Four Branches Define High Propensity, and the Four-Quarter Projection Falls Below Planned Wages

Six numbers for August 2026 landed on one page on September 11, and two of them are routinely quoted as though they measured the same event. Business Applications came to 531,728. Projected Business Formations within four quarters came to 28,501. Both describe the same month of filings for an Employer Identification Number. One is about 18.7 times the other.

That distance is neither an error nor a revision. It is the span between a form reaching the Internal Revenue Service and a payroll record appearing for the filer, and the Business Formation Statistics measure both ends of that span with twelve national series. Four of them count applications: 531,728 for the total, 145,387 for high propensity, 34,263 for planned wages, 41,075 for corporations. Eight count or project formations. Follow 531,728 forward and it narrows to a screen with four branches, two of which carry their own published series and two of which do not.

One Month of Filings, Twelve Series

The outer boundary comes first, because the headline total is already a filtered number. The U.S. Census Bureau, Business Formation Statistics: Key Definitions sets Business Applications out as a subset rather than a census of filings: “Includes all applications for an EIN, except for applications for tax liens, estates, trusts, or certain financial filings, applications outside of 50 states and DC or with no state-county geocodes, applications with certain NAICS codes in sector 11 (agriculture, forestry, fishing and hunting) or 92 (public administration) that have low transition rates, and applications in certain industries (e.g. private households, civic and social organizations).”

That exclusion list does two different jobs. Tax liens, estates and trusts are removed because they are not business starts. Within sectors 11 and 92 the cut is narrower and empirical: it reaches applications with “certain NAICS codes” in those sectors “that have low transition rates” — the same criterion that drives the high-propensity screen further down. Boundary and subdivision rest on one idea, transition probability, applied twice at different strengths. The 531,728 figure is already the output of a propensity judgment before any of the four rows is reached.

The U.S. Census Bureau, Business Formation Statistics, August 2026 states the headline plainly: “Business Applications for August 2026, adjusted for seasonal variation, were 531,728, a decrease of 7.8 percent compared to July 2026.” The formation line in the same release reads: “Projected Business Formations (within 4 quarters) for August 2026, adjusted for seasonal variation, were 28,501, a decrease of 4.6 percent compared to July 2026.” Both lines carry the same seasonal adjustment and the same reference month, and neither revises the other.

Four Branches Decide High Propensity

The second row, 145,387, is 27.3 percent of the total, and it is where the definition does its work. The screen read here is the one the page labels for 2012 onward. The Key Definitions page describes the screen as a characteristic test taken from the filing itself: “The identification of high-propensity applications is based on the characteristics of applications revealed on the IRS Form SS-4 that are associated with a high rate of business formation.” Then it enumerates the branches: “High-propensity applications include applications: (a) from a corporate entity, (b) that indicate they are hiring employees, (c) that provide a first wages-paid date (planned wages); or (d) that have a NAICS industry code in accommodation and food services (72) or in portions of construction (237, 238), manufacturing (312, 321, 322, 332), retail (44, 452), professional, scientific, and technical services (5411, 5413), educational services (6111), and health care (621, 623).”

Four branches, joined by “or.” An application needs one of them. That matters for what the row can be read to mean, because the branches are not the same kind of evidence. Branches (a), (b) and (c) are statements the filer made about itself. Branch (d) is an industry classification, and it admits an application on the base rate of its sector rather than on anything the filer declared.

Branches (b) and (c) look interchangeable in summary and are separate boxes on the form. The Internal Revenue Service, Instructions for Form SS-4 handles the employee count at Line 13: “Complete each box by entering the number (including zero (-0-)) of Agricultural, Household, or Other employees expected by the applicant in the next 12 months.” The wages date sits at Line 15: “If the business has employees, enter the date on which the business began to pay wages or annuities. For foreign applicants, this is the date you began to pay wages in the United States. If the business doesn't plan to have employees, enter 'N/A.'” One box asks for a headcount in a forward window. The other asks for a date, and its N/A option is conditioned on the business not planning to have employees.

Branch (a) is narrower on the form than the word corporation suggests. Line 9a reads: “This box is for any corporation other than a personal service corporation. If you check this box, enter the income tax form number to be filed by the entity in the space provided. If you entered '1120-S' after the Corporation checkbox, the corporation must file Form 2553 no later than the 15th day of the 3rd month of the tax year the election is to take effect.” So the corporate box excludes personal service corporations and admits a filer that will elect S-corporation treatment later. The Census row re-merges what the form separated: Business Applications from Corporations is “High-Propensity Business Applications (HBA) from a corporation or personal service corporation, based on the legal form of organization stated in the IRS Form SS-4.”

Where the Two Published Subsets Stop

The release prints a row for branch (a) and a row for branch (c), and nothing for branches (b) and (d). Business Applications from Corporations came to 41,075 in August. Business Applications with Planned Wages came to 34,263. Add them and the total is 75,338, which is 70,049 short of the 145,387 high-propensity row.

That shortfall is a floor rather than a measurement. Both subsets sit inside the same high-propensity row, so an application that checked the corporate box and also entered a wages date belongs to both. The at-a-glance table carries four application rows and no intersection row, so the published figures support only a bound: the union is 75,338 or less, and the share admitted by branch (b) or branch (d) alone is 70,049 or more.

The regional columns show why that distinction is not cosmetic. In the Northeast, corporations contributed 9,284 applications against 4,824 with planned wages; in the Midwest the ordering reverses, 4,676 against 6,067. A reader holding one subset would describe the month differently from a reader holding the other. The high-propensity share moves too: 23,048 of the Northeast's 74,664 applications cleared the screen, 30.9 percent, against 61,792 of the South's 232,595, 26.6 percent.

The month-over-month figures diverge the same way. The national total fell 7.8 percent, the high-propensity row 4.1 percent, the planned-wages row 1.9 percent and the corporate row 8.1 percent. In the Northeast the total fell 4.9 percent and the corporate row fell 19.2 percent. A sentence about applications falling in August is true of the total and the corporate row and much weaker as a claim about planned wages, the hazard that faultlinesweekly on benchmark revisions landing on one month traced through the payroll data.

The Four-Quarter Projection Lands Below the Wages Count

Set the formation side beside the application side and a specific ordering appears. Projected Business Formations within four quarters for August was 28,501; Business Applications with Planned Wages was 34,263. The projection is 5,762 below the count of applications that named a first wages-paid date.

That comparison has to be stated exactly, because the two figures do not share a base. The Key Definitions page builds the projection on the whole total, not on a subset: Projected Business Formations within four quarters is “The projected number of employer businesses that originate from Business Applications (BA) within four quarters from the month of application.” So the model ran over 531,728 applications and still returned a four-quarter figure beneath the 34,263 that declared a wages date. The declaration on Line 15 is a statement of intent with a date attached; the projection is a probability-weighted expectation over a fixed window.

Widen the window and the ordering flips. The eight-quarter projection for August was 39,292, which is 5,029 above the planned-wages count and 10,791 above the four-quarter projection. On the shorter horizon intent exceeds projected formation; on the longer horizon projected formation exceeds intent. Nothing about the filings changed between those two readings.

As shares of the total the two horizons are 5.4 percent and 7.4 percent. The eight-quarter projection also runs 106,095 below the high-propensity row, but that difference is not a count of applications that failed: both figures rest on the same 531,728 total, and the projection is modelled over that total rather than over the screened row. High propensity names a screen output, not an expected outcome.

Which Series Is Counted and Which Is Modeled

The eight formation series split into counted, projected, spliced and duration pairs, and only one of those pairs is an administrative count. The U.S. Census Bureau, Business Formation Statistics Technical Documentation: Methodology describes the projection machinery: “The projections are based on an econometric model that generates estimates of the likelihood that a business application turns into an employer business.” The counted series has a hard stop written into its definition: “By definition, the end-point of this series is determined by the most recent quarter for which the administrative data identifies employer business startup activity based on first payroll observation.”

That end-point is why a projection appears in a monthly release at all, and why the spliced pair exists. Spliced Business Formations within four quarters is defined as a join: “This series combines (splices) BF4Q and PBF4Q to provide the entire time series for the actual and projected business formations within four quarters.” A spliced series is two estimators under one line, its internal boundary moving forward each quarter as administrative data arrives, and the seam is not drawn.

The duration pair measures the application-to-formation interval, though its published values were not retrieved. Average Duration from Application to Formation within four quarters is “A measure of delay between business application and formation, measured as the average duration (in quarters) between the quarter in which the month of business application falls and the quarter of business formation, conditional on business formation within four quarters.” The average is taken only over applications that formed inside the window, so it says nothing about the rest. The same documentation flags the frequency problem: “Appropriate caution is required in interpreting fluctuations since high-frequency weekly data are subject to fluctuations from seasonal factors including holidays and beginning and end of calendar year effects.”

An Independent Count of the Same Event

A payroll record appearing is also counted outside the Census program, on a different unit and a different clock. The U.S. Bureau of Labor Statistics, Business Employment Dynamics Overview defines its entry category this way: “These are either establishments with positive third month employment for the first time in the current quarter, with no links to the prior quarter, or with positive third month employment in the current quarter following zero employment in the previous quarter.”

Three things in that sentence keep it from substituting for the Census formation series. The unit is the establishment, not the applicant, so one new employer with two locations is two openings. The threshold is positive third-month employment, so a firm paying wages in a quarter's first month and none in its third does not open. And the second clause admits re-entrants, which are openings without being new. The clock differs as well: the data are “updated quarterly, usually six months after the reference quarter.”

So the two programs cannot be differenced to audit each other. They answer adjacent questions on different units, and the reference-period mismatch belongs to the same family as the one in faultlinesweekly on the August openings figure resting on one week.

Concrete Framework

Four steps turn a Business Formation Statistics headline into a defensible sentence.

First, name the series rather than the concept. Write WBA or CBA or PBF4Q, not new businesses. The four application rows for August were 531,728, 145,387, 34,263 and 41,075, and a sentence that omits the row cannot be checked.

Second, treat the subsets as nested rather than additive. Corporate and planned-wages are both subsets of the high-propensity row, so 34,263 plus 41,075 bounds their union instead of measuring it. Write at most 75,338.

Third, match the horizon before comparing. A declared wages date is a filing fact; 28,501 and 39,292 are model outputs over four and eight quarters, and the ordering against 34,263 reverses between them.

Fourth, check whether the series is counted, projected or spliced. A counted formation series stops at the most recent quarter with first-payroll evidence; a spliced series continues past that stop with modelled values. For a month as recent as August 2026, each formation figure on offer is a projection.

What to Watch Next Week

Three observable things will settle how much of this holds for September. First, whether the high-propensity row again falls more slowly than the total. In August the total fell 7.8 percent and that row 4.1 percent, lifting its share toward 27.3 percent; a reversal would mean composition is shifting rather than volume.

Second, whether the planned-wages row keeps outrunning the four-quarter projection. The August gap was 5,762. A narrowing gap implies either fewer declared wages dates or a higher modelled transition probability, and only one of those is a statement about filers.

Third, whether the corporate and planned-wages rows keep diverging regionally: 9,284 against 4,824 in the Northeast, 4,676 against 6,067 in the Midwest. Holding for a second month would make branch composition a regional feature rather than a monthly accident.

What Would Invalidate This

The arithmetic rests on one reading of one seasonally adjusted table, and four things would change it.

The 70,049 remainder is a floor, because the bound above assumes nothing about the intersection of the corporate and planned-wages rows. The release table read here carries four application rows and no intersection cell, so that overlap was not retrieved; were it published and small, the remainder would shrink.

The counted formation series was not retrieved. The current end-point quarter of the four-quarter and eight-quarter counted series, their published values, and the duration values all sat outside the material gathered. If a counted figure for an earlier month sits far from that month's projection, 28,501 deserves a wider band than anything here implies.

The not-seasonally-adjusted tables and the state detail were not opened either. The regional comparisons use the seasonally adjusted regional columns, and the table note states that “Details may not equal totals due to rounding.” Orderings as close as 4,676 against 6,067 could move on a different adjustment basis. The Puerto Rico application series was not examined, and the 531,728 total excludes applications outside the 50 states and DC by definition.

Finally, a comparison reaching back before the 2012 boundary of this high-propensity definition would be comparing two screens, and the branch counts here would not transfer.