Daily COVID-19 Data Is About to Get Weird

Based on what we’ve seen over the last eight months of state-reported COVID-19 data, we think two big, potentially misleading things are about to happen to the testing, case, and death numbers that allow us to track the pandemic in the United States. 

First, by Thanksgiving Day and perhaps as early as Wednesday, all three metrics will flatten out or drop, probably for several days. This decrease will make it look like things are getting better at the national level. Then, in the week following the holiday, our test, case, and death numbers will spike, which will look like a confirmation that Thanksgiving is causing outbreaks to worsen. But neither of these expected movements in the data will necessarily mean anything about the state of the pandemic itself. Holidays, like weekends, cause testing and reporting to go down and then, a few days later, to "catch up." So the data we see early next week will reflect not only actual increases in cases, test, and deaths, but also the potentially very large backlog from the holiday. 

Why the data wobbles

We compile and publish COVID-19 data organized by the date on which it's reported, rather than by date of specimen collection, data of symptom onset, date of death, etc. To see how holiday delays affect this data, we can look at the way weekends and holidays have caused predictable dips and rises in the numbers we compile every day from US states and territories.

If you’ve been following the data we report, you’ll probably be familiar with the day-of-week effects that make many state-reported COVID-19 metrics so jagged on the charts. On Wednesday through Saturday, we tend to see peak reporting for tests, cases, and deaths. Sunday and Monday, on the other hand, are usually very low in comparison. (This is the main reason we use—and advocate for the use of—seven-day averages for most COVID-19 metrics.) 

The reasons for these effects are many, and extend from test administration all the way through to the process of getting the data onto an official website. On weekends, fewer doctors’ offices and other testing sites are open, so fewer people get tested, which means that fewer tests make it to labs. The reporting systems, too, are affected: Fewer results are reported to health departments, and fewer health department staff are at their desks to turn those results into the data points we eventually see under tests and cases. 

For most states and territories, Saturday results are generally reported on Sundays, and Sunday results are generally reported on Mondays. At the national level, test reporting drops Sunday through Wednesday, with the biggest drop on Tuesday. Case reporting falls on Sundays and Mondays, with a smaller dip on Tuesdays. These weekday patterns suggest that most jurisdictions don’t fully catch up from the weekend until Wednesdays for cases and Thursdays for test reporting.

Death reporting, too, is affected, much more noticeably. Reporting for deaths already takes more than a week on average—and can take much longer in some states. On weekends, we see death reporting drop by an average of 44 percent from the seven-day average on Sundays, and 48 percent from the seven-day average on Mondays, with the highest reporting on Wednesday. (We've written previously about the ups and downs of reporting deaths by date reported, rather than by date of death.)

On Thanksgiving, we expect to see the equivalent of a double-weekend pattern. Far fewer people will be tested on Thanksgiving Day, and perhaps on the day after as well, and then the usual weekend pattern will begin. The same will be true for reporting. We don’t expect that most US jurisdictions will fully staff public health departments from Thursday through Sunday, so we anticipate that reporting of tests and cases will be delayed through at least Monday, November 30—and perhaps several days longer, considering the depth of the likely backlogs produced by a four-day slowdown during a major case surge. Death reporting, too, will slow down for an unknown number of days.

So how long will the reporting be knocked off course? It’s impossible to know for sure. We saw holiday effects over the summer on July 4 and Labor Day, both holidays that took place during periods of relative stability in the course of the US pandemic. For both of these holidays, reporting was only substantially affected on two days, the day following the holiday weekend and the day when the majority of the backlog rolled in.

It’s also possible that Thanksgiving will produce only relatively minor disruptions in the data, but the differences between this week and the summer holidays have us concerned. Thanksgiving falls on a Thursday and may disrupt staffing in many test sites, labs, and health departments on the Friday after the holiday as well—which, if it happens, is more than understandable given the endless crisis sprint so many healthcare, lab, and public health workers have been on since the spring. Massachusetts and Wyoming have already noted on their dashboards that they will not report on Thanksgiving Day, and we expect many other states to follow. We’re also not expecting that all US jurisdictions will manage to fully staff their reporting on Friday. Washington State has announced that it will not report between November 23 and November 30.

A four-day (or three-and-a-half-day) weekend could result in a backlog that takes extra time to process. We’re also in a very different place this November than we were in this summer and early fall. Tests and cases are dramatically higher now than they were over the summer, and deaths have been rising sharply several weeks behind cases, which suggests that the people who process death-reporting paperwork in states with severe outbreaks are also likely inundated.

What to watch

There are, however, a few metrics that we expect to remain relatively stable through the holiday. Current hospitalizations reported by states and territories have not shown as much volatility as tests, cases, and deaths, which makes sense—these numbers are reported by hospitals or hospital associations, and hospitals don’t get weekends off. (Even hospitalization metrics include some connections to the testing and case reporting systems, because COVID-19 patients aren’t confirmed to have the disease unless they receive a positive PCR test. Hospitalized patients’ tests generally receive priority, allowing their results to be processed and reported as quickly as possible.)

Likewise, the new admissions metric in the public hospitalization dataset from the US Department of Health and Human Services shows only moderate volatility, and will likely be an additional source of useful data through the expected holiday dip and subsequent spike in test, case, and death data.

So when will we see cases from holiday gatherings?  

The actual case increases from Thanksgiving exposures—people who got COVID-19 during the holiday weekend—probably won’t start showing up in the data until the second week of December. Succeeding waves of infections from holiday gatherings will roll in for weeks. From what we’ve seen so far, the virus can spread with remarkable speed, but there are delays at every step in tracing and reporting its spread: It takes time to get tested, time to get and report a result, time to trace close contacts—and to start the process over again with a new circle of exposures. 

Consider the now infamous Millinocket, Maine wedding superspreading event this summer: The day after the wedding, the “index case” wedding guest developed COVID-19 symptoms, according to reporting from the Los Angeles Times and a CDC report—though this person wouldn’t receive the results of their PCR test until six days after the wedding. Three days later, a worker at a long-term-care facility who hadn’t attended the wedding, but whose child had, began showing symptoms. The test for the LTC worker didn’t come back until 11 days after the wedding, by which time they had worked multiple days in the facility while ill. Their test results were presumably reported to the local health department immediately, but it usually takes another day or two to see cases roll up to the state level and get reported to the public. The LTC worker who started feeling ill only four days after the wedding might not have shown up as a case in Maine’s official state data for almost two weeks. By the three-week mark, 30 people with ties to the wedding had confirmed infections, four people were in the hospital, and one had died. None of the hospitalized patients had attended the wedding. 

The virus swept through the long-term care facility, sickening more than half of the facility’s residents, six of whom died. It reached a nearby jail, sickening 46 inmates and 22 staff. By a bit more than a month after the wedding, public health officials were able to trace 177 confirmed cases and seven deaths to the event. Notably, officials were able to move quite quickly to identify cases and outbreaks, in part because the case rate was extremely low in most parts of the state and it was possible to successfully contact-trace wedding attendees.

Responsible reporting during data irregularities

At The COVID Tracking Project, we’ll be flagging all our data reporting over the holiday weekend and much of the subsequent week with reminders that official state reporting is expected to drop and then spike as backlogs get processed. If you’re a reporter covering COVID-19, we recommend focusing on current hospitalizations and new admissions as the most reliable indicator of what is actually happening in your area and in the country as a whole. We also recommend reminding your audience that Thanksgiving infections are unlikely to be clearly visible in official case data until at least the second week in December, and that neither the holiday drop in cases nor the immediately following rise should be taken as evidence for anything to do with the pandemic itself.