Skip to main content
economy lab

Compared to a typical opinion poll, a U.S. jobs report comes from a massive sample. But it’s not big enough to get more than a rough idea about what’s really happening in the labour market of a country of more than 313 million people.

The significance placed on the labour market surveys is wildly out of proportion to the reliability of the data.

North American stock markets were lower Monday as investors had little to distract them from Friday's U.S. jobs report, which showed non-farm payrolls increased by 120,000 in March, weaker than most on Wall Street had expected and the lowest in five months.

Here's the thing: there's a 90 per cent chance the reading was off by about 100,000 in either direction. So investors could be dumping stocks when in reality the U.S. economy could have added in excess of 200,000 jobs. Or, the employers might have shed considerably more positions than 120,000, suggesting Monday's sellers should pick up the pace.

The margin of error around employment surveys is extremely wide.

The headline U.S. jobs figured is derived from a poll of about 141,000 businesses and government agencies, representing roughly 486,000 individual worksites, according to the Bureau of Labour Statistics (BLS).

Compared with your typical opinion poll, that's a massive sample. But it's not big enough to get more than a rough idea about what's really happening in the labour market of a country of more than 313 million people. Accordingly, the BLS explains that there is a 90 per cent chance that its results will be within 1.6 standard errors from the "true" value. In the case of the establishment survey that means the BLS is fairly confident that the reported number is within about 100,000, plus or minus, of the actual number of people hired or fired in the month.

"All of the U.S. labour surveys have such large errors that one needs to be careful in not getting carried away with monthly changes in job growth that may not be statistically significant," Derek Holt and Dov Zigler of Bank of Nova Scotia said in a report released last week.

None of this is a secret, exactly.

The BLS explains the limits of its survey at the end of each report, as does Statistics Canada when it releases monthly employment figures. (StatsCan's Labour Force Survey covers 56,000 households and comes with a standard error of 27,200. According to the Bank of Nova Scotia analysis, that translates into a 90 per cent chance that Canada's actual hiring in a given month is within 43,500, plus or minus, of the reported figure.)

So one must assume the investors trading off the March labour data Monday know they could be betting against a mirage. Rarely does the U.S. economy generate hundreds of thousands of jobs for months in a row without taking a breather. The U.S. mostly shed jobs through 2002 and 2003 in a delayed response to the 2001 recession. By 2004, the labour market had recovered. The U.S. added 337,000 jobs in March, 249,000 in April and 310,000 in May. Then hiring slowed to 81,000 in June and 46,000 in July. August brought a rebound to 122,000, and employers followed by creating 161,000 jobs in September and 348,000 in October.

The actual number of jobs created each month isn't important; it's the trend that matters. And so far, the trend in the U.S. labour market is positive. Some economists can't help but worry that the March figures represent a spring slowdown like the one last year that prompted the Federal Reserve to shift course. That's possible, of course. But it's impossible to know that based on number that could be 100,000 off the mark.