Half a million
filings.
Not half a million
new employers.
September's Census Bureau release shows a modest increase in business applications and a stronger rise in applications with planned wages. The important story is the distance between starting the paperwork and starting a payroll.
The employer-formation figure is a separate, forward-looking projection.
A busy front door, not a finished building
A new business can begin long before anyone opens a storefront, buys a machine or hires a first employee. It can begin with a tax identification application. That makes the paperwork useful economic evidence, but it also makes the headline easy to overread. An application is a signal of activity at the entrance to the business-creation process. It is not proof that a new employer has arrived at the other end.
The Census Bureau reported on October 14, 2026, that seasonally adjusted business applications reached 534,947 in September, up 0.5% from August. Its separate projection estimated that 29,022 employer businesses would emerge within four quarters from September's application cohort, up 1.4% from the equivalent projection for August. These are two different measurements with two different clocks. Neither should be presented as the number of businesses that opened their doors during September. [1][2]
One month of applications, a future window of formations
The application count groups filings by the month in which they were made. The projected formation count looks forward from that group. Census is estimating how many of those applications will become businesses with payroll tax liabilities within a specified window. The four-quarter figure therefore describes an expected outcome for September's applicants, not a completed count of all September startups.
That distinction matters even when the two figures move in the same direction. A month can generate many applications without producing an equally large increase in prospective employers. Conversely, an improvement in the characteristics of applications can strengthen projected employer formation without a dramatic jump in the total number of filings. Reading both measures is more informative than treating the larger number as the whole story. It also avoids turning a forecast about a cohort into a claim about jobs already created.
Similar names do not mean interchangeable measures
Census reported 147,331 high-propensity applications in September, up 1.3% from August. Applications with planned wages rose 4.3% to 35,786. Applications from corporations fell 0.2% to 40,951. All four headline application series in the release are seasonally adjusted. The stronger monthly gain in the planned-wage series is worth noticing, but it is still a change in applications containing a particular characteristic, not a count of employees hired. [2]
These categories are not four separate piles to add together. High-propensity applications are a subset of business applications. Planned-wage and corporate applications sit within the high-propensity group, and the categories can overlap. A chart that stacks them as if each represented a different set of firms would overstate the total. A diagram that presents them as mandatory successive stages would also be misleading: a business does not need to pass through every named category in a fixed order. [3]
What high propensity actually identifies
The high-propensity classification uses characteristics on an Employer Identification Number application that have been associated with becoming an employer. Those characteristics can include corporate organization, an indication of hiring, a planned first-wage date, or particular industry classifications. It is a statistical grouping, not a government certification that a business will succeed, create a certain number of jobs or receive financing. [3][5]
This is why the adjective should not be translated into a promise. A high-propensity application can fail to become an employer, while an application outside that category may eventually develop in ways that the initial form did not reveal. The measure is useful because it organizes signals available early. Its usefulness depends on keeping the signal separate from the eventual outcome. For readers following the economy, the relevant question is whether the application mix and subsequent formation evidence tell a consistent story over time.
A stronger monthly reading, with a narrow meaning
The planned-wage series rose in all four Census regions in September. The Northeast increased 6.6%, the Midwest 5.6%, the South 4.1% and the West 2.6%. Nationally, the 4.3% increase was larger than the 0.5% rise in total applications. This is a more employment-oriented signal than the broad total because the underlying applications include a first wages-paid date. It remains a statement of planned activity on a form. [2][3]
It does not tell us how many workers each applicant expects to hire, what those jobs would pay or how long they would last. Nor does it establish that every applicant subsequently met the indicated date. A useful follow-up would compare the signal with later employer-formation evidence. A less useful shortcut would multiply the number of applications by an invented average workforce and call the result new employment. The release does not supply the basis for that calculation.
A national increase can hide regional divergence
Total applications rose 2.6% in the Midwest, 1.2% in the South and 0.3% in the Northeast, while falling 1.8% in the West. The South accounted for 235,028 applications, compared with 130,070 in the West, 94,891 in the Midwest and 74,958 in the Northeast. These are levels for differently sized regions, not measures of entrepreneurial intensity per resident. A larger region can produce more filings without having a higher application rate. [2]
The West illustrates why the composition matters. Its overall application count declined, yet high-propensity applications increased 2.0% and planned-wage applications rose 2.6%. There is no contradiction in those movements because the series describe different subsets. The result does not establish a regional turnaround or decline by itself. It shows why a single aggregate can miss changes in the kinds of applications being filed.
Projected employers tell another regional story
The four-quarter formation projections also differed by region. Census projected 11,302 employer formations from the South's September applications, 8,319 from the West, 4,773 from the Midwest and 4,628 from the Northeast. The monthly changes were minus 0.2% in the South, plus 2.4% in the West, plus 1.9% in the Midwest and plus 3.0% in the Northeast. These are projections tied to application cohorts, not current counts of newly operating regional employers. [2]
The South's rising application total and slightly lower projected formation figure are a useful reminder that volume and expected employer formation are not equivalent. They do not, on their own, reveal why the relationship changed. Establishing a cause would require more evidence about the applications and the model, not a story inferred from two arrows pointing in different directions. This article treats the difference as a question worth following, not an explanation already proved.
Retail leads the count, but a filing is not a shop opening
Retail trade recorded 110,048 applications in September, up 4.7% from August. Professional services recorded 78,937, down 1.7%, while construction had 48,078, up 0.1%. Information applications rose 4.3% to 13,806. Manufacturing increased 2.2% to 8,114. The industry table is a view of applications classified by activity; it is not a census of newly completed stores, offices, construction projects or factories. [2]
The distinction is especially important for activities that can operate without a conventional commercial location. A large retail application count should not be converted into a claim that the same number of physical shops is about to open. Likewise, a rise in information filings is not evidence that a particular technology caused the increase. The release supplies a classification and a change, not a causal account of applicants' motivations or business models.
Industry labels can change as information improves
Census assigns industry codes using information available from administrative sources and its coding processes. The release explains that initial assignments can be revised as more detailed information becomes available. The next annual update is scheduled with the October data release on November 12, 2026, and is expected to revise industry information, formation data and seasonal factors. A revision is therefore part of the statistical process, not automatically evidence that the earlier release was unusable. [2]
For anyone maintaining a chart, this creates a practical discipline: compare data from a consistent release vintage. A number copied months ago may not match the latest historical series. Combining an old baseline with a revised current observation can create a change that partly reflects statistical maintenance. The right response is to label the retrieval date, retain the source and explain material revisions, rather than treating every difference between saved tables as a new economic event.
A definition change is not a sudden change in behavior
Census says that, beginning with the January 2026 monthly release, applications associated with internet sales were excluded from the high-propensity and corporate application series. The definitions were updated across the entire time series. That detail matters when interpreting historical comparisons: the current series should be compared with its consistently revised past, not casually spliced to an older definition. The broad application series and the narrower employer-oriented series answer different questions. [6]
A reader does not need to master every coding rule to use the release responsibly. But a chart should make clear which series it shows and whether its history is on the same basis. Otherwise, a methodological boundary can be mistaken for a sudden change in entrepreneurs' intentions. The broader lesson is simple: the label on a statistic is part of the evidence, not decoration beneath a graph.
Self-employment is not a failed version of an employer firm
The formation measure is designed around the emergence of wage-paying businesses. A person can operate a legitimate business without employing anyone else. Some sole proprietors without employees do not necessarily need an EIN, and businesses begun without one may not appear in these application statistics. Census also excludes specified types of applications and activities from the core measures. The dataset is extensive, but it is not a complete register of every way Americans earn business income. [4]
This is why the difference between applications and projected employer formations should not be described as a failure count. The arithmetic remainder contains more than one possible path, and the projection has a defined time horizon. Some applicants may operate without payroll, some may take longer to develop, and some may not proceed. The release does not divide the remainder into those outcomes. Calling all of it failure would turn an absence of that particular measured transition into a judgment the data do not support.
New business activity is larger than this one measure
Census explains that these formation measures focus on entirely new employer businesses originating from applications. They do not include new employer operations formed by existing firms in the same way. An established company opening another location and an entirely new firm hiring its first worker are different events, even though both can matter to local employment and commercial demand. Business dynamism includes several processes that should not be collapsed into one number. [4]
For a city watching its downtown, the relevant evidence might include occupancy, permits, employer records and actual hiring alongside applications. For a national reader, formation data can complement broader employment and business-demography measures. The point is not that the application release is inadequate. It is that using a specialized measure for its intended purpose is more informative than asking it to describe every kind of expansion, contraction and entrepreneurial activity at once.
Four quarters and eight quarters are different windows
Census also projected 39,703 employer formations within eight quarters from September's applications, up 1.0% from the analogous August projection. The longer horizon gives applications more time to become employers. It is not a second independent group that should be added to the four-quarter figure. Both measures refer to outcomes associated with the same monthly cohort, observed or projected over different windows. [2]
A longer window can be useful when businesses need more time to organize operations. But it also changes the question being asked. Comparing a four-quarter figure in one month with an eight-quarter figure in another would mix horizons and produce an invalid trend. The time window belongs in the headline or chart label, particularly when a number is described as projected business creation. Without it, readers cannot tell whether the claim concerns a near-term transition or a longer development period.
No sampling error does not mean no uncertainty
The release is based on administrative data rather than a sample survey, so Census says the statistics are not subject to sampling error. That does not make every number a perfect observation of economic reality. The agency identifies possible nonsampling errors, including recording, coding, reporting and coverage problems. The projected formation series also carries projection error. The report marks statistical significance as not applicable or not measurable for the displayed changes. [2]
A careful account can report a 0.5% increase without declaring a statistically significant surge. It can identify a stronger planned-wage reading without promising a hiring boom. Those are not timid qualifications; they define what the evidence can support. The most useful interpretation follows the same series over time, checks revisions and looks for corroborating outcomes, instead of asking a small monthly change to settle a much larger argument about the economy.
Existing owners face a different set of questions
A separate NFIB survey of its members found that small-business optimism slipped 0.8 points to 97.9 in September. Only 7% of respondents, seasonally adjusted, described it as a good time to expand, while 32% reported job openings they could not fill. The membership survey and the Census application data measure different populations and different phenomena. They should not be combined into a single score or treated as a direct test of the same firms. [7]
The contrast is still useful context. Entering the application process and deciding whether to expand an operating business are not identical choices. A flow of prospective ventures can coexist with caution among established owners. Neither dataset alone tells us how the other group will behave. The comparison is a reason to separate the stages of business development, not proof that one source is wrong.
From an application to a functioning workplace
The gap between paperwork and payroll is where many practical questions arise. Is there a workable product or service? Can the business find customers, suitable premises, equipment and the people it needs? These are general questions about turning a plan into an operation, not findings about the individual applicants in September's release. The data do not tell us which obstacle mattered most or whether any particular applicant encountered it.
For policymakers, this distinction encourages evaluation that follows outcomes rather than celebrating a single input. More applications may be welcome, but a program intended to support employers should also be assessed against relevant employer outcomes. For readers, it encourages skepticism toward claims that a filing total alone proves the success of a tax change, a technology or a local initiative. Causation needs evidence beyond the timing of a release and a plausible story.
Watch the transition, not just the headline
The next scheduled release is November 12, 2026. Because it includes the announced annual update, readers should look for revisions as well as the new month's observations. The useful comparison will keep the series, adjustment basis and projection horizon consistent. It will also distinguish a change in the current estimate from a change in previously published history. Census's release notice is the source for the schedule, which can change. [2][6]
September's result is neither a count of half a million new employers nor a verdict that most would-be businesses have failed. It is a timely view of applications, their characteristics and their expected transition into payroll businesses. The headline number tells us that the front door remained busy. The harder economic question is what happens after people walk through it. Answering that requires the patience to follow a cohort beyond the filing date.
Six questions to keep beside the numbers
Did 534,947 new companies open in September? No. That is the seasonally adjusted application count, not a count of openings. [1]
Did 29,022 employers start operating in September? No. It is the projected number emerging within four quarters from September's application cohort. [1]
Can the four application categories be added together? No. They are overlapping and nested measures, not mutually exclusive categories. [3]
Does a planned-wage application count employees? No. It identifies an application with a first wages-paid date, not a number of workers. [3]
Does a business without employees count as a failure? No. Employer formation is a specific statistical outcome, not a general judgment about a business. [4]
What should be checked in the next release? The new observations, revised history, definitions and seasonal-adjustment basis, with the same projection horizon used for comparison. [2][6]
Sources & reporting notes
- Census Bureau: September 2026 release, October 14, 2026
- Census Bureau: full September 2026 report and explanatory notes
- Census Bureau: definitions of application and formation series
- Census Bureau: frequently asked questions and coverage
- Census Bureau: methodology
- Census Bureau: annual updates and methodological announcements
- NFIB: September 2026 Small Business Economic Trends
The Census release, its PDF tables and methodological notes were cross-checked. They are parts of the same official statistical system, not independent surveys. NFIB is a separate membership survey and is used only as context, not as a validation of the Census formation projection. All business and people images are newly generated symbolic illustrations. No interviews with the depicted people are implied.
September 2026 data were released October 14, 2026. Comparisons are month over month and seasonally adjusted unless explicitly stated otherwise. The release identifies projection error and possible nonsampling error; its displayed changes are not described here as statistically significant.
Comments
Post a Comment