Two Frames, Two Ways of Seeing the Same Business
The BDM lifecycle shows how a local unit can open, expand, contract, or close without telling us by itself whether an independent firm was born. This chapter turns to the institutional question behind that distinction: what records allow the BLS and Census Bureau to observe those units, and how do they connect a location to a parent firm? Understanding the two paths gives readers a way to diagnose a count difference as either a coverage problem, an organizational-linkage problem, or both.
The comparison is not between a database that contains facts and another that contains opinions. Both are statistical frames built from administrative reporting and longitudinal links. A frame determines who can enter the business universe, which attributes are attached to a record, and how several locations are associated with an organization. Those decisions are part of the measurement, not invisible plumbing behind it.
The BLS path: from unemployment-insurance records to establishment flows
The BLS Business Employment Dynamics program derives its data from the Quarterly Census of Employment and Wages. The QCEW is built from summaries supplied through state and federal unemployment-insurance systems. Those records provide employment and wage information and identify an establishment’s location and industry. In practical terms, the path begins with reporting relationships created for unemployment-insurance administration and turns them into a statistical view of business units.
That origin matters for coverage. When a private-sector business hires employees in a covered industry in a state, it reports information used for unemployment-insurance taxes and records. State agencies provide the resulting employment and wage information to BLS. The BLS frame therefore has a direct connection to the employment relationships that state systems are designed to administer. It is not simply an estimate based on a sample of businesses volunteering to answer a business survey.
The record is not automatically a fully separated map of every physical work site. For an employer with one establishment in a state, the accounting record may provide enough information for statistical purposes. For an employer with multiple establishments or different activities in the same state, a statewide contribution report may not identify each work site separately. Multiple Worksite Reports help disaggregate that employer by listing employment and payroll at individual establishments. This is one reason the BLS path can represent an organization as a network of local units rather than only as one statewide employer record.
The distinction between an employer report and an establishment is important. A reporting system may know that an organization has workers and wages without automatically knowing how those workers are distributed across locations. A worksite report adds location-level detail, but the resulting establishment map still depends on the reporting rules and the quality of the linkage over time. The BDM then longitudinally links establishment and firm records so that it can compare employment across March observations and classify openings, expansions, contractions, and closings.
This creates several layers of observation that are easy to collapse in a headline. The state system begins with an employment and wage relationship. A statistical worksite record adds a place and an industry. A longitudinal link then determines how that place relates to the prior period and whether it is treated as a continuing or new unit under the annual definitions. Firm information adds another layer by grouping establishments under a common owner or employer identifier. Each layer answers a different question: whether employment exists, where it is reported, whether the location persists, and how it relates to other locations.
The layers also explain why more local detail is not automatically the same as more independent businesses. Multiple Worksite Reports can make an employer’s network visible as several establishments, which is valuable for local job-flow analysis. But the network may still belong to one firm. Conversely, if worksite detail is incomplete or organized differently, a collection of locations may appear less finely divided. The BLS path is therefore sensitive both to coverage of the employment relationship and to the way an organization supplies location-level information.
The BLS frame can also attach alternative characteristics to the same activity. Gross job flows are measured at establishments, while age and size information can be determined using establishment and firm information. This makes it possible to ask whether jobs are being added at young establishments, at locations belonging to old firms, or within particular establishment and firm size groups. It also means that a single headline can conceal which level of the BLS frame produced it.
The strength of this path is its connection to employment reporting and its ability to observe local operating units when the relevant worksite information is available. Its boundary is equally important: a QCEW-based frame is not a timeless, location-neutral inventory of all economic organizations. It reflects the scope of the underlying reporting systems, the establishment details supplied by employers, and the longitudinal links used to identify continuity.
The diagram places the two paths side by side without implying that their records are interchangeable. On the BLS side, state UI and QCEW records provide the base, while Multiple Worksite Reports can add establishment detail. On the Census side, IRS and supplementary records feed the Business Register and LBD. Both sides end with establishments and firm relationships, but the route to those objects is different.
The Census path: from the Business Register to firm relationships
The Census Bureau’s Business Dynamics Statistics is derived from the Longitudinal Business Database, which tracks establishments and firms through the Business Register. The Business Register relies primarily on Internal Revenue Service records and is supplemented by survey, census, and other administrative sources. Those inputs create a different business universe from the QCEW-based frame and provide a different foundation for identifying the relationship between a local establishment and its parent firm.
The Census path is designed to do more than record employment at a location. The Business Register supports a longitudinal record of establishments and firms, allowing the LBD to follow units over time and associate establishments with parent organizations. Organizational information is therefore central to the BDS view. It affects whether several locations are treated as part of one firm, how firm characteristics are assigned, and how establishment openings and closings are interpreted in relation to firm dynamics.
The distinction between the IRS-based and UI-based foundations does not mean that one system observes “real” businesses and the other observes administrative artifacts. Administrative records are the means by which both systems observe the economy. But administrative records are created for different purposes. A payroll-tax form, a state unemployment-insurance report, and a worksite questionnaire can all describe a business while organizing its locations and responsibilities differently. The statistical frame inherits some of those differences.
The Census path therefore has its own layers of observation. The Business Register supplies the business frame; the LBD links annual establishment records into a history; and the BDS publishes selected aggregates from that longitudinal structure. The firm relationship is not an afterthought added to a location count. It influences which establishments are grouped together and which firm characteristics are available for comparison. A table of establishments and a table of firms may be built from related records while answering different questions about organizational scale.
This also explains why a disagreement can survive even when both frames identify many of the same businesses. Agreement that a parent organization exists does not guarantee agreement about how many locations belong to it. One source may make a network of worksites more visible, while another may attach more activity to a smaller number of reporting units. The resulting difference is not necessarily a disagreement about whether the firm operates; it can be a disagreement about the structure of its establishment portfolio.
This is the point at which two kinds of disagreement must be separated. A coverage disagreement occurs when an employer or establishment appears in one source but not the other. If a set of small service employers enters the UI-derived frame but lies outside the Census employer universe, the two systems will report different totals even if both link their own records consistently. The question in that case is, “Who is in the frame?”
An organizational-linkage disagreement occurs when both sources contain the same underlying firm but assign it different numbers of establishments or different single-unit and multi-unit statuses. One frame may link several locations to a common parent; another may record fewer locations or attach them differently. The question in that case is, “How is a covered organization divided into units?” The underlying firm need not be missing for the establishment counts to diverge.
These mechanisms can coexist. A frame can omit a class of employers and also represent some common firms as more or fewer establishments. That is why a comparison of published totals cannot identify the cause by itself. The total is the symptom. To diagnose the mechanism, researchers need linked records that can distinguish firms present in both sources from firms present in only one and can compare their establishment structures.
A useful diagnostic follows from that distinction. If the difference is concentrated among firms found in only one source, the first question is coverage. If the same firms appear in both but one source assigns them more establishments, the first question is linkage and worksite organization. If both patterns appear, a single explanation will be misleading. This diagnostic does not decide which frame is preferable; it identifies the kind of evidence needed before making that judgment.
The linked analysis in the Census paper makes that distinction explicit by comparing the LBD with establishment-level QCEW data in the LEHD Employer Characteristics File. The purpose is not to declare one frame universally superior. It is to determine whether the divergence comes from different sets of firms, different establishment counts within common firms, or both. That is a stronger question than asking which published total appears more plausible.
The two frames also differ in the kind of organizational story they make easy to see. A system that records more location-level units can make an economy look more decentralized, with more small establishments distributed across communities. A system that attaches more employment to fewer establishments can make the same economy look more concentrated in large work sites or parent organizations. Those narratives may be partly about the underlying economy and partly about how the frame allocates activity across units.
No single criterion decides which frame should be used for every question. If the question concerns employment flows at local operating units, the BDM’s QCEW foundation and establishment-level measures are directly relevant. If the question concerns parent-firm structure and the distribution of establishments across firms, the LBD/BDS linkage is directly relevant. If the question is about changes in the national establishment total, the analyst must first ensure that the two series have comparable scope, reference dates, industry coverage, and unit definitions.
“Same business” also needs a careful meaning in a comparison like this. It can mean the same legal firm, the same establishment address, the same employment relationship, or the same economic activity. Those are related identities, but they are not interchangeable. A comparison that matches firms may still find different establishment counts. A comparison that matches locations may still find different parent-firm assignments. Before interpreting agreement or disagreement, the analyst needs to say which identity is being held constant.
This is why the public aggregate tables and the linked microdata serve different purposes. Aggregates show that the two frames produce different totals and distributions. Linked records can investigate whether the difference comes from exclusive coverage or from alternative organization among common firms. Neither level replaces the other: the aggregate establishes the scale of the puzzle, while the linkage supplies evidence about its mechanism.
The public descriptions establish the architecture of the disagreement, but not a complete audit of every inclusion decision. The linked microdata analysis is bounded by the years and private-sector scope studied in the paper, while the published aggregate series extend over different periods and may be presented with rounded totals. The frame comparison can therefore tell us why a headline deserves investigation; it cannot, by itself, supply one corrected national count.
The practical consequence is a better reading of the next question. When the BDM and BDS report different numbers of establishments, the first response should not be to average them or choose the larger number. It should be to ask whether the difference reflects who is covered, how common firms are partitioned, or a combination of the two. Only after that diagnosis can the aggregate gap be interpreted. The next comparison will show how large it is—and why more establishments in one frame can coexist with more employment in the other.