How to Read the Path from Intention to Employment

The final question is not whether entrepreneurship is real when no employees have yet been hired. It is how to interpret the different signals that appear between an intention to start, an application, employer entry, first-year jobs, and continued operation. The answer is a disciplined one: a business application records a plan and a forecast; it becomes evidence of an employer only when a later record shows positive paid employment, and even then the scale and durability of that employment remain separate questions.

That answer preserves two truths that are easy to force apart. Early expectations contain information about what may happen, so discarding them throws away a potentially useful signal. But those expectations are selected, context-dependent, and imperfectly calibrated, so treating them as employers or promised headcount turns a probability into a fact. The evidence is most useful when each signal is kept attached to the stage, population, horizon, and outcome it actually measures.

The correct unit is a transition

The application-to-employer path is best understood as a sequence of thresholds rather than a single event. A person may move into business activity. They may file an application that states an intent to start a new business and reports expected employment. They may or may not become an employer within the relevant period. If they do, the business may reach a particular first-year employment scale, create jobs at the population level, and remain active for a year. Each threshold filters the cases that reach the next one.

The first threshold is administrative and cognitive: someone takes the step of applying and reports a plan. The next is operational: the linked employment record shows at least one paid employee. That distinction is the precise answer to the title’s question. A business application is not yet an employer because the application describes what an applicant expects to do; employer status describes what a later record shows has happened. The form can be an important early observation without being the outcome it anticipates.

The Census study makes this transition visible by linking nearly 17 million applications to start new businesses, filed from April 2017 through December 2021, to subsequent employment observations. The study’s employer definition is deliberately narrow and testable: positive paid employment within a matched one-year window. It treats an application that becomes a nonemployer business and an application that produces no observed employment alike for the employment-generation outcome. That choice is appropriate for asking whether paid employment appeared, but it does not say that nonemployer activity had no economic value.

Once the entry threshold is recognized, the application data have to be read on two margins. The extensive margin asks whether employer entry occurs at all. The intensive margin asks how much employment follows, conditional on entry. The distinction matters because an application can fail at the first margin while being overconfident about entry, or cross the first margin while underestimating the scale it will reach. A single count of “startups” cannot preserve both pieces of information.

The four expectation-and-outcome groups provide a compact test of the first margin. In the linked sample, 84.1 percent of applications neither expected nor achieved employer entry within the study window. Another 8.9 percent expected entry but did not realize it, while 3.6 percent realized entry despite reporting no positive expectation. Only 3.4 percent both expected and achieved entry. The shares are not a forecast for every future application; they are the observed composition of this defined population and horizon. Their value is to show that a large application population contains a large number of cases that do not cross the employer threshold during the measured period.

The off-diagonal groups also reveal direction. Expected entry exceeds unexpected entry by 5.3 percentage points, calculated as 8.9 percent minus 3.6 percent. If the concern is any mismatch rather than its direction, the two groups add to 12.5 percent. Those statistics answer different questions: one describes net overprediction of entry, the other describes the frequency of wrong classification. Neither permits the stronger claim that every application with no observed employment was permanently abandoned, because delayed entry and nonemployer operation remain possible outside the window.

The intensive margin produces a second correction. Across all applications, expected maximum employment averages 0.530 workers and realized maximum employment 0.388, an overall overprediction of 0.142. But among the small group that both expected and achieved entry, expected employment averages 4.779 while realized employment averages 6.683, an underprediction of 1.903.