Counting the number of jobs is very, very hard.

In the Morning Jolt yesterday, I wrote something I want to unpack a little more:

It’s easy to armchair-quarterback the estimates after the fact, but if someone could produce a more accurate jobs report than the BLS, there are lots of people on Wall Street who would pay big bucks for it, because it would give them an upper hand in economic forecasting.

It’s tempting to say that private organizations could better estimate the number of jobs. Private-sector alternatives to government usually are better! But the BLS was genuinely much better than the private-sector alternative at estimating employment in real time. And we know this because the private sector told us so every month by closely watching the BLS jobs report.

The profit motive here is strong: Better information would allow for investors to make better decisions, and the state of employment is important information to have. Investors don’t rely solely on the BLS reports, of course. They look at many different indicators, publicly and privately produced. But they wouldn’t bother to look at the BLS reports if they weren’t any good.

And sure enough, the BLS beats the pants off the top private-sector alternative. Gary Wagner, a professor of economics at the University of Louisiana at Lafayette, made the following graph:

ADP is a major payroll-processing company. It has a pretty good idea of what the job market looks like, and it puts out its own labor market data. They’re good! But the BLS data have been much better.

Wagner is comparing the estimates from the BLS and ADP to the Quarterly Census of Employment and Wages (QCEW). It’s what the BLS uses to revise its estimates from the monthly jobs report. It offers better-quality data, but because it’s a census rather than a survey, it takes longer to collect and only comes out once a quarter.

The BLS also produces the QCEW, which is what makes the entire spat over the revisions somewhat comical. The allegation is that the BLS is biased because it makes revisions, but the only way to know revisions are needed is a different product from the BLS. If the BLS were actually biased, why wouldn’t the QCEW be biased, too? You could get sucked into a wormhole of conspiracy theories, or just use Occam’s razor and say they’re revising estimates based on the later availability of better data.

You can see on the graph that the BLS has been getting a little bit further away from the zero bound in the last few years. That’s in large part due to declining survey response rates. And the errors were usually larger between 2011 and 2013. The size of BLS errors don’t demonstrate any relationship to which party is in power, and the fact that they have increased and then decreased demonstrates that the agency has been able to learn and improve over time. The agency is working on addressing the falling response rates (with the help of an advisory group that Secretary of Commerce Howard Lutnick disbanded earlier this year).

You can also see that ADP’s estimates have been getting a lot better. Maybe someday they’ll be consistently better than those of the BLS! But they aren’t, which is why the private sector continued to pay attention to the BLS even though the ADP data were always available.

Counting the number of jobs is very, very hard. And the BLS was very, very good at it. For the worst miss to be less than 1 percent and the median miss to be 0.29 percent is an incredible testament to the agency’s professionalism and competence, built over years of experience. Tearing that down is a whole lot easier than was building it up.