Every month, the Bureau of Labor Statistics releases employment data that shapes market sentiment, economic forecasts, and policymaking. From interest rate decisions at the Federal Reserve to unemployment insurance triggers at the state level, the influence of BLS data is far-reaching. And yet, with each release, one often overlooked note quietly accompanies the data: subject to revision.
To the uninformed, this might suggest inaccuracy or even manipulation. But the reality is rooted in how data is collected, processed, and interpreted. The act of revising economic data is not a flaw but a fundamental feature of any statistical system that prioritizes accuracy over speed. This article explores why the BLS revises its data, the mechanics of seasonally adjusted vs. not seasonally adjusted numbers, and how a real-world dataset employment among African American women in 2025 illustrates the complexity of labor market measurement.
The Bureau of Labor Statistics, founded in 1884, is the principal fact-finding agency for the U.S. federal government in the field of labor economics. Its core function is to measure labor market activity, working conditions, and price changes in the economy.
Among its most closely followed outputs is the monthly Employment Situation Report, which contains data on job growth or loss, unemployment rates, participation rates, and hours worked. These figures often headline national news and affect everything from political discourse to stock market performance. But the collection and interpretation of labor data is a dynamic process. No matter how carefully designed the surveys are, initial data releases are based on incomplete information and statistical models that must later be refined.
The BLS relies on two primary surveys to produce monthly employment estimates:
- Current Population Survey (CPS): Also known as the household survey, this samples about 60,000 households and is the source for data on unemployment, labor force participation, and demographic breakdowns.
- Current Employment Statistics (CES): Also known as the establishment survey, this collects payroll data from roughly 122,000 businesses and government agencies, covering over 666,000 worksites.
Because of tight deadlines for monthly releases—typically the first Friday of the following month—some employer reports are late, households may be unreachable, and administrative records may not yet be available. As more responses arrive over time, the BLS incorporates the additional data, which leads to two monthly revisions: a first revision one month later and a second revision two months after the initial release.
There is also an annual benchmark revision, where the BLS aligns employment data to comprehensive counts derived from state unemployment insurance tax records, which cover nearly all employers.
These revisions are not signs of incompetence or hidden agendas. Rather, they reflect the reality that high-frequency data collection must balance timeliness with completeness. Initial estimates are snapshots; revisions bring the picture into higher resolution.
Understanding Seasonally Adjusted vs. Not Seasonally Adjusted Data
Another common source of confusion is the distinction between seasonally adjusted (SA) and not seasonally adjusted (NSA) figures.
- Not Seasonally Adjusted (NSA): These are raw numbers taken directly from survey results. They reflect real, unaltered employment counts.
- Seasonally Adjusted (SA): These figures are modified using statistical models that remove predictable seasonal fluctuations—such as increased hiring in December or reduced construction jobs in winter.
Seasonal adjustment allows for clearer comparisons of month-to-month changes without the noise of recurring seasonal events. For example, employment traditionally rises in retail in November and December and drops in January. Without adjustment, these fluctuations could lead to misinterpretation of actual trends.
However, seasonal models rely on historical patterns. If a new shock occurs—such as a pandemic, atypical weather events, or irregular policy shifts—these models may not capture reality perfectly, requiring future refinements and adjustments to the seasonal factors themselves.
A Practical Example: African American Women’s Employment, 2025
To illustrate how data revisions and seasonal adjustments interact, consider the seasonally adjusted number of employed African American women over five consecutive months in 2025:
- March 2025: 10.300 million
- April 2025: 10.260 million
- May 2025: 10.332 million
- June 2025: 10.248 million
- July 2025: 10.247 million
At first glance, this dataset may seem inconsistent. Why the decline in April, a surge in May, and subsequent declines in June and July?
Several points are worth unpacking:
- Magnitude of Monthly Change
These monthly movements, ranging from about 50,000 to 80,000, may appear marginal, but in labor market terms, they represent significant shifts. These could be due to school-year employment cycles, changes in public-sector hiring, or temporary retail and service jobs. - Temporary Effects
The uptick in May could represent a short-term employment increase due to localized or sector-specific conditions—perhaps related to summer hiring, public campaigns, or fiscal year-end budgeting by employers. However, this doesn’t necessarily indicate a sustained improvement, as shown by June and July numbers. - Plateau, Not Decline
While there are ups and downs, the broader range—from 10.248 to 10.332 million—suggests a labor market that is relatively flat during this period. The volatility may be more reflective of sector churn than structural change.
If a future revision updates, for instance, July’s figure from 10.247 to 10.280 million, that revision would adjust interpretations about labor market strength. It may indicate more robust hiring than originally estimated. Conversely, a downward revision could reinforce a stagnation narrative. Revisions are standard practice across all major economic indicators. GDP figures are revised multiple times. Inflation statistics may be reweighted to reflect changing consumption patterns. The Census Bureau revises retail sales and trade data regularly.
In labor market statistics, revisions are particularly common because of the sheer scale and complexity of the data. Millions of businesses and households are involved, each contributing a piece of the larger puzzle. Moreover, revisions are conducted transparently. The BLS publishes revision histories, explains methodological changes, and allows the public to compare original and revised estimates. This openness is central to the integrity of the data, even if the revisions themselves can be politically or emotionally misunderstood.
A revision of 50,000 jobs may not seem impactful in an economy with over 150 million employed people. But such changes are statistically meaningful. For example, Federal Reserve interest rate decisions are often influenced by whether job growth appears to be accelerating or decelerating. A 0.1% change in employment may be the tipping point for a policy decision affecting credit costs for millions.
Revisions also matter for planning and budgeting by states, corporations, and local governments. Employment trends influence tax revenues, hiring plans, and social program allocations. A misinterpretation of the underlying data even if unintentional can have ripple effects through the economy.
While the BLS aims for statistical precision, the public conversation around the data is often shaped by headline figures and political narratives. This can result in overemphasis on preliminary numbers, even though they are explicitly marked as subject to change. It is important for observers, journalists, policymakers, and analysts to understand that early data is an estimate. Just as weather forecasts become more accurate as the date approaches, labor statistics become more reliable as more data is incorporated and models are refined.
Understanding the architecture behind the data helps prevent premature or inaccurate conclusions about the state of the economy. The BLS operates under dual pressure: provide timely data and ensure its long-term accuracy. These goals are inherently in tension, but both are critical. Without timely data, markets and policymakers would be flying blind. Without accuracy, trust in the data would erode, leading to poor decisions and broader skepticism of institutions.
Revisions are not a sign of error. They are the result of a methodical, transparent process aimed at refining the initial picture of the economy into a more complete and accurate one. For analysts and observers, the lesson is simple: understand the process, treat early numbers with caution, and always look at the data—both in the moment and over time—as a moving picture, not a still fram
Disclaimer: This article was assisted by ChatGPT.
HBCU Money 