The partisan duel over the latest Bush trick to embellish economic data.

The Bureau of Labor Statistics’ monthly payroll and unemployment numbers have become the hottest economic data in the presidential campaign. Republicans cite any uptick as a sign of the incipient Bush boom, fueled by the brilliantly calibrated tax cuts, and Democrats tout any payroll decline as further evidence of miserable failure. (In September, for the first time since January, the economy added payroll jobs—some 57,000. Temporary jobs, a potential harbinger of growth increased by more than 30,000.)

But in recent weeks, there’s been an emerging campaign, mostly by Bush partisans, to discredit the BLS payroll numbers. It is both a serious economic inquiry and a political effort to distract attention from the ugly job figures.

Last Friday, on the Wall Street Journal editorial page, Allan Meltzer, an economist at Carnegie Mellon and visiting scholar at the American Enterprise Institute pleaded: “Don’t believe the widely reported loss of millions of manufacturing jobs since the Bush administration took office.” Why? “All these alleged facts are either wrong or greatly exaggerated, based on the same faulty source.”

That “faulty source” is the BLS Establishment Survey, which provides the payroll jobs number. Instead of relying on the Establishment Survey, Meltzer says, we should pay attention to the BLS Household Survey, on which the unemployment rate is based. “For the year ending in August, the Establishment Survey shows a loss of 463,000 jobs. The Household Survey shows that the economy added 313,000 new jobs in the same period.”

Writing on the same page earlier in September, Brian Wesbury, chief economist at Griffin, Kubik, Stephens & Thompson, noted that “the Household Survey shows that 1.186 million new jobs have been created this year,” while the payroll data find that several hundred thousand jobs have been lost. The difference, Wesbury says, “can be explained by the fact that sole proprietorships and other small companies are starting up at a faster rate today than they did even in the go-go ‘90s. They don’t show up on the Establishment Survey’s radar because they’re too small and too new.”

And a non-polemical paper issued in July by John Kitchen, an economist with the House Budget Committee, found that, historically speaking, when the economy is coming out of a down cycle, new jobs generally appear first in the Household Survey and only later in the payroll data. In the past, the government has been forced to revise its payroll numbers sharply upward. “The analysis in this paper suggests that nonfarm payroll estimates could be revised up by roughly 500,000 to 700,000,” Kitchen concluded.

What gives? It turns out the Bureau of Labor Statistics each month takes snapshots of the labor market using two different cameras. Each ends up photographing slightly different phenomena, and each has its own flaws and quirks.

To compile the Establishment Survey, BLS gathers data from some 400,000 businesses around the country and then estimates the total number of jobs. This survey explicitly does not include people who work in agriculture. The payroll data don’t capture the self-employed, newly formed businesses, or domestic employees. For that reason, the survey might be expected routinely to undercount the number of adults who are actually working. In September, it found that there were 129.9 million Americans with “jobs.”

By contrast, the Household Survey is based on surveys of individuals in 60,000 households. It includes farm workers, the self-employed, and people who may work off the books, such as nannies. And it has a rather liberal interpretation of what constitutes a job. “If you did any work for pay or profit in their own business, even as much as one hour of work during the survey reference,” that counts as a job, says BLS economist Karen Kosanovich. (For a full explanation of how the government calculates unemployment, see here.) In September, the Household Survey found there were 137.6 million Americans with jobs.

That’s a big difference—nearly 8 million—especially at a time when the numbers have such large political consequence. So, you can see why Bush partisans might want to boost the Household Survey.

By relying on existing businesses, the Establishment Survey may miss potential growth, especially when a recovery is underway. But it doesn’t only underestimate. There are ways in which the Establishment Survey can overcount.Say you work 20 hours at Starbucks and 30 hours at Wal-Mart in a week, you’ll be counted as two payroll jobs when you’re really only one employed person. There are about 7 million such multiple job holders in the United States. Plus, while it’s slow to count new companies, the Establishment is also slow to register dying companies. “The payroll survey probably overstates the weakness in the job market. I just don’t think it overstates it by any significant amount,” says Mark Zandi, the straight-shooting chief economist of

And even with its flaws, the Establishment Survey is “measurably more accurate than the Household Survey,” says Zandi. It’s bigger and more comprehensive. BLS says it has become more adept at latching onto new companies, by, for example, checking state records more frequently.

The Household Survey has its share of fuzzy math, too. BLS economists make calculations based on sampling and the census’ estimate of population, which is a constantly moving target. One of the reasons the household figures look so good is that a census revision caused it to add more than 500,000 working Americans in January.

It’s important to remember that the monthly numbers BLS reports are preliminary and the bureau often revises the monthly data it has already reported. In his paper, John Kitchen suggested that in a period when the Household Survey is rising, we should expect the payroll surveys to be revised sharply upward—in other words, BLS should provide ex post facto acknowledgement that it misread the true health of the job market. But so far in this cycle, that has yet to happen—in fact, BLS is actually revising recent figures slightly downward—suggesting that the Household Survey increases aren’t yet as significant as its advocates hope.

The Establishment Survey may not be the best means for predicting job growth in the future. But the Household Survey, it turns out, isn’t necessarily a better means for predicting job growth in the past.