Business name search data counts filings, not businesses: where the counts drift

Registry data invites counting. It is tabular, it has dates, it covers a whole state, and it is free. So people count it: new businesses this quarter, companies in an industry, the most popular words in company names, how long a business lasts.

Nearly all of those numbers are measuring something other than what they are labeled. Not because the data is bad, but because the unit of a register is a filing, and the unit people want is a business. Those two things drift apart in ways that are systematic rather than random, which means bigger samples do not fix it.

What to take away

Why registrations are not businesses

One business, several entities. Holding structures, per-property or per-project entities, and separate entities for different lines of work are all ordinary. Counting entities counts the structure, not the business.

Entities with no operations. An entity can be formed and never trade. Nothing on the register distinguishes it from one that opened the next day.

Dormant entities that stay on the list. An entity that stopped trading years ago may remain in good standing as long as somebody keeps filing, and may sit as delinquent for a long time before any administrative action.

Foreign qualifications. A company registered in its home state and qualified in three others appears on four registers. Any national total assembled by adding state registers counts it four times.

Bulk formation. Formation services file large numbers of entities, and some jurisdictions attract registrations from businesses that operate nowhere near them. A register reflects where people choose to file, which is not where work happens, and the two are separate steps in the federal guide to starting a business for exactly that reason. The same filing-versus-business gap shapes any business name search by year, where a name's history belongs to filings rather than to the company.

The metrics that break, and how

The last row deserves a warning of its own. Where a register captures an industry or purpose field at all, it is usually chosen by the filer at formation and never revisited. Treating it as a current classification of what the company does is an easy mistake to make and a hard one to detect afterwards.

Entity type is a slightly better field and still not what people want from it: the categories are legal forms rather than lines of business, as the IRS overview of business structures makes plain.

Lag has a direction

If the errors were random, larger datasets would help. They are not.

Formation filings arrive before the business does anything, so a formation series leads real activity. Dissolutions arrive long after the business stopped, or never arrive at all, because a company that has quietly ceased trading has no reason to file. Administrative dissolutions arrive on the office's own schedule, in batches, when a deadline passes rather than when anything happened.

Reinstatement makes it worse, because it can reopen a record that a previous extract counted as closed. Any panel you build from repeated snapshots will show entities coming back from the dead, and the reason is procedural rather than commercial. The vocabulary behind all of these transitions is set out in registration status.

What name frequency actually measures

Counting words across a register's names feels like measuring the economy. It mostly measures three other things.

The trade vocabulary of the period, because most business names describe what the business does. The geography, because place names are the second most common ingredient. And the office's own rules, because a distinguishability test that permits near neighbors produces clusters of nearly identical names that a stricter test would not.

Comparing name patterns between states therefore compares rulebooks as much as markets. Comparing them across decades compares filing conventions, since suffix formatting and permitted characters have changed over time.

There is also a survivorship effect built in: registers hold the names of entities that failed alongside the ones that thrived, with nothing marking the difference except a status field that means something narrower than it appears.

If you are going to analyze it anyway

The honest version of the output

You can say: this many entity records were on this register, in this condition, on this date. You can say: this many formation documents were processed in this window. Those are real and occasionally useful.

A register cannot show how many businesses exist, how many started, how many closed, or what they do. Statistical agencies collect that data using sampling frames built for the purpose.

Those agencies exist because administrative records do not answer these questions. If a number matters, take it from a body that publishes it as a statistic, not from filings.

Common questions

Can I at least use registry counts as a trend, if not as a level?

Cautiously, within one state, over a period where nothing about the office's process changed. Fee changes, deadline changes, and enforcement drives all move filing volumes without moving the economy, which is exactly the kind of change described in secretary of state search data.

Two commercial datasets give different counts for the same state. Which is right?

Probably neither, and the difference is likely to be definitional: which statuses were included, whether foreign qualifications were counted, and when each extract was taken. Ask both for their definitions before you ask which is accurate.

Is entity data useful for anything quantitative?

Yes, for questions about the register itself: how many of your own vendors are currently in good standing, which of your entities have a filing due, where your portfolio is registered. Operational questions about a known set of entities are what this data answers well.