A friend of mine spent four months last year applying to a posting that had been live since the previous September. Same listing, same requisition number, reposted every thirty days. He was interviewed twice. In March somebody in HR told him, in the tone of a person doing a favour, that the role was on hold but the listing would stay up because taking it down and putting it back created paperwork.
That listing is in the vacancy statistics. It has been in the vacancy statistics for fourteen months.
I have spent a decade reading labour market data for a living and I have watched the job-openings series become the single most cited number in American economic commentary — invoked to prove that workers have leverage, that the labour market is tight, that anyone unemployed is unemployed by choice. It is a good series. It is not measuring what it is being used to measure, and the gap has widened enough that I no longer think the number can bear the weight put on it.
Start with the definitional problem. A vacancy, as collected, is a position the employer says is open, could start within thirty days, and is actively recruiting for. Every one of those conditions is self-reported and none is verified. The listing my friend applied to satisfied all three on paper for fourteen months while satisfying none of them in fact.
Then consider what changed to make that more common. Posting a job used to cost money — a newspaper advertisement, a recruiter's fee — and cost imposed discipline. Posting is now free and nearly effortless, and the incentives around it have inverted. Firms post to build candidate pipelines for roles they might open next quarter. They post to benchmark salary expectations. They post because an applicant tracking system is configured to keep requisitions open until formally closed by a manager who has moved to a different team. They post to look like they are growing. None of these are conspiracies; they are ordinary institutional behaviour with no counterweight, because nothing bad happens to a company that leaves a listing up.
The counterargument I take most seriously is that this has always been true to some degree, that the series is a consistent measure of a consistently noisy thing, and that consistency is what matters for tracking change over time. If the phantom share is stable, the trend still tells you something even if the level is inflated.
That would be persuasive if the phantom share were stable. I do not think it is, and here is the specific evidence that moved me: the relationship between vacancies and hires has come apart. For decades, the ratio of hires to openings sat in a fairly narrow band — more openings meant proportionally more hiring, because openings were roughly a real thing. Since about 2021 that ratio has fallen and stayed fallen. There are far more openings per hire than there used to be. Either American firms have become dramatically worse at converting an opening into a hired person, or a rising share of the openings were never conversions waiting to happen.
Both stories are consistent with the ratio. But the first requires believing that recruiting got substantially harder in a way that has persisted for years across every sector simultaneously, including sectors with abundant applicants. The second only requires believing that free posting plus no penalty produces more posting. I know which of those I would bet on.
A reader who does this work for a living wrote to say that both can be true at once and that I had set it up as a choice when it is a decomposition — some of the gap is genuine matching friction and some is residue, and the interesting question is the ratio. That is a better framing than mine and I have not been able to improve on it.
The consequence is not academic. This number sets interest rates. A central bank looking at high vacancies concludes that labour demand is strong and that wage pressure is coming, and tightens accordingly. If a meaningful share of those vacancies are administrative residue, the tightening is aimed at demand that does not exist, and it lands on the people who lose jobs when rates rise — who are, reliably, the same people the "workers have leverage" story was told about.
I want to be honest about the limits of my own claim, because I have watched this argument get overextended. I am not saying the labour market was weak. It plainly was not; quits were high, wages moved, people did have leverage, and the vacancy series was picking up something real. I am saying the series has drifted from the thing it is a proxy for, that the drift is measurable in the hires-to-openings ratio, and that the confidence with which the raw number gets quoted has not adjusted at all.
There is a fix, and it is boring. Add one item to the existing survey — how many of your open postings resulted in a hire in the last ninety days — and publish the conversion rate alongside the level, in the same release, on the same day. Not a new collection and not a new agency. One item on a form that already goes out. It would cost almost nothing, it would make the phantom share visible, and it would let anyone using the series see immediately whether they were looking at demand or at paperwork.
My friend eventually took a job at a smaller firm that had not posted the role publicly at all. He found it because someone he had worked with in 2019 called him. That transaction appears in the hires number and never appeared in the openings number, which is the other half of the same problem and a harder one to fix.
