The First Conversion: From Filing to Paid Employment
This chapter follows the first observable conversion in the business-formation pipeline: how a filed plan becomes, or fails to become, a business with paid employees. The key question is not whether an applicant had a serious idea, but what the linked records can establish about employer entry within a defined time window. Understanding that boundary lets the reader interpret the evidence as a measured transition rather than as a promise hidden inside a form.
What the application actually says
A business application is an unusually early observation of an entrepreneurial decision. In the Census study, the population is built from applications to the Internal Revenue Service for Employer Identification Numbers, or EINs, that report an intent to start a new business. The analysis covers nearly 17 million such applications filed in the United States between April 2017 and December 2021. About 90 percent of all business applications in the period reported that new-business intent; applications made for banking purposes, purchasing an existing business, or changing organizational form were excluded because they face a different kind of uncertainty.
The form contains more than a name and an address. It records an intended start date, legal form, principal business activity, and, importantly for this study, the highest number of employees the applicant expects in the next 12 months. That number is a forecast of maximum early employment. It is not a hiring contract, a payroll record, or a guarantee that the applicant will proceed.
The administrative setting helps explain why the response is informative without being definitive. The Internal Revenue Service uses the expected employment figure mainly to help determine payroll-tax filing arrangements. The answer is not an enforceable commitment and has limited immediate financial consequences at the application stage. It can also be updated as the business develops. Applicants often complete the form before they confront binding requirements from lenders, landlords, investors, regulators, customers, or potential employees. The response can therefore provide a window into what the applicant believes at the outset, before later scrutiny has reshaped the plan.
That does not make the number a pure or perfectly candid measure of belief. A person may be uncertain about when hiring will occur, may interpret the 12-month horizon differently from another applicant, or may report a plan that changes after demand and financing become clearer. The useful claim is narrower: the form captures an expectation at a consistent early stage, and that expectation can be compared with a later employment record. It gives researchers an early signal, not privileged access to the applicant’s true business quality.
The outcome record adds the event the application cannot provide. The study links applications to the Census Bureau’s Longitudinal Business Database, a panel that records firm identifiers, location, industry, age, quarterly employment, and payroll. A linked application counts as employer entry when the associated business is observed with positive employment in the study’s forward-looking window. An employer business, in this definition, has at least one paid employee. The definition is deliberately concrete: the question is whether the plan generated paid employment, not whether the idea was sincere or whether the owner earned income.
This distinction also clarifies what a zero means. If an application reports no expected employees, it may describe a nonemployer business, a plan to hire later, or uncertainty about early staffing. If the linked data show no positive employment in the window, the employment outcome is recorded as zero. The study groups nonemployer formation with no employer formation because both produce no observed paid employment for the outcome being analyzed. That coding is appropriate for measuring job creation, but it does not imply that a nonemployer business is economically empty. A self-employed person can operate a viable activity without hiring anyone.
The application and outcome records thus answer different questions. The first asks, in effect, “What do you expect to happen in the first year?” The second asks, “Was positive employment observed within the comparison window?” Neither record alone can tell us the complete story of a business. Together, they permit a disciplined comparison between expectation and realization.
The form also preserves the difference between a plan’s existence and its execution. An intended start date says when the applicant hopes to begin; a planned first wage-payment date says when hiring is expected; a legal-form choice says something about the proposed organization. None of these fields is a payroll event. They are pieces of a plan assembled before the business has demonstrated that customers, financing, regulations, suppliers, and workers will line up as expected. The form is valuable precisely because it captures that pre-outcome state, when uncertainty has not yet been resolved.
This timing gives the expectation both its strength and its weakness. It is early enough to reveal what an applicant believed before the outcome was known, but early enough that the belief may rest on incomplete information. A later payroll filing can confirm that employment happened, but it cannot reconstruct what the applicant thought before hiring. Conversely, the application can reveal an expectation that later disappears, but it cannot tell us from the form alone whether the change came from a failed demand hypothesis, a financing problem, an administrative delay, or a deliberate decision to remain a nonemployer. The linked design identifies the mismatch; explaining its cause requires additional evidence.
It is also important not to read the expected number as a promise about the business’s eventual workforce. The question asks for the highest number of employees expected within the first year, so the reported value is a planned maximum over a horizon, not an average headcount and not a measure of long-run employment. A business that expects a temporary early team, a business that plans to hire gradually, and a business that expects no employees are making different forecasts. Comparing each forecast with the same kind of early maximum makes the comparison coherent, but it does not turn the measure into a general forecast of the firm’s future size.
Where “entry” is observed
The study aligns the applicant’s 12-month expectation with four quarters of observed employment, including the quarter of filing. Entry occurs if the linked business records positive employment in any of those quarters. Realized maximum employment is the highest quarterly employment observed during the same window. The design does not pretend that a quarterly measure captures every moment of a firm’s first year. It makes the horizon explicit so that an expectation about the next year is not compared with an outcome measured over an unrelated period.
Timing within the filing quarter creates a subtle problem. An application submitted near the end of a quarter has less calendar time before the next quarterly employment observation than one submitted near its beginning. If both were assigned exactly the same four-quarter window without adjustment, the later applicant would be given less practical time to become an employer. The researchers use the week of application to adjust the observation window and mitigate this within-quarter timing difference. The purpose is not to make the data perfect; it is to make the comparison fairer for applicants who enter the dataset at different points in a quarter.
A fixed horizon is therefore a comparison rule, not a claim about the only meaningful definition of entry. A business may hire just after the window closes, may delay hiring because it is still testing demand, or may employ people intermittently in a way that quarterly records do not capture as the applicant imagined it. The study’s result is strongest when stated in its own terms: positive employment was or was not observed during the matched period. Extending that result into a claim about permanent nonexistence or long-run failure would require later observations and, ideally, information about the business’s operating status and owner income.
The choice of a maximum also matters for interpretation. If a business records positive employment in one quarter and then reduces its workforce, the maximum captures that early peak rather than the path back down. If it starts with no employees and hires later, the maximum records the later peak if it falls within the horizon. The measure is designed to align with the applicant’s maximum-employment question. It is not intended to describe stability, job quality, total labor demand, or the distribution of employment across the year. Those questions belong to different outcome measures.
The linked record does not eliminate selection; it makes selection observable enough to study. The applications are not a random sample of all people who might consider entrepreneurship. Filing requires some intention, information, and willingness to take an administrative step. Among those who file, applicants who report positive employment expectations are not necessarily comparable with those who report zero. Some may have more concrete plans, while others may be exploring a nonemployer activity or postponing a hiring decision. The comparison is valuable because the study can see both groups, but the groups’ differences must not be treated as caused by the reported number alone.
The state transitions are easier to understand than the administrative vocabulary suggests:
The diagram separates expectation from state change. “Positive expected employment” is a statement made at filing. “Employer entry within the matched horizon” is a later observation. The path from zero expectation to employer entry is open because a person can hire despite reporting no expected employees. Conversely, a positive hiring plan can end at no employer entry within the window. The outcome categories are therefore not simply high and low confidence; they are combinations of a forecast and a realized state.
The one-year horizon also places a boundary around the word “failure.” An application that has not generated paid employment within the window has not met this study’s definition of employer entry for that period. It may become an employer later, may have formed a nonemployer business, or may never operate. The data support the first statement directly and do not, without a longer follow-up, distinguish all of the latter possibilities. Treating a delayed transition as permanent failure would extend the evidence beyond its observation rule.
The same caution applies to scale. The realized measure is the maximum quarterly employment recorded during the window, not a complete account of all work performed, revenue earned, contractors engaged, or jobs created outside the linked employer record. It is well suited to the study’s question about early paid employment. It is not a universal measure of business value. A business that remains a one-person operation and a business that never gets off the ground can share a zero employment outcome even though their economic circumstances differ.
The linked design is nevertheless powerful because it connects an early forecast to a later, administratively recorded state at unusually large scale. Survey questions about entrepreneurial intention often cannot be followed to a precise employment outcome. Payroll records alone usually do not reveal what founders expected before they entered. Linking the two makes it possible to ask whether expectations predict entry, how much they overshoot or undershoot, and whether the errors differ across kinds of applicants and places.
It also prevents a common statistical shortcut. Counting applications estimates the volume of people who crossed an administrative threshold. Counting employer businesses estimates the volume that crossed a later employment threshold. The difference between those populations is not merely noise to be removed; it is part of the phenomenon. It may reflect uncertainty, financing, demand, regulation, timing, nonemployer intent, or a change of mind. The data establish that the thresholds are different. Explaining why they differ requires the next layer of evidence.
For now, the essential conclusion is methodological and substantive at once: filing is the beginning of an observation, not the observation of an employer. The application supplies a time-stamped expectation, while the linked employment record supplies a later test of whether paid employment appeared within a comparable horizon. Once those two states are kept apart, the next question becomes measurable: how large is the gap between the hiring plan and the employment actually reached, both for applications that cross the entry threshold and for those that do not?