The Startup
Published in

The Startup

Europe Controlled Covid-19 While America Failed. Or Was It the Opposite?

Each American state and European country was presented with its own unique pandemic challenge. How well did each area respond?

Photo by Chris Karidis on Unsplash

An underappreciated aspect of the pandemic is the point at which each state and country started. Some areas began with low infection rates and relatively small problems to solve; other areas started with astronomical infection rates that implied once-in-a-lifetime challenges. The starting point in New York was not the same as the starting point in Wyoming.

Opinions vary about how preemptively each area should have responded to the pandemic before the first death, but everyone agrees that areas should have been taking strong corrective action by the time the first death in each area occurred.

To provide a common basis for evaluating each area’s response, I’ll review how effectively each American state and European country controlled the virus relative to its specific starting point.

We’re Not in Kansas Anymore

I will use the phrase initial infection rate to refer to the number of infections that were present in each specific area as of the day of the first death. (For details on how to calculate the initial infection rate, see “Why Covid-19 Deaths in New York Were Inevitable.”)

The pandemic challenge levels faced by American states ranged from initial infection rates of almost zero in Kansas, Montana, North Dakota, South Dakota, and Wyoming to rates of 50,000 or more in Illinois, Louisiana, Massachusetts, Michigan, and New York. New York stands out as having the most challenging starting point, with about 450 thousand infections on the day of the first death in that state.

Figure 1 shows the initial infection rate in each state.

Figure 1

It is no coincidence that the states with the highest initial infection rates also experienced the highest early death tolls. New York had by far the highest initial infection rate of any state, and it also experienced the highest death toll of any state.

European countries also varied in their initial infection rates. Belgium, Germany, Italy, the Netherlands, Portugal, Romania, Spain, and the United Kingdom all had initial infection rates of 50,000 more. The highest initial infection rate, in Spain, was more than 250,000, but that is still almost 200,000 below New York’s initial infection rate. Figure 2 provides more details for European countries.

Figure 2

It Isn’t How You Start; It’s How You Finish

One way to look at how effectively an area has controlled the pandemic is to compare its initial infection rate to its total deaths. An area that had a high initial infection rate but controlled the pandemic toward a low death count can be seen as managing the pandemic well. An area that had a low initial infection rate but that nonetheless ended up with a high death count can be seen as managing the pandemic poorly.

Figure 3 shows a plot of American states in terms of initial infection rates on the X-axis and total deaths on the Y-axis. The total death data is from Johns Hopkins University and was current as of November 5, 2020.

Each state’s position on the X-axis shows its starting point. That state’s position on the Y-axis shows how well it did, relative to its starting point. If the starting points didn’t make any difference, you’d see a random scatter plot. But as the trend line in the graph illustrates, the points follow a generally diagonal pattern from lower left to upper right. That suggests that states’ starting points have influenced how well they have been able to control the pandemic.

The R-squared value of a linear relationship between initial infection rate and total deaths is 0.40, and R-squared value of a simple power-function relationship is 0.63. (The traditional interpretation of “R-squared” is that the R-squared number describes how much of one factor’s variability is explained by the other factor. In this case, it would be interpreted as the initial infection rate explaining 63% of the variation in total deaths.)

Figure 3

Notice that the bottom half of the graph is empty, i.e., the area around the “better” label. That suggests that if a state started with a certain level of infections, there was a limit to how few deaths it could have, no matter how well it controlled the pandemic from that point forward. The specific limit line is shown in Figure 4. States can do arbitrarily worse than the limit line, but they can’t do any better.

Figure 4

States vary significantly in how well they performed. The states that stand out as performing poorly are South Dakota and Florida, which are far above the limit line.

States that performed especially well — shown as being close to the limit line — include Vermont, Michigan, and New York. This is not intuitive at all, because New York has had the highest death total of any state by far. But it also had the worst starting point, by far.

Seen from this vantage point, New York looks like it addressed the pandemic fairly well, considering the deadly hand it was dealt. Indeed, New York hit peak contagiousness only 2 weeks after its first death, reduced infections quickly, and has not resurged since. (Check out the state infection graphs at the Covid Complete website to see just how effectively New York reduced its infection rate compared to any other state.)

You might be thinking, “The states that waited the longest to take action probably had the highest death rates, because more people became infected while they waited for the state to take action. If states waited until their first death to take action, of course the states that had their first deaths later also had the highest death rates.”

The data does not bear that out. Figure 5 shows each state’s initial infection rate, sorted from left to right by date of first death. (Note some dates are repeated.) Louisiana, New York, Michigan, Illinois, and Massachusetts had the highest initial infection rates, and their first deaths were within three days of one another.

Figure 5

Visually, there’s no correlation between day of first death and initial infection rate. The low R-squared value between those factors of 0.04 indicates the factors are not related.

We’ll Always Have Paris

Conducting the same analysis of initial infection rate vs. total deaths for European countries produces the graph shown in Figure 6. As with the American states, there’s significant variability in how well European countries have controlled the pandemic. France, Russia, Ukraine, Slovakia, and Poland stand out as having particularly ineffective responses. The country that started in the worst position, Spain, has had one of the more effective responses — notably better than Italy, the UK, Russia, and France.

Figure 6

This analysis differs from much of what has been written about the pandemic in Europe: Germany and Sweden (upper right) do not stand out as having done especially well compared to other countries; they have done about as well as would be expected based on their starting points.

Blue, No Yellow — auuuugh!

Americans have compared ourselves to Europeans endlessly throughout the pandemic. How have American states performed compared to European countries? Figure 7shows the states and countries all on the same graph.

Figure 7

The diagonal line that limits the minimum death count based on initial infection rate is the same for American states and European countries. A few states and countries are close to or on the line. The line is limited by small European countries at the bottom and large American states at the top.

A visual inspection of the points suggests that the American states are clustered more toward the middle whereas the European countries show more variation both on the better side and on the worse side. But overall the points are highly intermixed, with no clear advantage to either American states or European countries.

Controlled Deviation

Maybe I’ve spent too much of my professional life staring at productivity graphs, but, after a while, Figure 6 started to look to me like a statistical process control chart. If I draw ±1 standard deviation lines, I get the chart shown in Figure 8. In SPC terms, ±1 standard deviation lines can be viewed as upper and lower control limits — UCL and LCL. The darker diagonal line in the middle is the median.

Figure 8

Interpreting Figure 8 as if it were an SPC chart suggests areas of focus. In SPC, the variation among the points inside the control lines would be considered “common cause” variation — variation that’s due to inherent randomness in the system. Points outside the control lines would be a result of “special” variation. An SPC perspective suggests that we don’t stand to learn much by investigating points inside the control lines. We can improve the system by studying the points that show special variation, outside the control lines.

SPC would suggest investigating why France, Russia, Ukraine, Bulgaria, Slovakia, Florida, and South Dakota have performed so poorly. Similarly, it would suggest identifying the factors that led to New York, Michigan, Vermont, Andorra, Estonia, San Marino, and Monaco performing so much better than average.

I do not have the epidemiological background to investigate the root causes of certain states and countries performing so much better or worse than the rest, unfortunately. I hope that someone reading this article does have the right background, and I hope this analysis helps focus their investigation.

More Details on the Covid Complete Website

I lead the team that contributes the CovidComplete forecasts to the CDC’s Ensemble model. For more graphs, forecasts at the US and state-level, and forecast evaluations, check out the Covid Complete website.

--

--

--

Get smarter at building your thing. Follow to join The Startup’s +8 million monthly readers & +756K followers.

Recommended from Medium

CSVs Are Overrated! I Give up Some of Its Benefits to Gain More.

The Aiven Museum of Modern Monitoring Art

A woman admires a painting from data’s late expressionist period in the Aiven Museum of Modern Monitoring Art.

Drip by Drip: Creating a Water Commons

Seasonality Analysis and Forecast in Time Series

Data science productionization: portability

Data Science For Product Managers (DS4PM)

How B+Tree Indexes Are Built In A Database?

Recursive Feature Selection: Addition or Elimination?

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store
Steve McConnell

Steve McConnell

Author of Code Complete and More Effective Agile, Contributor to CDC Covid19 Forecasting, CEO at Construx, Dog Walker, Motorcyclist, Cinephile, DIYer, Rotarian.

More from Medium

How I Became the Ultimate Influencer

How do I ‘pitch’ my book?

The Gaping Hole That is You

Based on anecdotal research: now would be a good time to “phone a friend.”