The government has overstated by thousands the number of jobs it has created or saved with federal contracts under the president's $787 billion recovery program, according to an Associated Press review of data released in the program's first progress report.

The discrepancy raises questions about the reliability of a key benchmark the administration uses to gauge the success of the stimulus. The errors could be magnified Friday when a much larger round of reports is released. It is expected to show hundreds of thousands of jobs repairing public housing, building schools, repaving highways and keeping teachers on local payrolls.

The White House seized on an initial report from a government oversight board weeks ago that claimed federal contracts awarded to businesses under the recovery plan already had helped pay for more than 30,000 jobs. The administration said the number was evidence that the stimulus program had exceeded early expectations toward reaching the president's promise of creating or saving 3.5 million jobs by the end of next year.

But the 30,000 figure is overstated by thousands - at the very least by nearly 5,000, or one in six, based on AP's limited review of some of the contracts - because some federal agencies and recipients of the money provided incorrect job counts. The review found some counts were more than 10 times as high as the actual number of jobs; some jobs were credited to stimulus spending when, in fact, none were produced.

The White House says it is aware there are problems. In an interview, Ed DeSeve, an Obama adviser helping to oversee the stimulus program, said agencies have been working with businesses that received the money to correct mistakes. Other errors discovered by the public also will be corrected, he said.

"If there's an error that was made, let's get it fixed," DeSeve said.

Within minutes of the publication of AP's story, the White House released a statement that it said was the "real facts" about how jobs were counted in the stimulus data distributed two weeks ago. It said that had been a test run of a small subset of data that had been subjected only to three days of reviews, that it had already corrected "virtually all" the mistakes identified by the AP and that the discovery of mistakes "does not provide a statistically significant indication of the quality of the full reporting that will come on Friday."

The data partially reviewed by the AP for errors included all the data presently available, representing all known federal contracts awarded to businesses under the stimulus program. The figures being released Friday include different categories of stimulus spending by state governments, housing authorities, nonprofit groups and other organizations.

As of early Thursday, on its recovery.org Web site, the government was still citing 30,383 as the actual number of jobs linked so far to stimulus spending, despite the mistakes the White House has now acknowledged and said were being corrected.

There's no evidence the White House sought to inflate job numbers in the report, but the administration embraced the flawed figures the moment they were released.

The figures released earlier this month claimed jobs linked to roughly $16 billion in federal contracts, an initial report on a small fraction of the total stimulus program. DeSeve said federal officials had only a few days to go through the data for errors before they were made public.

It's not clear just how far off the 30,000 claim was. The AP's review, which was not an exhaustive accounting of all 9,000 contracts reported by the government so far, homed in on the most obvious cases of jobs wrongly tied to the stimulus because of duplications or misinterpretations of how the jobs should be counted.

While the thousands of overstated jobs represent a tiny sliver of the overall economy, they represent a significant percentage of the initial employment count credited to the stimulus program.

Administration officials say they are trying to head off such problems before the new figures are released Friday.

"Part of this is, it's an unprecedented effort," said Tom Gavin, a spokesman for the White House budget office. "It's as new to recipients who have to do it as it is to the American people who are able to view this data for the first time."

Some businesses actually undercounted jobs funded with stimulus money, the AP's review shows, because they reported only new jobs created, not existing jobs saved. But by far the most reporting errors were found in the number of jobs credited to the stimulus.

"I'm not trying to say one balances out the other," DeSeve said. "We don't like either of them."

In one major miscount found by the AP review, Colorado-based Teletech Government Solutions had worked with the Federal Communications Commission to come up with a job count for its $28.3 million contract for call centers fielding consumer questions about conversion of televisions to receive digital signals. The company reported creating 4,231 jobs - the highest number listed in the first stimulus accounting - even though 3,000 of those workers received a paycheck for five weeks or less.

"We all felt it was an appropriate way to represent the data at the time," company president Mariano Tan said.

Now the job count is being adjusted to less than 1,000, Tan said, to meet the requirement that a job reported is equal to a full-time, 40-hour-a-week position held for one year.

The Toledo, Ohio-based Koring Group also received two FCC contracts to help people make the switch to digital television. The company reported hiring 26 people for each of the two contracts, bringing its total jobs to 54 on the government's official count.

But the company cited the same 26 workers for both contracts, meaning the same jobs were counted twice. The job count was further inflated because each job lasted only about two months, so each worker should have counted as one-sixth of a full-time job.

The FCC spotted the problem and called company owner Steve Holland, who now says the actual job count is closer to five, not 54.

"We're just trying to be accurate. All of this has happened so fast," Holland said. "It is a little confusing. We're new to government contracting."

The AP's review identified nearly 600 contracts claiming stimulus money for more than 2,700 jobs that appear to have similar duplicated counts.

DeSeve said he's pleased that the FCC and other agencies are working with businesses to fix the errors.

Barbara Moore, executive director of the Child Care Association of Brevard County in Cocoa, Fla., reported that the $98,669 she received in stimulus money saved 129 jobs at her center, though the cash was used to give her 129 employees a 3.9 percent cost-of-living raise. She said she needed to boost their salaries because some workers had left for better paying jobs.

"They were leaving because we had not been able to give them a raise in four years," Moore said.

Officials at East Central Technical College in Douglas, Ga., said they now know they shouldn't have claimed 280 stimulus jobs linked to more than $200,000 to buy three semi-trucks and trailers for commercial driving instruction, and a modular classroom and bathroom for a health education program.

"It was an error on someone's part," said Mike Light, spokesman for the Technical College System of Georgia.

The number of jobs should be zero, Light said. The 280 count represents the number of students who would benefit, he said.

The San Joaquin, Calif., Regional Rail Commission reported creating or saving 125 jobs as part of a stimulus project to lay railroad track. Because the project drew from two pools of money, the commission reported that figure twice, bringing the total to 250.

Spokesman Thomas Reeves said the commission corrected the data Tuesday and changed the total to 73, although the count is not corrected in the government's official job tally. He said officials incorrectly added some indirect job creation to reach its initial 125 total. He said the number should not have been doubled.

DeSeve said he's confident the job counts in the first report will be corrected and future reports will have fewer errors.

"What we want is the most accurate total available," he said.