There seems to be a weekly rhythm to covid-19 deaths and new cases?
June 12, 2020 4:34 PM Subscribe
There seems to be a sine wave in the daily covid-19 data for USA daily deaths and daily new cases. Is this an artifact of data collection procedures, or is there a correlation with day of the week? I'm looking at data from NYTimes, Johns Hopkins, & the EU CDC. Here are a few charts that show the trends I'm asking about. What gives?
Best answer: It's been the same in every country since the start. Lower numbers at the weekend, because it's the weekend and there aren't so many people working (daily deaths aren't deaths that actually happened in the previous 24 hours, they're deaths that were reported, and you need people working to do that reporting). Then on Monday/Tuesday catching up on that lag.
posted by cincinnatus c at 4:43 PM on June 12, 2020 [3 favorites]
posted by cincinnatus c at 4:43 PM on June 12, 2020 [3 favorites]
Best answer: Data collection/reporting isn't real time, and data collection in particular doesn't happen 24/7. It can be anywhere from a few days to a week or two behind, depending on jurisdiction.
on preview: basically what cincinnatus c said.
posted by pdb at 4:49 PM on June 12, 2020
on preview: basically what cincinnatus c said.
posted by pdb at 4:49 PM on June 12, 2020
Best answer: Yeah, as others said basically the per-day counts aren't necessarily accurate because the reporting of each case and death isn't precise to the day. The reporting systems seem to be more designed to ensure that the cumulative number of cases up to a given day represents the best estimate for the number of cases that have occurred, and in fact this is what the Johns Hopkins dataset that most people seem to use actually reports, not per-day numbers.
If you look at deaths instead of cases, this is effect is more obvious because of the relatively smaller numbers, and in fact sometimes the reported deaths in a particular area can even be negative. Sadly this doesn't indicate people are returning from the dead, but just reflects normal adjustments to reported numbers for a variety of administrative reasons, which keep the cumulative total count accurate even when the per-day count isn't necessarily. This is why most of the time people report a moving average for the per-day cases and deaths: it smooths out most of the administrative slop caused by imprecision in when each given case/death is reported. Most of the example plots you show report this, and it's probably better to focus on this line rather than the daily numbers. Probably better still would be to just report weekly rather than daily numbers, but the 7-day average is a pretty good approach.
What this means in practice is that the nature of the data we have access to for COVID cases (at least for the Johns Hopkins dataset that everyone seems to use) is such that differences in case counts or deaths on a time scale shorter than about a week or so don't really have much meaning. It's reasonable to ask whether cases have gone up or down since last week, but not really to ask whether they've gone up or down since yesterday. I've seen news articles reporting things like "spikes" in cases locally within the previous three days or something, and this just doesn't really mean anything, and sure enough if you wait a few days and look at the local numbers again they generally haven't changed much. (Which isn't to say there isn't anywhere that cases have started increasing again: Arizona in particular is back to a disturbingly rapid growth in cases.)
posted by biogeo at 5:41 PM on June 12, 2020 [4 favorites]
If you look at deaths instead of cases, this is effect is more obvious because of the relatively smaller numbers, and in fact sometimes the reported deaths in a particular area can even be negative. Sadly this doesn't indicate people are returning from the dead, but just reflects normal adjustments to reported numbers for a variety of administrative reasons, which keep the cumulative total count accurate even when the per-day count isn't necessarily. This is why most of the time people report a moving average for the per-day cases and deaths: it smooths out most of the administrative slop caused by imprecision in when each given case/death is reported. Most of the example plots you show report this, and it's probably better to focus on this line rather than the daily numbers. Probably better still would be to just report weekly rather than daily numbers, but the 7-day average is a pretty good approach.
What this means in practice is that the nature of the data we have access to for COVID cases (at least for the Johns Hopkins dataset that everyone seems to use) is such that differences in case counts or deaths on a time scale shorter than about a week or so don't really have much meaning. It's reasonable to ask whether cases have gone up or down since last week, but not really to ask whether they've gone up or down since yesterday. I've seen news articles reporting things like "spikes" in cases locally within the previous three days or something, and this just doesn't really mean anything, and sure enough if you wait a few days and look at the local numbers again they generally haven't changed much. (Which isn't to say there isn't anywhere that cases have started increasing again: Arizona in particular is back to a disturbingly rapid growth in cases.)
posted by biogeo at 5:41 PM on June 12, 2020 [4 favorites]
Best answer: Washington state's coronavirus page has a data visualization with a lot of information about coronavirus in the state, including testing, and yes, there are very clearly far fewer test results coming in on weekends.
Most of the "time based" data is when the information was received, not when people actually got sick or died. So if you're trying to draw conclusions from it, you need to keep that in mind.
I'm not sure if it's even possible to get solid information on when people actually contracted the disease in any sort of rigorous fashion. Some part of the Washington State data on deaths they are trying to report by actual date of death, but they say there's about a week lag where the data is unreliable because they're still getting reports and organizing the data, and I've noticed that in reality, it's a bit more than that because reports keep trickling in from people who died a while ago. And the information that ends up in the national/international level databases on Washington state is dated by when the deaths were reported anyway, just to try and keep it consistent with the data they are getting from other states and countries.
posted by Zalzidrax at 7:44 AM on June 13, 2020
Most of the "time based" data is when the information was received, not when people actually got sick or died. So if you're trying to draw conclusions from it, you need to keep that in mind.
I'm not sure if it's even possible to get solid information on when people actually contracted the disease in any sort of rigorous fashion. Some part of the Washington State data on deaths they are trying to report by actual date of death, but they say there's about a week lag where the data is unreliable because they're still getting reports and organizing the data, and I've noticed that in reality, it's a bit more than that because reports keep trickling in from people who died a while ago. And the information that ends up in the national/international level databases on Washington state is dated by when the deaths were reported anyway, just to try and keep it consistent with the data they are getting from other states and countries.
posted by Zalzidrax at 7:44 AM on June 13, 2020
« Older What are the Medicare eligibilty regulations for a... | Children's book about a house Newer »
This thread is closed to new comments.
posted by duien at 4:42 PM on June 12, 2020 [4 favorites]