The US outlier problem in three graphs or why the impact of omicron will be significantly more severe in the US
This makes it pretty obvious, doesn’t it.
I’m not sure you fully appreciate what a smooth curve fit can reveal. I don’t frequently reference Wikipedia, but on this subject, they have a nice animation at the top of the page. Click on it and it’ll show the development of a curve fit to unsmoothed data. The final fit in the animation sequence is what I was talking about. The location of an inflection point in the rising part of the curve is easily identified with the final curvefit, for example, as is the nominal peak.
That’s not relevant to reading these charts.
There are random variations from day to day, plus relative differences for days of the week due to more results being reported on certain days, e.g. some weekend results aren’t reported until Monday. A seven-day moving average smooths those enough to see a trend if there is one.
You seem to be imagining that it’s possible to read more out of the data than is there.
Not reading more than what’s there, but helping to reveal trends in the data. (I edited my post after your reply to clarify what I meant.)
What trends actually exist in the messy real-world data are visible in a seven-day moving average. In hindsight you can imagine that some of the noise was the start of a trend.
Well, I’ll concede that a 7-day average is better than none at all. But personally, I’d like to see the 7-day average data smoothed out, to help identify an inflection point in the rise and the peak in these data sets when it happens. That’s what I’ve been trying to do by visual or pencil-in smoothing. A couple of days ago I mentioned that I thought I could detect an inflection point in one of the 7-day average plots, which I thought was a good thing, for instance, because inflection points in previous covid surges have been about 1/3 to 1/2 the way to the peak.
There’s no such thing as an inflection point in that kind of complex, messy, real-world data. Things are bouncing around in multiple directions.
Yeah, that’s sort of the problem with not vaccinating, like, the entire global population at one time… You might get a virus that’s evolved to evade the vaccine… 
1 Like
Here are two global projections from healthdata. The first is global infections. (The peak is at 5M.) According to this projection, the world is very near the peak.
The second is a projection of global hospitalizations. (The peak is at 3M.) According to this projection, the world is at or near the inflection point right now.
And here’s a plot of actual hospitalizations in the US (not a projection), from Our World in Data:
This third plot shows all hospitalizations in the US to Jan. 15. It’s a naturally smooth curve due to the overlap in patients present in hospitals at a given time.
The major winter surge in the US hospitalizations on this plot began right at or just after Christmas Day. There’s a clear inflection point on 1/4 (or 1/5). As of 1/15 (this past Saturday), the peak had not yet been reached, but it appears that it is being closely approached. The peak looks like it might be reached later this month, or the beginning of February at the latest.
Trends in the curve for deaths in the US (below the hospitalization curve) are harder to see due to the compressed vertical rise on the y-axis scale used on this plot. While hospitalizations represent hard data, deaths usually occur two or three weeks after initial hospitalization, so those data aren’t as meaningful (i.e., probable undercount) as hospitalizations after about Jan 1 or slightly earlier. Deaths compared to cases have been relatively low, though.
I see all of this as good news.
=====================================================
Edit added 1/19: I’ve corrected the link to healthdata in this post.
I think something’s wrong there. The US Covid hospitalization curve became a logarithmic spike around January 9.
I’ve been watching this and similar graphs. The thing that’s annoying is that most of them show the data over such a long time period that it’s hard to see any detail in the current surge. Do we really need to see all of the past data all of the time? (That’s rhetorical.) Plots over the past few months are much easier to read, and more relevant to what’s happening now. That’s what we’re all interested in. Comparisons with past waves are of interest but we’ve all seen them – over and over. Let’s look at what’s been happening since November 2021.
So, yes, if you look at the CDC data plotted over a long time period, the 7-day averages of cases look like they go almost straight up. There are regularly spaced notches in the case graph that you posted yesterday, and I think that what looks like a peak at the top of that curve is just another notch. We probably haven’t reached the maximum number of cases nation-wide yet, actually. (I’m not sure but I think that the data for the US that Our World in Data used came from the CDC.)
I also think that case plots are a lot less meaningful than hospitalization plots, and that death plots can be misleading if you just look at the tail of the curves (last three weeks).
Case plots aren’t really very meaningful other than in a general way because they depend on how many reported infections have been documented, and also on how many tests have been conducted at the testing sites. A huge number of people are getting tested now, more than earlier in the pandemic for various reasons, so there are more positive tests.
Hospitalizations are clearly meaningful, much more so than case plots, because that’s what we hope won’t max out or stress out the health care systems beyond their capacity, and some hospitals are pretty close. Death plots are of great interest, of course, but there are more deaths in this wave due to the larger number of cases and hospitalizations, as people have been saying over and over. But happily, the fraction of deaths continues to be relatively low.
I think @honkman had already referenced diminishing returns w/ additional boosters.
You can look at the data in the CDC chart by right-clicking and choosing “Show as table.” The previous peak was 3.70 on August 26. Then it went gradually down to 1.52 on November 5. Then it went gradually up to 2.40 on Dec. 13 and hovered around there until Dec. 21, after which it started increasing rapidly, doubling to 4.50 in 12 days (Jan. 2).
The dip at the end is subject to change as more data comes in.
Maybe the 7-day average (not "naturally smooth) in the Our World in Data chart is enough to mask the size of the decline from Oct. 1 to Nov. 5. There’s no reason to use a 7–day average for hospitalizations, admissions don’t drop off on the weekend the way test results do.
1 Like
In a roundabout way, this isn’t a bad thing. Speaking strictly for myself, I just got an mRNA booster three months ago, so talk of a fourth shot isn’t especially appealing. I think I’m adequately protected against croaking from Covid and so I’m not particularly worried about catching it. I’ve had colds, I’ve had flu, and like them, I can be sick for a week with covid and stay away from others to keep from infecting them. I can deal with the idea of getting a booster shot once a year, but every few months? That might get really old really fast.