No, the Biden Administration Is Not Manipulating Jobs Data

The Bureau of Labor Statistics issued its annual revision of the jobs numbers today, showing that it had overestimated job growth between March 2023 and March 2024 by 818,000 jobs. These revisions happen every year, but this was a bigger miss than normal.

Given the direction of the miss, some are attributing it to politics. Donald Trump called it a “massive scandal” and posted on Truth Social that “the Harris-Biden Administration has been caught fraudulently manipulating Job Statistics to hide the true extent of the Economic Ruin they have inflicted upon America.”

This seems unlikely to be the case, for the following reasons:

1. The premise of Trump’s complaint doesn’t make sense. Nobody “caught” the BLS. The BLS is revising its own data, on the same schedule it does every year.
2. If the BLS were trying to cover up worse jobs numbers for Democrats, it would not make sense to issue the revision now, as regularly scheduled, before voting begins. A cover-up would look something like delaying the revision until December.
3. The BLS missed in the other direction earlier in the Biden administration. In August 2022, the BLS revised job growth upward by 462,000 jobs, meaning it had undercounted job growth under Biden from March 2021 to March 2022. That was in the lead-up to the midterm elections, so at that time Democrats could have complained that the BLS was biased against them.
4. In August 2019, when Trump was president, this same regularly scheduled annual revision showed that the BLS had overcounted job growth by 501,000 jobs. Was the BLS manipulating data then to make Trump look better? Doubtful.
5. If the BLS were biased in favor of Democrats, it probably wouldn’t publish data showing that the union-membership rate is at a record low despite Biden’s efforts to run the most pro-union administration in American history and the media’s efforts to spin a “union resurgence.”
6. People vote based on what is actually happening, not based on what statisticians in Washington report. Even if this were some kind of scheme (which it is not), it would be ineffective and therefore unlikely to have been attempted.

The BLS revises its data because it publishes multiple jobs-related products, as economist Jeremy Horpedahl explains on his blog. The monthly Current Employment Statistics (CES) report, usually referred to in the media as “the jobs report,” is a survey of a sample of employers. The Quarterly Census of Employment and Wages (QCEW) includes nearly all employees.

Because the QCEW data is of better quality, the BLS uses it to “benchmark” the CES data. “The media reports the CES monthly data more prominently, because it is more timely and usually pretty close to correct — but benchmarking is the process to see just how correct those initial surveys were,” Horpedahl writes.

Here’s a fuller explanation from the BLS itself in a 2019 blog post (warning: very boring, skip if easily bored):

The CES is a monthly survey of approximately 142,000 businesses and government agencies composed of approximately 689,000 individual worksites. As with all sample-based surveys, CES estimates are subject to sampling error. This means that while we work hard to ensure those 689,000 worksites represent all 10 million worksites in the country, sometimes our sample may not perfectly reflect all worksites. So the monthly CES estimates aren’t exactly the same as if we had counted employment from all 10 million worksites each month. To fix this problem, we “benchmark” the CES data to an actual count of all employees, information that’s only available several months after the initial CES data are published.

In essence, we produce employment information really quickly from a sample of employers, then anchor that information to a complete count of employment once a year.

The primary source of the CES sample is the BLS Quarterly Census of Employment and Wages (QCEW) program, which collects employment and wage data from states’ unemployment insurance tax systems. This is also the main source of the complete count of employment used in the benchmark process. QCEW data are typically available about 5 months after the end of each quarter.

Each year, we re-anchor the sample-based employment estimates to these full population counts for March of the prior year. This process—which we call benchmarking—improves the accuracy of the CES data. That’s because the population counts are not subject to the sampling and modeling errors that may occur with the CES monthly estimates. Since the CES data are re-anchored to March of the last year, CES estimates are typically revised from April of the year prior up to the March benchmark. Then estimates from the benchmark forward to December are revised to reflect the new March employment level.

As Horpedahl points out, if you divide the 818,000 jobs over twelve months, it comes out to 68,000 jobs per month. “Over that time frame (March 2023 to March 2024), the average monthly job growth was 242,000 jobs. Crucially, the lowest monthly job growth during that period was 165,000 jobs, in October 2023,” he writes. So even if you subtracted 68,000 from each month’s jobs growth, the U.S. still added jobs in every month. In no way does the revision change our understanding of the economy to suggest it was ever in a recession.

There’s a larger question here about how to improve economic data to be more accurate and consistent. I wrote a Corner post in 2022 about the gaps between the household survey and the employer survey for the jobs market, and between measurements of gross domestic product and gross national income. Those are interesting problems for statisticians to solve, but they aren’t political in nature.

In the guessing game that precedes these data releases, J.P. Morgan Chase forecast a negative revision of 360,000 jobs, and Goldman Sachs forecast as many as 1 million. But forecasters agreed that a big negative number was coming, and that’s what happened.

That’s probably why markets didn’t really care about the revision, with the S&P 500 index closing up 0.4 percent on the day, and that’s probably the right reaction. The BLS regularly misses on both sides in its data work, which indicates that the misses are due to genuine statistical error rather than bias. Collecting data on a labor market of about 170 million souls is hard work.