Humans, not robots
One reason for the gap is that humans, not robots, drive the test cycles that measure fuel economy in city and highway driving. Skilled drivers can smooth out the tests, accelerating and braking gingerly in order to save fuel.
"The industry has some very good drivers, and we've noticed," Grundler says.
It is not cheating, Grundler says -- it is just taking full advantage of the rules.
Indeed, says John German, a fuel economy expert who worked as an engineer at Chrysler, Honda and the EPA, this is a natural consequence of car buyers' and regulators' increasing demand for better fuel economy.
"Any time you have stringent standards that require expensive changes to automobiles, the automakers are going to look for the cheapest way to comply," he said. "And if that means taking advantage of the rules, they're going to do it."
Now the EPA is working with SAE International to come up with new testing standards to narrow the gap. The agency also has started demanding that automakers submit their "drive trace," an electronic log of their driving patterns, so the EPA can see if drivers are leveling out the hills and valleys in test cycles.
There's a risk for automakers in trying to do too well on the test. Under EPA regulations, an automaker can use its own numbers on a window sticker as long as its estimate is within 3 percent of the EPA's readings. If the EPA catches an automaker with a bigger discrepancy, which now happens routinely, the automaker can demand a retest.
"When they demand a retest, we confirm the original number that we got nearly every time, and that's what goes on the label," Grundler says.
Even if the agency is vindicated, says Grundler, "that's a problem for me, because it doubles our workload."
One way to bring more consistency to the tests would be to program robots to operate the accelerator and brake, precisely following the EPA test cycle. But that could cause new problems, experts say. Without the wild card of a flawed human driver, car companies would have an even easier time tailoring their cars to the test, and the results would be even more removed from real-world conditions.
Says Grundler: "People aren't robots."