CDC reporting of Omicron numbers

After initially estimating that nearly three-quarters of U.S. COVID instances between Dec. 11 and Dec. 18 have been attributable to the omicron variant, the CDC revised that quantity on December twenty eighth to simply 23%. Since then, they’ve revised it as soon as extra to 38%. Besides they didn’t: none of these estimates have been as exact as we simply made them sound. For instance, the present estimate is someplace between 31.4% and 44.7%.

Polls printed in the course of the pandemic counsel that belief within the CDC is faltering each amongst health professionals and the general public. These and different polls don’t get into the explanations for distrust, however specialists typically agree that trustworthiness will depend on the solutions to 5 questions: do I’ve purpose to assume this particular person or establishment is: competent, dependable, honest, benevolent, principled?

Occasions over the previous two years have brought on many individuals to query every of these judgments. Huge revisions to accounts, similar to these cited above, name into query the CDC’s reliability, particularly. However this specific mishap was preventable–with higher reporting.


The core narrative that the omicron variant unfold shortly and is now dominant has confirmed true. Nevertheless, particularly in mild of the mass quantity of misinformation round Covid, getting the numbers fallacious additional erodes public belief within the one establishment that each the general public and the media ought to be capable to rely upon. That is the place the place accuracy is probably the most essential.

There are two simple issues the CDC–and each information outlet that picked up the preliminary 73% quantity–may, and will, have performed otherwise. Each of them improve transparency about uncertainty, which is a key half of all rising science. We’re half of a group of social scientists and journalists who work intently collectively to encourage better reporting about statistics. Particularly, we’re making an attempt to determine methods to assist folks with out statistical experience draw extra correct conclusions. Right here’s what we’ve realized, and the way it may have improved this case.

First, all visualizations and information studies about omicron’s prevalence ought to have included confidence intervals. Statistics is, primarily, the science of estimating one thing a few inhabitants by wanting rigorously at a randomly chosen subset of that inhabitants (generally known as a “pattern”). Even when the pattern is completely consultant, there’ll all the time be some uncertainty. That’s simply how chance works, and a confidence interval is a means of representing this sort of uncertainty. It tells us the vary the place the true worth is prone to fall. Though confidence intervals are sometimes generally known as “margins of error,” they don’t imply somebody has made a mistake. They merely assist us see the variability inherent in all estimates.

The CDC’s prediction web page consists of this info in a sidebar, however their visuals don’t present it. The 73% determine made headlines partially as a result of it was so shockingly excessive—however the authentic estimate was between 34% and 99%. That’s a very monumental vary, and one that ought to have been included within the CDC’s personal visualization–and each single information report. In any case, revising from a spread that features 34% all the way down to 23% appears a lot much less dramatic than revising from 73% all the way down to 23%. At the least we’re seeing some enchancment within the reporting: the NPR story on the revision notes that the interval stays massive.

Second, all reporting ought to have centered on the restrictions of the strategies. The CDC’s variant info web page notes in small textual content that “information [from the last few weeks] embrace Nowcast estimates, that are modeled projections that will differ from weighted estimates generated at later dates.” However the news coverage didn’t embrace such caveats, for probably the most half. And even the later weighted estimates depend on native public well being reporting, which varies by jurisdiction, in addition to statistical procedures for imputing lacking information.

The position of information is to maintain the general public knowledgeable, and which means being as clear about what we don’t know as what we do. In phrases of public well being, it also needs to imply working with the CDC to make sure that everyone seems to be adhering to greatest apply in statistical communication. All of this uncertainty is a component of how science works. Leaving it out isn’t simply simplifying the story – it’s actively deceptive.

Jena Barchas-Lichtenstein, Ph.D., is a linguistic anthropologist who leads the media analysis at Knology. They’re co-PI of Significant Math, a four-year NSF-funded collaboration with PBS NewsHour to enhance statistical reporting.


John Voiklis, Ph.D., leads analysis on behaviors, norms, and processes at Knology. He educated in social and cognitive psychology and has taught statistics for researchers and analysis strategies for information scientists.