Some Background on Accuracy
As I’ve watched the tracker at 1point3acres approach one million cumulative cases, a milestone it will reach sometime later today, I’ve become increasingly aware of the disparity in numbers between the trackers. For example, here are the most recent numbers from several trackers:
|New York Times||965,214|
Some of these differences are easy to explain. For example, the CDC number is so much lower because it only really updates on weekdays at 4pm. Since I’m writing this on a Monday morning, that means it’s numbers are more than two and a half days old. However, the other trackers had all updated within an hour of my noting their numbers and there is nearly 30,000 cases difference in their reports. That comes to about a 3% difference, which sounds less startling, but it still amounts to about a day’s difference on when they will reach one million cases.
There’s an easy answer to the question of which tracker we should believe: None of them. All of these numbers are just approximations based on interpreting inadequate data. Each tracker has to make decisions about when to include or exclude a data point in its count. 1point3acres is the most aggressive of these in including cases, CDC is the most conservative in doing so. At best, these trackers give us an estimate of known cases, but it is increasingly apparent that a large majority of cases are unknown. This isn’t a surprise, even early reports indicated a high proportion of asymptomatic and mild cases. For example, nearly half of those who tested positive on the Diamond Princess were asymptomatic. Because the passengers on the Diamond Princess were generally older, and older people are more likely to be symptomatic, we should expect the majority of cases to be asymptomatic. In mid-March the WHO said that about 80% of cases seemed to be either asymptomatic or mild.
Given the lack of testing in the US, those with mild or no symptoms are unlikely to be confirmed to a degree needed to show up on these trackers. The trackers will also miss those who have symptoms but who, for a variety of reasons, are never diagnosed. Again, lack of testing has a role here. Many locales recommend that even those with symptoms try to stay at home as long as possible so as not to overwhelm hospitals. Further, we know that some deaths have been missed, though not exactly how many.
My working guess, and it’s only a guess, is that the trackers probably only catch about half of symptomatic cases and about one tenth of all cases. I guess they also only catch about half of all COVID related deaths. These aren’t blind guesses, but it would be a stretch to call them educated. They happen to match with a few data points, but they’re driven as much by a desire for nice round numbers as they are by looking at the relevant research. But given those guesses, a difference of 3% between trackers seems almost trivial.
Changing Primary Tracker
Starting today, I’m shifting my primary tracker, the one I’ll base my projections on, from 1point3acres to Johns Hopkins. There are a couple of reasons for this.
My goal in these posts has been to help people manage expectations by making the numbers more understandable. Whatever the real numbers, it can be confusing to see one source say we have a million cases and another source say we’re still 30,000 shy of that. Johns Hopkins is more often referenced in media reports, so it will track better with what you’re likely to see there. This should keep the projections more closely aligned with common sources, which should make it more helpful.
John Hopkins shares its data set publicly, 1point3acres doesn’t. By using Johns Hopkins’ data I can run different comparisons, an ability that will be increasingly important as we shift from trying to get a lid on explosive growth to just trying to keep the lid on it. For example, I can use their data on hospitalizations and testing to get a better idea of the scope of the epidemic. This gives me the opportunity to burnish some old skills and learn some new ones, something I enjoy doing. I’ll try to have some new things to show soon, but for today, it’s just numbers.
It will take me some time to fully switch to the new dataset, so today’s projection is something of a hybrid. I’ll be starting with Johns Hopkins’ numbers from last night, but I’ll be projecting based on the growth rates from 1point3acres. This isn’t ideal, but it should be fairly close. If it turns out to be an issue, I’ll post an update once I start to get the new system running.
Here’s the 10 day projection
By the way, in the time it has taken me to write this, the 1point3acres tracker has risen to over a million, the Johns Hopkins tracker is still just under 980,000.
Thanks for reading.