Covid-19 Update, Data-Focused (7/23/20)
My personal diary on Covid-19 — with a heavy emphasis on graphs and data, and no politics.
This is an informal update that I’ve been providing a few times a week to a Covid-19 data science group. If you find value in this sort of informal update, please let me know by clapping, sharing the article, or emailing me at stevemcc@construx.com.
Hospitalizations, Test Results, and Estimated Cases
I thought I’d start today with the graph that shows hospitalizations, among other things. The graph is kind of a screwy format in that it plots each item against its overall average. In other words, it shows where the factor is at each point in time relative to its long-term average.
Points that are interesting to me about this graph:
- Positive tests are at an all time high, and the number of positive tests declined slightly for two months from early April to early June.
- Total tests have been climbing steadily since the beginning. Total tests more than tripled during that two-month period when positive tests were declining.
- Estimated cases (my estimates of cases) are about 60% of their all time high.
- Deaths are at about 40% of their all time high.
- Hospitalizations are at their all-time high.
These last two points, together, seem super interesting. You’d think there would be a better relationship between hospitalizations and deaths. But as I’ve complained before, hospitalization data was really spotty in the early days, so I didn’t pay much attention to it.
I’m inclined to interpret the hospitalization part of the graph as hospitalization data is getting reported more regularly than it was early on, but not that hospitalizations are truly at an all time high.
Other National Graphs
In other news, the increase in positive tests is clearly declining, as shown below. They’re still increasing, but at a much slower rate.
Deaths are pretty much stable, as shown below. Definitely not increasing, and definitely not increasingly meteorically, as some media reports would have us believe. Of course this varies by area, and in some regions the deaths are increasing.
The naive CFR continues to hover around 1.5%.
The State Picture
The state-by-state picture is very much in flux.
The Good
FL, AZ, and SC appear to be getting things under control (the solid green lines in the figure above).
Florida looks like it’s turned the corner in terms of positive tests, but the 14-day lag between test results and death reports mean we’re still going to have another 10–14 days of bad death reports from FL before things start to look better on the death side.
Arizona is looking a lot better. It’s easy to see the 14-day lag between positive tests and deaths in AZ’s smooth tests/death graph:
South Carolina’s tests have been down for about a week, but as you can see from its graph it likely has a few more days of increasing deaths before the death rate declines:
The Bad
NV, MS, AL, CA, TN are not in good shape and appear to be getting worse (the solid yellow lines).
NV, MS, AL, and TN are all increasing, but that’s in relative terms. Their absolute numbers are not that large. CA is the really ugly one.
CA’s deaths have increased, and positive tests have increased 25% over the past 14 days, so I expect to see the death rate follow suit. In addition, I have CA at peak contagiousness now. This all spells another 14 days of bad news for CA’s death rate. Hopefully the positive test rate has plateau’d and will begin to decline.
The Potentially Really Ugly
The area I’m really keeping my eye on is the northeast. We’re starting to see a bit of resurgence in RI, DE, MD, DC, VA. I would love to think that NY and NJ have established herd immunity, at lower infection levels than people thought were required (i.e., at 25–30%, not 80%). If positive test rates stay low in NY and NJ, there might be some merit to that idea. If those states start to take off again, I’m going to interpret that as there being no real way to keep the virus in check without either a vaccine or getting to herd immunity levels.
My Estimates
Deaths are tracking very closely to the low end of my most recent estimation range. My cumulative number is off by only 1.1% after 4 days.
I’ve received numerous comments on social media about my fatality estimates being optimistic. Far from it! I’m having trouble coming up with a principled basis for making my estimates low enough.
More Details on the Covid-19 Information Website
For more details and state-level data, check out my Covid-19 Information website.
My Background and Data Sources
I have been focused for 20 years on understanding the data analytics of software development, including quality, productivity, and estimation. The techniques I’ve learned from working with noisy data, bad data, uncertainty, and forecasting all apply to COVID-19.
The data used to create the graphs in this article is all readily available from the Covid Tracking Project and JHU. While the data is easily accessed, it is not very comprehensible in its native form. That is why I have created these data presentations.