Few people missed the headlines when the latest employment figures were unveiled earlier this month by the U.S. Bureau of Labor Statistics. The addition of only 32,000 new jobs, or 200,000 fewer than expected, alarmed everyone. Stocks swooned, reporters wrote economic obituaries and President Bush's political opponents crowed.

But hardly anyone noticed a new study, published the same day by the same federal agency, that shows payroll job losses have been consistently overestimated since the 2001 recession. How can this be? The problem lies in the way we measure jobs.

The BLS has two ways to do that. One is the payroll survey, which charts the number of jobs that employers report to the government. It shows we've lost 1.24 million jobs since March 2001 (including the 32,000 in July). The second is the household survey, in which the Census Bureau queries Americans directly about their job status. It shows we've added 1.81 million jobs since March 2001, including 629,000 last month. Why the discrepancy?

Part of the reason is that the payroll survey overlooks some segments of workers that have been growing in today's economy - such as the self-employed and consultants. None of these show up on payrolls. The economy is structurally different than it was five or 10 years ago, and it may be that the payroll-jobs numbers are weak simply because the economy now relies less on payrolls for engaging the labor force.

But the new BLS study now acknowledges another reason: job-changing.

When workers change jobs, they wind up being counted twice in the payroll survey - as employees for their former and new employers. When turnover rates are high, as they were in the late 1990s, it looks as if there are many more jobs than there actually are. But when turnover declines, as it has since 9/11, the jobs figure is closer to reality. Thus, some jobs are cited as "lost" when, in fact, employees have merely stayed at the same jobs.

And there's a third reason to suspect that the payroll survey isn't serving us well: Other economic indicators point to recovery. Real wages are rising. Unemployment is low and declining. (Indeed, the same day the 32,000 figure came out, the unemployment rate was reported to have dropped from 5.6% to 5.5%.) Jobless claims have been declining for the last year and are holding at about 340,000 a week. Levels below 400,000 are widely perceived to reflect a healthy, growing workforce.

There's more. The last few months also have brought much-needed relief to long-term unemployed workers. The number in this group declined by more than 300,000 workers since March. And the length of unemployment dropped sharply in July to 8.9 weeks from 10.8 weeks in June.

Critics note that the household survey's employment level varies widely from one month to the next. However, variability isn't caused, as is so often charged, by the small sample size. It's caused by the fresh faces in the sample - 15,000 new respondents per month. Over the long term, the instability disappears, and trend lines are clear. Not only is employment growing by millions, but the labor force is too.

Now for the question of scale: How much does job-changing distort payrolls? The BLS says it overstated payroll job losses by a quarter of a million, but it uses cautious assumptions. And let's not forget the fact that payrolls overlook many workers, as mentioned above. As a result, the study from BLS offers the lowest possible estimate of the overstated job losses. Its acknowledgment is just the tip of the iceberg.

But does it really matter how big the hole is in the side of the Titanic? The point is, the payroll survey is now officially suspect. At the very least, it shouldn't be viewed as superior to the other sources of economic data. Analysts have little choice but to reevaluate all their economic assumptions.

For now, though, the household survey should be seen as the standard for long-term analysis, and payrolls should be kept in perspective. If the economy is going to take center stage in the political debate, we need to ensure we're arguing from accurate figures. Thanks to the Bureau of Labor Statistics, we're getting closer to doing just that.

Kane is a research fellow in the Center for Data Analysis at The Heritage Foundation. He is a veteran Air Force intelligence officer, and former San Diego software entrepreneur.