US kindergarten vaccination 2026: exemptions are the real signal
US kindergarten vaccination coverage fell 0.1 points. Nonmedical exemptions jumped 0.8. One is a lagging metric, the other is a leading one — and the CDC published no report.
US kindergarten vaccination rates for the 2025–2026 school year fell again, and almost everyone reading the headline will focus on the wrong number. MMR coverage dropped from 92.5% to 92.4%. That is a rounding error. The number that actually moved is the nonmedical exemption rate, which went from 3.2% to 4.0%.
Ars Technica's Beth Mole reported the release on 17 August 2026. I want to talk about why the small number and the big number tell completely different stories, and what that has to do with how you read any metric.
📊 What the numbers actually say
Here is the whole release, compressed:
| Metric | 2024–2025 | 2025–2026 | Change |
|---|---|---|---|
| MMR coverage | 92.5% | 92.4% | −0.1 pt |
| DTaP coverage | 92.1% | 92.0% | −0.1 pt |
| Nonmedical exemptions | 3.2% | 4.0% | +0.8 pt |
| Medical exemptions | 0.2% | 0.2% | flat |
Two more figures from the CDC data worth holding on to:
- 10 states are at or above the 95% MMR threshold: West Virginia, Connecticut, Maine, New York, Mississippi, Rhode Island, Maryland, Massachusetts, California, New Mexico.
- 24 states now have exemption rates at or above 5%, up from 14 states in 2023–2024.
More than half of all states saw coverage decline across MMR, DTaP, polio and varicella. Coverage has been sliding since 2019–2020, when the US was around 95%.
🧮 The 5% ceiling is a constraint, not a target
This is the part I find genuinely interesting as an engineering problem, because it is arithmetic rather than politics.
Exempt children are opted out by their parents. No clinic, no outreach van, no SMS reminder campaign converts them. So the best coverage any location can reach is bounded:
max_achievable_coverage = 100% − total_exemption_rate
Run the national numbers: 4.0% nonmedical plus 0.2% medical gives 4.2% total exemptions, so the ceiling is 95.8%. That is my arithmetic on their figures, not a CDC statistic, but it is the same logic behind what the CDC calls "potentially achievable" coverage. The US is at 92.4% against a ceiling of 95.8%. There are 3.4 points of headroom left, and it shrinks every year.
Cross 5% exemptions and the 95% target becomes mathematically unreachable. Not hard. Not expensive. Unreachable.
Key takeaway: 24 states have already crossed the point where perfect public-health execution still fails the target. You cannot outwork a broken denominator.
If you run services, you have seen this shape before. It is an error budget. Once you have burned it, no amount of heroic on-call work buys the quarter back — the only remaining lever is the thing that generated the burn. Exemption policy is that lever, and it is the one nobody is pulling.
⏱️ Coverage is lagging, exemptions are leading
A 0.1-point coverage drop looks like noise, and if you tracked only coverage you would reasonably report "roughly flat." But the two metrics measure different moments in time.
| Coverage rate | Exemption rate | |
|---|---|---|
| Measures | outcome | intent |
| Moves when | children go unvaccinated | parents opt out |
| Reversible by outreach | often | almost never |
| Signal type | lagging | leading |
Coverage is the state of the system today. Exemptions are a commitment about the next several years, because an exemption filed in kindergarten does not usually get withdrawn in grade three. A 0.8-point jump on a 3.2 base is a 25% relative increase in permanent opt-outs in one school year. That is the number a monitoring system should page on, and it is the one buried three paragraphs down.
The general lesson, worth stealing for whatever you are building:
- Rank your metrics by how reversible the thing they measure is.
- Alert on the irreversible ones, even when they are small in absolute terms.
- Never report a percentage-point change without the relative change beside it. 3.2 → 4.0 sounds like nothing. +25% does not.
📄 Raw data is not a publication
The other detail in the story matters more to us than to most readers. In previous years the CDC published its analysis in the Morbidity and Mortality Weekly Report, peer-reviewed inside the agency. This year, as with last year, it posted the numbers online with no full analysis and no vetted publication.
A dataset without an analysis layer has not been published. It has been leaked into the open, with the interpretive work quietly transferred to whoever downloads it next.
For anyone who consumes public data — and if you build tools, you do — this changes your risk profile in three concrete ways:
- No provenance narrative. You get the values but not the caveats: sampling changes, which states reported, how "up to date" was defined this year.
- No stable citation. An MMWR article has a DOI and never moves. A page of numbers on a website can be edited or removed with no diff and no notice.
- You are now the analyst. Any conclusion you draw is yours to defend, without an authoritative source to point at.
The practical response is boring and effective: snapshot every upstream public dataset you depend on, with the retrieval date, and diff it on each refresh. Every calculator on this site carries a LAST_VERIFIED constant for exactly this reason. If the source moves, we find out from the diff, not from a reader.
🇱🇰 What a Sri Lankan reader should take from this
I am not going to quote a Sri Lankan immunisation coverage figure here, because the source above does not contain one and I will not invent one. What I will say is that the failure mode in this story is worth understanding before it becomes locally relevant.
The classic coverage problem in a developing health system is access: distance, cost, stock-outs, a missed clinic day. Those are solvable with logistics, and logistics is a thing governments and NGOs are reasonably good at. The problem in the US data is refusal, which no amount of logistics touches. A country can therefore have excellent delivery and still lose coverage, and the delivery metrics will look fine right up until they do not.
If you have a child in the system and want to check where you stand against the national schedule, we maintain a Sri Lanka child vaccination schedule calculator with the sources cited in the page. It is free, needs no signup, and runs in your browser.
💡 What this means for you
- Read the metric that moved, not the metric in the headline. MMR fell 0.1 points. Exemptions rose 25% relative. Only one of those is news.
- Find the hard ceiling in your own numbers. Every system has a quantity that caps the best achievable outcome regardless of effort. Know yours before you spend a quarter working below it.
- Alert on irreversibility, not magnitude. Small changes to things that do not revert are more urgent than large changes to things that do.
- Archive your upstream sources. When the analysis layer disappears, the raw file you saved last year becomes the only record of what the number used to be.
The CDC still publishes the data. It just stopped doing the reading for you. That is a job that does not vanish when an agency drops it, it only moves downstream.