety and reduce the overall costs of collisions.

In this case, private incentives and social incentives are aligned.

The dynamics involved is the same as that of an arms race: if other families buy bigger vehicles, then you will want to as well, if only in self-defense.

To see where we are in this arms race, Ms. White examined crash data maintained by the National Highway Safety Administration at www.nass.nhtsa.dot.gov/nass.

Using this data, Ms. White was able to estimate how the probability of fatalities or serious injury varied with the types of vehicles involved in collisions.

For example, in a two-car accident, the probability of a fatality in the car is 38 percent less than in a car-light truck accident. However, in car-light truck accidents, the probability of fatalities in the light truck is 55 percent less than it would be in a truck-truck accident.

If a light truck hits a pedestrian or a cyclist, the probability of fatalities is about 82 percent greater than if a car is involved.

Ms. White then asked what the impact would be of replacing a million light trucks with cars. She considered two models for driver behavior. In the first, she assumed that the former drivers of light trucks would have the same number of accidents as they did when driving trucks. In the second, she assumed that the former drivers of light trucks would have the same accident probabilities as other car drivers.

Using conventional methods for value-of-life calculations, she finds that each light-truck owner who switches to a car saves about $447 in total expected costs of accidents.

Ms. White examines various policies that might persuade drivers to adopt such changes, including changes in liability rules, traffic rules and insurance rules.

Unfortunately, each of these policies has its problems, so there are no easy solutions.

One interesting way to reduce the arms race problem would be to link automobile liability insurance to gasoline taxes. This means drivers whose cars use more gasoline and those who drive a lot would pay more for their insurance - not unreasonable, since, on average, they impose more costs in accidents.

Aaron Edlin, a professor of economics and law at the University of California, Berkeley, has argued that such "pay at the pump'' insurance premiums would have many other benefits (www.bepress.com /aaronedlin/contribution5/). So this type of payment scheme is worth considering for a variety of reasons.

Hal R. Varian is a professor of business, economics and information management at the University of California at Berkeley.