A single statistic is circulating in public health circles this week: as of September 2, 2026, the CDC estimates that COVID-19 infections are growing or likely growing in 47 states, declining in 0, and not changing in 3. Forty-seven states. That's the kind of number that gets screenshotted, shared, and stripped of every qualifier the CDC carefully attached to it.
I covered the Rt statistic back in July, when the number was 37 states. The jump to 47 is real. But the way this number is being used — as a proxy for "COVID is bad right now" — is the same methodological sleight of hand it was then, just with a bigger numerator. The denominator problem hasn't changed. What has changed is that we now have enough context to see exactly where the framing breaks down, and why fixing it matters more than the headline figure.
What Rt Actually Measures — and What It Explicitly Does Not
Start with the definition, because the CDC is unusually clear about it and most coverage ignores the clarity entirely.
The CDC's Rt page states this directly: "Epidemic trends indicate direction only and do not reflect the burden of disease, therefore they should be used alongside other surveillance metrics (such as the percentage of ED visits) for a more complete picture."
Rt — the time-varying reproductive number — tells you whether the number of infections is currently growing (Rt > 1), shrinking (Rt < 1), or holding steady. It is a directional signal. It is not a severity signal. It is not a hospitalization signal. It is not a "how sick are people getting" signal. The CDC built this caveat into the page's architecture, placing it in a dedicated "What Rt can and cannot tell us" section. That section exists because the agency has watched this number get misread enough times to know it needs a warning label.
The Rt estimate itself is derived from daily incident emergency department visits reported through the National Syndromic Surveillance Program — not from confirmed case counts, not from hospitalizations, not from deaths. ED visit data has its own selection biases: it captures people sick enough to seek emergency care, which in a population with high prior immunity and available antivirals is a different denominator than it was in 2020 or 2021. A growing Rt from this base means more people are showing up to EDs with COVID symptoms relative to recent weeks. It does not mean the same thing it meant when the baseline population had no immunity.
This is the comparison problem wearing an epidemic alert.
The Burden Question the Headline Number Skips
Here's the arithmetic that should accompany every Rt report but rarely does.
The CDC's respiratory illness data dashboard shows that nationally, respiratory illness activity is currently "very low." The same agency publishing a 47-state growth signal is simultaneously reporting that the overall burden of respiratory illness sits at its lowest categorical level. Both numbers are true. They are measuring different things. The 47-state figure measures trend direction; the "very low" designation measures burden magnitude.
A disease can be growing from a very low baseline and still represent very low burden. If ten people had COVID last week and fifteen have it this week, that's 50% growth over one week and Rt well above 1. It's also fifteen people. The Rt number tells you the first fact. It tells you nothing about the second.
The CDC's own summer outlook notes that regions which did not experience substantial COVID activity during the most recent winter — specifically the South and West — are expected to see increases in summer months. This is a modeling prediction about seasonal patterns, not an alarm about a novel threat. Summer waves in the South and West have been a documented feature of post-2022 COVID epidemiology, driven by indoor crowding during heat events and regional immunity gaps from lighter winter exposure. A predicted, seasonal, regionally-concentrated uptick showing up in the Rt data is not the same story as an unexpected national surge.
The 47-state figure flattens all of that. It makes a seasonal signal in Florida look identical to an unexpected outbreak in Minnesota. The map is doing work the number cannot support.
The Variant Caveat Nobody Is Quantifying
The CDC's respiratory data page includes a specific hedge that deserves more attention than it's getting: a larger summer increase remains possible "particularly if a variant that the immune system no longer recognizes becomes more common." That conditional is load-bearing. The current growth signal is consistent with ordinary seasonal dynamics. It would look different — and the Rt number would not tell you which scenario you're in — if a substantially immune-evasive variant were driving it.
This is not a hypothetical concern unique to COVID. The same interpretive problem appears across epidemic surveillance systems. The WHO's One Health framework, which convened its expert panel in Lisbon this month, exists precisely because single-pathogen metrics routinely miss the biological drivers that determine whether a trend signal is a blip or a turning point. Variant surveillance is the biological-driver layer that Rt, by design, cannot supply.
The Rt would be higher, probably, if a novel immune-evasive variant were driving growth — but the number itself doesn't carry variant information. You need genomic surveillance data for that, which the CDC tracks separately. This is a case where three different data streams need to be read together: Rt (trend direction), ED visit percentages (burden magnitude), and variant prevalence (biological driver). The 47-state headline is one-third of the picture. Reporting it alone is like publishing a stock's daily price movement without its price, volume, or sector context.
The forensic question here is: who benefits from the simplified version? The answer is everyone with a stake in a clean, alarming number — and nobody who needs to make an actual decision about behavior, policy, or resource allocation.
What the Statistical System Around This Number Is Actually Doing Well
I want to be precise about what I'm criticizing, because the CDC's Rt infrastructure is, methodologically, genuinely good work.
The agency is transparent about its source data (NSSP ED visits), its estimation method (a time-varying reproductive number model, documented in a public "Behind the Model" page), and its uncertainty (the categories are probabilistic — "growing or likely growing" is not the same as "growing," and the distinction matters). The CDC publishes historical estimates on data.cdc.gov. It maintains separate seasonal tracking for influenza and RSV, which ended for the 2025-2026 season as of June 5, 2026, and will resume for the next season — a transparency about data gaps that most surveillance systems don't bother with.
The problem is not the data. The problem is the translation layer between the data and public understanding. And that translation problem is not unique to epidemic surveillance. The Peterson Institute for International Economics has documented a related failure mode in economic statistics: when political pressure or media simplification strips technical caveats from government data, the number that circulates bears only a family resemblance to what the agency actually measured. The mechanism is different — PIIE is writing about Census Bureau integrity under political interference — but the underlying dynamic is identical. A carefully constructed statistical product, built with documented methodology and explicit uncertainty ranges, gets reduced to a headline figure that the original researchers would not recognize as their finding.
A number like "47 states" is designed — by the human brain, not by the CDC — to feel like a severity claim. It has the grammatical structure of a severity claim. "Growing in 47 states" sounds like "bad in 47 states." The CDC's careful hedging lives three paragraphs into a government webpage that most people sharing the statistic have never visited.
The Comparison That Would Actually Help
If you want to know whether the current COVID situation warrants concern, here is the comparison the Rt number cannot give you but that the broader data ecosystem can.
First: compare current ED visit percentages for COVID to the same week in prior years. The CDC's respiratory data page displays this. If the percentage of ED visits attributable to COVID is at or below historical summer baselines for this point in the calendar, a growing Rt is consistent with normal seasonal dynamics. If it's above, that's a signal worth escalating.
Second: compare hospitalization rates, not case counts. Hospitalization data captures severity in a way that ED visit-derived Rt does not. A growing Rt with flat or declining hospitalizations tells a very different story than a growing Rt with rising hospitalizations. The BLS's archived news release infrastructure offers a useful model here: every major economic indicator release is accompanied by confidence intervals, revision flags, and explicit notes on what the number does and does not measure. The employment situation report doesn't just say "jobs up" — it breaks out the household survey from the establishment survey, flags statistical significance thresholds, and notes where the 90% confidence interval includes zero. The Census Bureau's economic indicators page does the same, explicitly marking estimates where "the Census Bureau does not have sufficient statistical evidence to conclude that the actual change is different from zero." Epidemic surveillance could learn something from that disclosure architecture.
Third: watch the variant surveillance data. If a novel immune-evasive variant is gaining share in genomic sequencing, the growth signal has a different interpretation than if the circulating variants are familiar ones the immune system has seen before.
None of these comparisons are exotic. They're all available from CDC data. They just require clicking past the headline number, which is apparently the hardest thing in public health communication.
The Number to Watch Is Not 47
Here's where I land, after working through the data.
The 47-state Rt figure is a real number derived from real surveillance data using a documented methodology. It means COVID infections are trending upward in most of the country as of early September 2026. That is worth knowing.
It does not mean COVID is surging. It does not mean the situation is worse than prior summers. It does not tell you whether a novel variant is driving the growth. It does not tell you about hospitalization burden. The CDC says all of this, explicitly, on the same page where the number lives.
The number to watch — the one that would actually change the public health calculus — is the ED visit percentage trend alongside the hospitalization rate. The CDC's respiratory dashboard shows overall respiratory illness at "very low" nationally. Until that designation moves, or until variant surveillance shows a substantially immune-evasive strain gaining share, the 47-state figure is a directional signal from a low baseline, not an alarm.
Reporting it as an alarm is a denominator error. The denominator is the baseline burden. The baseline burden is very low. Forty-seven states growing from very low is not the same sentence as forty-seven states in crisis — even though both sentences contain the same number.
The most dangerous phrase in this particular episode of public health communication is "growing in 47 states." Not because it's false. Because it's true in a way that sounds like it means something it doesn't.
