In this series of posts I’ve chronicled my own learning about the coronavirus pandemic, focusing in particular on estimating the prevalence of the virus especially in the US population. My schooling and professional background gave me a head start on this exploration; my demographic vulnerability and my curiosity have motivated me to persist.

It’s possible that readers of these posts could have come away not only with the results of my investigations, but also with a battery of methods and resources for conducting their own analyses. It’s also possible that, having written these posts, I could organize what I’ve learned and how I learned it into a class for teaching others — call it Applied Epidemiology: Covid. There wouldn’t be many lectures, nor would there be many assigned background readings. Instead, the class would consist mostly of a series of projects, progressing from simple to complex. The work would be cumulative, with knowledge and skills and resources acquired in earlier projects being brought to bear on subsequent projects.

Here’s a project that would be ideal for this hypothetical Applied Epidemiology class.

The Canadian government just released summary findings from a nationwide covid seroprevalence survey. The assignment: evaluate this survey.

The article includes both descriptive and evaluative material, giving the student a head start. But anyone who’d have taken this class would have developed hands-on competence in using a variety of other evaluative criteria and resources that go beyond the article.

 

More Continuity

State-specific data analyses for the 14 days between August 24 and September 7 largely confirm findings from prior intervals. The average death rate across all states over the last two weeks is the same as it was during mid-August. As reasserted any number of times in this series of posts, death counts are the most accurate indicator of infection. And as the state-by-state analyses have shown repeatedly, the best predictor of a state’s current death toll is its prior death toll. That’s the case again this time for predicting state-specific deaths per 100K of population: the infection rate from the preceding ten days correlates highly with current death rate (r = .79), but the correlation of deaths prior with deaths present is even higher (r = .92). Continuity prevails: high infection rates cause high death rates, and high infection rates self-perpetuate across time through contagion.

Hot Spots

There has been some change in the collection of states saddled with the ignominious title of High Outlier for test-positives and deaths.

States with test-positive rates more than 50 percent above the across-state average: Alabama, Arkansas, Georgia, Iowa, Kansas, Mississippi, Missouri, North Dakota, Oklahoma, South Dakota, Tennessee.

States with death rates more than 50 percent above average: Alabama, Arizona, Arkansas, Florida, Georgia, Louisiana, Mississippi, Nevada, South Carolina, Texas.

The southeastern states remain hot spots, as they have for most of the summer. Now though there’s a test-positive surge in the central states, extending from the western Gulf Coast north to the Canadian border. Is this fallout from last month’s annual Harley rally in Sturgis South Dakota? Is it the disproportionate cultural and financial influence of Texas — an outlier throughout the summer — through this vertical band of the country?

Comparison with the EU

Recently I ran an analysis comparing (unfavorably) the US versus Western Europe in controlling the covid pandemic. A friend suggested that I might be cherry-picking the most affluent European nations to make the US look bad. Wouldn’t it be more apt to compare the entire EU and all of its member countries with the US and its individual states? Very well: here goes.

Here’s a link to a table showing the most recent 14-day covid death rate per 100K of population of all the EU/EEA countries plus the UK. The average across those 31 countries is about 0.4 deaths per 100K. Over that same 14-day span the average US state experienced 3.0 deaths per 100K — 8 times as high as Europe. The top five Euro countries for death rates are Romania, Bulgaria, Spain, Malta, and Croatia: they averaged 1.6 deaths per 100K over the past two weeks. The ten US high-outlier states identified earlier in this post averaged 7 deaths per 100K — nearly 5 times as high as the high-outlier Euro countries. Of the 50 states + DC, only 4 — Connecticut, Maine, New Hampshire, and Vermont — had death rates as low as or lower than that of the average European country.

A Procedural Note

I’ve been doing these periodic updates primarily in order to understand the interrelationships among the corona aggregate indicators — diagnostic tests conducted, test-positive rate, percent test-positive, death rate, geography, time. And in turn I’ve used those interrelationships among measurable indicators in order to construct a fairly reliable and valid and straightforward estimate of infection rate. I think I’ve got a pretty good handle on it by now. If the US were to conduct a nationally representative random sample of diagnostic tests or antibody tests, then my understanding might change dramatically. Now it’s a matter of fine tuning what I already know. The effective contagion rate goes up and down over time, the specific states experiencing spikes gradually shift. The body count goes up and up. It might be time for me to discontinue the periodic updates, or to spread them out over longer time spans.