Some ideas fail when tested on real people, who sometimes react unpredictably and aren't always the rational maximizers of utility assumed by classical economics. (For a demonstration of this, see our interactive quiz.)

So when testing policies in the lab, Chen and his fellow researchers look for unintended consequences, which alert them to loopholes that need to be closed or ideas that need to be scrapped altogether.

"The experiments are a cheap way of removing your mistakes," says CalTech economist Charlie Plott, who helped develop the field and was a teacher of Chen's. "All the big expensive mistakes are confined to the laboratory."

Plott recently used his expertise in auctions and markets to consult for Ford Motor, finding a clever way for Ford to profitably comply with federal fleetwide gas-mileage regulations despite the automaker's decentralized structure. To address the problem, he designed and tested a market for fuel-efficiency credits that would be used just within the company.

Governments have used the experimental approach for a long time to guide the development of regulatory policies and allocate resources. But few corporations have historically utilized it because of the cost and uncertain payoff. Hiring the staff and developing the software to run an in-house lab can easily cost hundreds of thousands of dollars or more a year, according to Sam Dinkin, a former I.B.M. research economist who now works for Power Auctions.

But the cost-saving work of researchers like Chen has caused more companies to take notice.

One of Chen's recent projects involved finding a way for H.P. to more accurately predict demand from its nine distributors, who collectively sell as much as $3 billion worth of H.P.'s products. The problem? Its distributors' forecasts for demand were frequently off by as much as 100 percent, wreaking havoc on H.P.'s production planning.

Chen's solution to the planning problem, which H.P. intends to test soon with one distributor, was to develop an incentive system that rewarded distributors for sticking to their forecasts by turning those forecasts into purchase commitments. In the lab, the overlap between distributors' forecasts and their actual orders using this system increased to as high as 80 percent. "That's pretty astonishing given that the underlying demand is completely random," Chen says.

"You model the environment and try different strategies, and the experiments tell you which way to go," says Chen of his basic approach. "If you can understand how people make decisions, you'll be able to zero in on useful ideas a lot faster."