Measuring Economic Growth
How fast or how slow is the economy growing?
One metric of growth to which analysts frequently turn is the monthly change in payroll employment. By that measure, the economy is just fine.
Until August 21, 2024, the country had seen strong employment growth: from July 2023 through July 2024, the economy produced 2,512,000 jobs, or about 210,000 per month. From March 2023 through March 2024 (dates we will come back to), jobs grew by 2,900,000, for a monthly gain of 242,000.
But did the economy really produce that many jobs? If not, is our view of the state of the economy what we think it is? The answers are no, and not really.
It turns out the U.S. economy did not produce as many jobs during that March-to-March period as we thought.
The BLS Downward Revision
The Bureau of Labor Statistics
announced a preliminary downward revision of 818,000 jobs for 2024 on August 21, 2024. BLS published the draft results of its annual benchmarking exercise, which stated that employment gains were previously overstated by 818,000 through March of 2024.
This preliminary revision, should it hold up to the final revision announcement in February 2025, means that the estimates for job gains between March 2023 and March 2024 will be lowered, as will all the months since then for which BLS published employment numbers.
This is the biggest revision since 2009, when the Bureau reduced non-farm employment by 902,000. As you can see from Figure 1, the Bureau always finds a reason to revise its employment estimates, but big changes are rare. In fact, the average revision is a miniscule 0.1 percent of total employment. This 0.5 percent change is noteworthy.
Figure 1
Source: Author’s chart using BLS data.
This revision means that total job growth between March 2023 and March 2024 now stands at 2,082,000 and the monthly average falls to 173,000. If BLS applies the same level of downward revision on the July 2023 to July 2024 period, the total job growth will fall to 1,694,000, reducing the average monthly gains down to 141,000.
[1]
Unfortunately, the economy currently needs to produce an average of 180,000 jobs per month to keep up with population growth, and neither revised number does that.
Some avid watchers of BLS’s monthly jobs report have noted the increasing prevalence of downward revisions to initial jobs estimates. BLS collects survey returns over a three-month period, which means that the first two months are preliminary estimates before being finalized in month three.
Figure 2 shows the total revision each month between March 2023 and March 2024.
Figure 2
Source: Author’s chart using BLS data.
As seen in Figure 2, additional data collected over the two months after initial collection led to downward revisions in 9 out of 12 months.
[2]
This may mean that the past year’s economy was weaker than the BLS projections model expected. The projections model BLS employs to produce initial estimates use extensive historical data. Thus, repeated misses on the downside could indicate weakness in the economy when compared to history; that undetected weakness could also have been at the root of the big benchmark revision.
BLS Downward Revisions by Sector
In addition to the preliminary change to total employment and the pattern of monthly revisions, we also got a glimpse at revisions by sector. This offers us an additional clue for what caused the negative benchmark.
The biggest changes occurred in the following sectors:
| Sector |
Revision |
| Professional and Business Services |
-358,000 |
| Leisure and Hospitality |
-150,000 |
| Retail Trade |
-129,000 |
| Manufacturing |
-115,000 |
What likely happened to cause these big downward revisions? These sectors, particularly the first three in the table above, are dominated by small firms; they are historically very sensitive to changes in consumer and business consumption patterns. In this case, the sluggish economy may have caused more firms to dissolve than BLS originally estimated.
I strongly suspect that BLS’s “birth/death” model is to blame for the lower employment estimates.
A Challenge for the Model
This model is used by the agency to estimate the number of firms that are being formed or disappearing between the March benchmark months.
BLS looks at a complete count of firms each March called the Quarterly Census of Wages and Employment (QCEW) to get an actual count by sector of the number of firms with employees. The QCEW is based on Unemployment Insurance records, which means the QCEW is a complete snapshot of active firms, but it leaves the March count as the only complete count. BLS uses the birth/death model between March of each year to estimate business formations and business dissolutions.
This usually reliable model is challenged, however, when the US economy is recovering from slow times or descending into sluggish economic activity. The model underestimates employment gains from new firms when the economy is warming up and overstates gains when it is cooling down.
That is what I suspect happened this time. The economy between March of 2023 and March of 2024 was a good deal cooler from an employment standpoint than the model predicted, or experts believed.
BLS likely saw this overstatement once they compared their projections to the actual count of active firms.
What Happens Next
What happens between now and the final revisions that will be published in early February of 2025?
First, BLS will be carefully checking its work announced on August 21st. The BLS experts who do the revisions are some of the most careful statisticians in an agency of well-above-normal careful statisticians.
Second, the revisions staff will be aligning published estimates since March to the lower March number.
All these changes will be revealed in February when the Bureau publishes a lengthy essay explaining why it changed total non-farm employment.
For additional background prior to that publication, I recommend reading the
benchmark article for 2023.
Options for Congress to Consider
Major revisions have major consequences and Congress certainly has a role to play in this process going forward. So, what can Congress do?
First, these benchmark revisions matter to public policy, since policymakers rely so heavily on BLS labor market data in formulating the country’s economic policies. This revision is large enough to change the way we think about the current economy. Members are likely to make different policy decisions based on their picture of the economy. Congress should find out what went wrong in the initial estimates by conducting thorough oversight of the matter.
Second, Congress should act on those findings. BLS has plans for modernizing its surveys and the many ways the agency estimates the monthly numbers. Congress should use this opportunity to fund and prioritize critical modernization initiatives at BLS through the annual appropriations process.
Third, don’t shoot the messenger. The staff at BLS are not servants of either political party and maintain a strictly non-political, non-partisan workplace. To achieve better performance by BLS’s hardworking staff, Congress should show more, not less, support for their data-driven mission.
[1] We do not know how BLS will distribute the 818,000 fewer jobs by month. The Bureau does not simply apply a monthly average and subtract that number for the unrevised total. Thus, we must wait until the final announcement in February to see how the benchmark is applied to the period March 2023 through October 2024, the last month to be affected by this year’s revision process.
[2] Some analysts have added these monthly changes to the 818,000 benchmark revision to produce an even larger number. That is incorrect. The final monthly number achieved over three months of data collection is the number that the benchmark will be adjusting, not the initial monthly number that is revised twice. For example, if the initial estimate for April 2023 was 153,500,000 jobs, or a gain 200,000 jobs from the previous month and two months later had been reduced to 153,450,000; then some portion of the 818,000 total revision will be applied to this final non-farm employment number. The monthly revisions get us to a final total employment number, and that is the number that the benchmark revises.
