Having declared last June that the recent recession had eliminated many more jobs than anyone had realized, the Labor Department is now canceling that view.
The department says it overstated by 540,000 the number of jobs that were created in the late 1980s. And then it overstated how many jobs had disappeared in the recession.
Not until early this year did the Labor Department realize that any of these figures were incorrect, and that data processing companies that deliver payroll information to the government had miscounted.
Among other things, they were counting paychecks rather than the people getting the checks -- so that a person getting a paycheck and then an overtime check came out in the data as two workers.
The Labor Department's Bureau of Labor Statistics will formally correct the job numbers in June, the month it sets aside for correcting mistakes.
"We blamed the recession for eliminating jobs that had never existed in the first place," said Thomas Plewes, an associate commissioner of the bureau. "It just dawned on us recently, after we had gone through millions of records, what really happened."
Beyond jobs -- and fresh doubts about the reliability of government economic data -- the corrections appear to shed little new light on the economy.
The unemployment rate, 7 percent through March, is derived from a different source: a survey of households. Economic growth, or gross domestic product, is calculated from revenue and income data, not job holders. Only productivity, which measures worker output, might be revised in June.
"We now find that productivity fell trivially during the recession," said Edwin Dean, the bureau's productivity expert. "We might change this to a slight drop."
But the error might have given President Clinton some help in winning the election. The "accounting mistake," as the bureau now labels it, came at a bad moment for George Bush. Until last June, the bureau had calculated the job loss during the $l 1990-1991 recession at 1.2 million people.
When last June rolled around -- the annual moment for officially correcting mistakes -- the bureau ruled that the recession had cost more jobs than earlier data had suggested, and that the real job loss during the recession, which officially lasted nine months (from June 1990 to April 1991), had been 1.7 million, not 1.2 million. The bureau had issued what it now says was an incorrect correction.
Seizing the new 1.7 million figure, Mr. Clinton charged that Mr. Bush had callously minimized the recession's impact -- a charge that no longer stands up so well.
But Mr. Bush will not emerge exonerated when the bureau issues its new corrections next month.
He had been credited with some of the 540,000 extra jobs counted earlier in his term. Now his record for job creation will go from a meager 1.33 million new jobs during his four-year term down to 1 million or so. The bureau is still working out the exact number.
Counting American jobs took on a new complexity in the 1980s. Six million American companies, almost every employer, are required to deliver payroll and tax data to the Internal Revenue Service and state tax agencies, and some of this information is also used by the Bureau of Labor Statistics to count job holders.
Until the 1980s, the companies handled most of this work themselves. Then the data processing companies got into the business, signing contracts with roughly a million companies employing 30 million people, Mr. Plewes said.
Somehow, a number of the data processors made similar mistakes: Their software counted payrolls instead of people, or counted data for an entire month rather than the single week in mid-month that the bureau specifies.
This meant that a regular paycheck, for example, counted as one job holder, Mr. Plewes said. An overtime check to the same person came out in the data as a second job holder. And a bonus check, to the same employee, became a third job holder. In addition, counting paychecks for an entire month, rather than a single week, meant that people who received two or more paychecks were counted as two or more job holders.
With more than 100 million Americans holding jobs, the 540,000 represented a divergence of half a percent -- enough to notice when corrected all at once, but hard to spot over the years with only a few thousand mistakes made each month.