ISSUE 012
What Survival Rates Actually Show
Everybody has heard that ninety percent of businesses fail. Nobody I have asked can say by when, or against what. The Bureau of Labor Statistics has been following every cohort of new American establishments since 1994 and publishes the survival curve by state, Missouri included, which means the claim is checkable. It took an afternoon. The answer is that the number is wrong at the horizon people mean it, roughly right at a horizon nobody means it, and less interesting than what sits underneath it — which is that restaurants outlast software companies by a distance.
An idea worth thinking about
A death, in these statistics, means the payroll went to zero
Before any survival number means anything you have to know what counts as dying. The Census Bureau's Business Dynamics Statistics methodology page gives the definition in one line: an establishment exit is an establishment with zero employment in the current year and positive employment in the prior year. Entry is the mirror image. Everything is measured in the payroll period that includes 12 March, so every rate is a March-to-March change.
That is a measurement rule, not a judgement. An owner who sells up and retires comfortably at sixty-eight produces an exit if the buyer sets up under a new employer identification number. An owner who lays off their last employee and carries on alone, profitably, as a one-person business produces an exit. A firm that moves across a state line produces an exit in one state and an entry in the other. Census does say most closings are true exits, meaning operations ceased at that location — but true exit and failure are not the same word either.
So the survival curve is a count of continuing payrolls, not a count of people whose businesses worked out. Both numbers are worth having. They are not the same number, and the gap between them runs in one direction only: the failure rate implied by these tables is an overstatement of how many people got a bad outcome.
One caveat Census puts on its own page and I will pass on: it says the methodology documentation is no longer current, because the Bureau is evaluating alternatives to the noise infusion it uses to protect confidentiality. The definitions above are the ones the published data rests on, but the page is flagged by its own publisher as out of date.
A tool I disappeared into
BLS will build you a Missouri series in about four clicks
The one-screen data search for Business Employment Dynamics lets you pick a state, an industry, an ownership class and a measure, and get a quarterly series back as a table or a chart. No account, no download, no spreadsheet. If you want to know how many private-sector establishments opened in Missouri in a given quarter, or how many jobs were lost at closing establishments, it is about four clicks.
The limit is important and the tool does not announce it. What you get here is the quarterly flow data — gross job gains and losses, openings and closings, expansions and contractions. The survival tables, the ones that follow a single year's cohort of new establishments forward for thirty years, are published separately as static files and are not in this tool. Every survival figure in this entry comes from those files, not from here.
A second limit worth carrying: BED counts establishments with employees. A business with no payroll is invisible to it, which, as this pillar's fifth entry found, is most American businesses. When you read a survival rate you are reading about the minority of businesses that ever hired anybody.
What it is genuinely good for is sanity-checking a mood. If you have the feeling that everything in your county is closing, pull the Missouri openings and closings series for the last few years and see whether the feeling is in the data. Usually it is not; usually both lines are large and close together, which is what churn looks like from the inside.
Learn something
How to check a survival claim yourself in about ten minutes
The Business Dynamics Statistics time-series tables go back to 1978 and are published as plain CSV files, free, no account. There is a file by state, a file by state and firm age, files by sector, by firm size, by county. The state file is a few hundred kilobytes and opens in anything.
The skill worth acquiring is small and it transfers. Open the state file, filter to Missouri — state code 29 — and you have one row per year with establishment counts, entries, exits, entry and exit rates, job creation and destruction, and firm deaths. Next time somebody quotes you a failure statistic, you can see in thirty seconds whether the stock of businesses in your state is actually falling, which is the thing the statistic is usually being used to imply.
What the file cannot tell you is why, and this is the honest ceiling on the whole exercise. There are no owners in it. No prior industry experience, no capitalisation, no education, no sector switching. The columns are counts and rates. Anyone who tells you what predicts survival is importing that from somewhere else, and the somewhere else is usually a survey of people who were still around to be surveyed.
Two practical notes. The 2023 vintage is the current one, so you are looking at data roughly two years behind the present. And the firm-age file is the one to use if you want survival-flavoured numbers from Census rather than BLS, because it lets you see how many Missouri firms sit in each age band — which is a different question from how many of a given cohort lasted, and worth not confusing.
What we’re seeing
13,188 Missouri establishments closed in 2023. 13,817 opened.
Those two numbers are one row in the Census Bureau's state file for 2023. Missouri: 102,345 firms, 138,037 establishments, 2.63 million employees. Exits 13,188, an exit rate of 9.576 percent. Entries 13,817, an entry rate of 10.032 percent. Firm deaths — whole firms, not single locations — 9,796, taking 46,508 jobs with them.
Run the establishment count across the period and the stock barely moves: 139,141 in 2019, 138,185, 138,942, 139,770, and 138,037 in 2023. Roughly one in ten Missouri establishments closes every year and roughly one in ten opens, and the total sits almost still. That is churn, not collapse, and it is the first thing the ninety percent claim obscures.
Now the cohort view, which is the one the claim is actually about. BLS has followed every annual cohort of new private-sector establishments since 1994 and publishes a Missouri table. Of the Missouri establishments that opened in the year to March 2016, 85.1 percent were still going a year later, 44.8 percent at five years and 28.6 percent at ten. The 2019 cohort: 74.8 percent at one year, 43.2 percent at five. The 2021 cohort: 72.7 percent and 39.9 percent. Nationally the same table gives 79.6 percent, 50.2 percent and 34.6 percent for the 2015 cohort. Missouri's recent five-year rates run several points below the national figure and I cannot tell you why — these tables carry no owner, industry-mix or firm-size breakdown at state level, so any explanation would be me inventing one.
Against that, ninety percent. About sixty-five percent of American establishments are gone by year ten, not ninety. To reach ninety you have to wait roughly thirty years: of the establishments that opened in the year to March 1994, 12.0 percent nationally and 13.0 percent in Missouri were still operating at March 2026. So the claim is wrong at the horizon people mean it — the first few years, when they are deciding whether to start — and approximately right at a horizon that describes a career rather than a venture. That comparison is my reading of the BLS cohort rows, not a statement BLS makes. I have quoted their figures rather than linking them because the BLS site refuses automated requests and this archive only links what it can verify.
The finding I did not expect is in the industry tables. Accommodation and food services — restaurants, the trade everyone is sure fails — had 88.2 percent of its 2015 cohort surviving at one year, 58.5 percent at five and 41.6 percent at ten, comfortably above the all-industry average. Information, the sector that contains software companies, managed 80.8 percent, 46.8 percent and 29.9 percent. At the thirty-two-year mark the 1994 food-service cohort is at 14.1 percent and the 1994 information cohort at 8.0 percent. Restaurants outlast software companies, and have done for three decades. Whatever people mean when they say restaurants always fail, the establishment data says the opposite.
The useful version for a business in Cape Girardeau or Kirksville: your odds at five years are a bit under even, your odds at ten are around three in ten, and your trade matters more than the general statistic does. Go and find the survival number for your own industry before you accept a number for all industries. And if you are already past year five, the curve flattens — the annual loss rate after year six is far lower than it is in years one to three, which means the thing you survived was the hard part.
One question
How old is your business, in years?
Just the number. I want to know where this list actually sits on the curve.
More in this pillar as it's written.
— Shana
Sources link to the original publisher, never to a summary.