There will be two large revisions in the next two job reports that will hugely affect the picture on both job creation and the number of people employed. It is important to understand that these revisions are part of the Bureau of Labor Statistics (BLS) standard process of collecting data, and not some arbitrary adjustments made by officials at the agency.
The first adjustment — which will be incorporated into the January jobs report that will be released this month — will be to the job growth that has been reported over the last two years. The job numbers that we hear reported every month come from a survey (Current Employment Statistics, or CES) sent to approximately 120,000 establishments. It is a very large survey, and usually is reasonably accurate, but it does fail to account for new businesses that are not included in the sample and businesses that do not respond because they have closed. To account for the missing businesses, BLS has a “Birth-Death” model that projects job gains or losses based on other economic data. This model tends to be far off the mark when the economy hits a turning point, as it did in the 2008-09 recession.
We find out how far the CES is from the mark when it is compared to the Quarterly Census of Employment and Wages (QCEW). BLS does this every year, producing a new benchmark for March of the prior year based on the QCEW data.
The QCEW data come from the unemployment insurance filings that businesses must send in regularly for all their workers. More than 99 percent of all employees are covered by unemployment insurance, so we can be confident that the QCEW data are giving an accurate picture of total employment.
BLS releases a preliminary figure for the benchmark revision in the summer. They announced this benchmark in September, saying it would show 911,000 fewer jobs than had been previously reported for March of 2025. The revised data are also likely to show somewhat slower job growth in subsequent months, since the birth-death model will be revised based on the new QCEW data.
The preliminary benchmark revision is usually, but not always, close to the final benchmark revision. In the benchmark for March of 2024 the preliminary figure showed an overstatement of 818,000 jobs, while the final benchmark revision incorporated into the data in January of 2025 showed an overstatement of 589,000 jobs. Nonetheless, it is reasonable to expect that the revised data for December of 2025 will show somewhere around 800,000 fewer jobs in the economy than had been previously reported.
That is a larger error than we would like to see. This is not a failing of BLS; it is a problem of trying to measure employment in an economy with more than 170 million workers. BLS is constantly trying to improve its methodology and responds to suggestions from academic and business economists. If someone has a better method, they should put it on the table. They would have a very receptive audience at BLS.
Population Controls in the Household Survey
BLS adjusts its population controls for the household survey every year based on data compiled by the Census Bureau. Usually, these adjustments are put into the survey in the jobs report released for January; due to the shutdown, the new controls will not be put in place until the February jobs report that will be released in March.
These controls provide the underlying number for both population totals and for specific groups like Black or Hispanic people. BLS surveys 60,000 households every month, but BLS needs the population controls to try to make the survey representative of the whole population.
For example, if the population controls from Census Bureau indicate that Black people are 14 percent of the population, but only 12 percent of the people who answered the survey are Black, they will increase the weight of each Black person in the survey so that the sum of their weights comes to 14 percent.
The controls also provide the basis for translating the ratios found in the survey into actual numbers. If the survey tells us that 60.0 percent of the population is employed, we need the population controls to tell us that this corresponds to 160,000,000 people being employed (60 percent of 266.7 million).
The population controls also impute a specific growth rate based on the rate of population growth we had been seeing. We know that the population controls put in place in January of 2025 hugely overstated the population at that point in time, and even more the subsequent growth in population. The reason is that the tighter immigration policies first put in place under Biden in June of 2024, and the even tighter one imposed after President Trump took office last January.
As a result, the civilian non-institutional population is likely to be around 580,000 lower (per Jed Kolko’s estimate) when the new population controls are put in place with the February jobs report. This will reduce the number of people reported as being employed by around 350,000. (*Note: see correction below.)
Again, this is a much larger error than is desirable, but it stems mostly from an extraordinary change in policy that could not have been anticipated. Here also, if economists or demographers have suggestions for making the survey more accurate, they would be warmly received at BLS.
We all would like accurate and timely data, but the numbers don’t just present themselves at our doorstep. BLS has to put large amounts of time in conducting its surveys and then making sense of the raw data to get meaningful statistics on things like unemployment, job growth, and wages. On the whole, it does a very good job.
Surely there are ways they can improve, but the key point is that the agency is open and transparent. Anyone who cares can learn exactly how they got any specific statistic — like the unemployment rate, the number of people working in manufacturing, or the rate of wage growth for restaurant workers. And an honest statistical agency is a tremendously valuable commodity.
*Correction:
Jed Kolko, who really knows this stuff, corrected me on the size of the downward revision to employment in the household survey. He puts the downward revision with the new population controls in the range of 350,000-380,000. My original figure of 4 million is obviously hugely overstated.
I got my original number not by any careful calculation, but rather a back of an envelope calculation based on a news report. After seeing Jed’s careful analysis, I realize I need a new envelope. Anyhow, this will still be a substantial downward revision, even if far smaller than I had calculated. The important point is that BLS is following long-standing procedures, it is not cooking the books.
More By This Author:
The AI Bubble: Consumption Is Outrunning WagesDoing Well By Doing Good: Dump Your American Stocks
Donald Trump’s $300 Billion Temper Tantrum Over Canada
Comments
Log in or sign up to join the conversation.