Interpretation Guide

The Trend graphs (sparklines) showing the Daily Delta (day-to-day change) in the Raw Numbers for the Test Results should be interpreted in isolation – not in comparison to other states, because they have individual vertical scales to maximize visibility. The horizontal time scale in days, however, does line up for comparing time occurrences.

The area under the curve for a Trend graph is equal to the to corresponding Raw Number total (see picture, below). Additionally, the Maximum (peak) Daily Value is highlighted (see picture, below) with a darker line in the bar graph (sparkline). These techniques are also used for the U.S. Total Trend graphs on the far right side of this visualization.

The Trend graphs (sparklines) showing the Test Results Normalized Per Population (per capita results) allow for meaningful  comparisons across states, because they are displayed in the same percentage scale for all states, as well as for the U.S. Average at the upper right of the visualization. In other words, the normalized (per-capita) percentage value shown by the height of each bar in the Trend graph (sparkline) for, say – Deaths (Dead) – is displayed at the same scale for all of the Trend graphs for Deaths across all states and the U.S. Average.

Note that the types of results are at different scales. Thus, the Trend graphs for Positive Test Results (infection rate), Negative Test Results (not infected), Hospitalized (hospitalization rate), Deceased (death rate), and Total Test Results – all use different scales. The Current Daily Value in the Trend graph is highlighted by a darker color, and this value is equal to the percentage shown in the Test Results – Normalized Per Population, to the left of the Trend graphs (see picture, below).

For a given percentage value shown with four decimal points of precision, you can easily interpret these as “per million” value. For example, in the picture below, Washington state (WA) has a normalized (per capita) Death rate of 0.0079%. This can be equivalently stated as 79 Deaths Per Million of population in Washington state.

Analysis of COVID-19 Epidemic

This visualization of the data shows that COVID-19 outbreaks in the U.S. are highly localized to urban metro areas. Four metro areas – New York City, New Orleans, Detroit, and Boston make up less than 10% of the U.S. population, but account for 70% of COVID-19 deaths. This observation has continued to be accurate since this analysis was first posted on 3-29-2020.  

New York City’s outbreak crisis shows up in NY, NJ, and CT state data. Urban density (lack of social distance), heavy use of mass transportation (subways, trains, and buses), high levels of international travel, and the 1-25-2020 Lunar New Year celebration in NYC’s Chinatown are drivers of the epidemic in New York City. Problem areas shown by the numbers and visualization in the states of NY, NJ, and CT are driven by the outbreak localized to the vicinity of New York City. Ideally, more detailed data by county, voting districts, cities, and zip codes will be made available to make better visualizations and analysis.

New Orleans has the next largest COVID-19 outbreak after NYC, which is reflected in Louisiana (LA) state data. The outbreak was fueled by the recent Mardi-Gras celebration on 2-25-2020 with many visitors from China and Europe. New reports speculate that the high death rate is driven by ill health with higher rates of obesity and diabetes in New Orleans.

Detroit (shown in MI state data) and Boston (visible in MA state data) are the metro areas with the third and fourth, respectively, most severe outbreaks. News analysis of Detroit suggest multiple factors including: direct flights to Wuhan China thru late January, Sander’s campaign event on 3-6-2020 with 6,000 people, primary election on 3-10-2020, poverty, ill health (obesity, diabetes, and asthma), and poor air quality as possible drivers. News accounts of the Boston outbreak suggest causes including: air travel from Wuhan and elsewhere in China thru late January, significant air travel from Italy and Spain thru early March, and outbreaks at homeless shelters.

The initial early outbreaks in Washington state (WA) and Vermont (VT) have leveled off, and are, hopefully, under control.

Analysis of the Data

My data source is The COVID Tracking Project, which has done a great service for our country. Started by three people, it has grown to over 1,000 volunteers. They have accomplished a great deal of complex work to coalesce the data from 50 states and 5 territories, because they all report their data in different ways using many different web and database technologies.  However, because the project as well as the health departments of all the states involve people and many steps, there is opportunity for errors and uncertainty to be present in the data. I have corrected a few dozen data errors, which I detected – how else – by visualizing the data! Most of the errors involved negative numbers for daily changes in the test results. Because I am visualizing only the cumulative data, daily changes can only be zero or positive numbers.

Just by looking at the Trend graphs of Daily Deltas (rate of change) you can plainly see that Pennsylvania (PA) has done the best job of consistently reporting Positive, Negative, and Total Test Results each day. In like fashion, you can clearly see that Ohio (OH) has done the best with reporting Hospitalization numbers each day, while New York (NY) has done the best job of the grim task of reporting daily death numbers. Overall, New York (NY) has done the best with daily reporting.

You can also clearly see that Hospitalization numbers for most states are very sporadically and inconsistently reported. This seems bad because, how can you allocate resources like ICU beds and ventilators if you don’t have proper data collection and reporting? To the credit of The COVID Tracking Project, they did add ICU and ventilator statistics to their data model after a couple weeks, but only a handful of states are reporting these additional details. Nebraska (NE) and Puerto Rico (PR) have still not reported any Hospitalization numbers. About 8 states have only reported Hospitalization numbers 1 or 2 times in the past month. Only about 16 states are consistently reporting every day.

What Does the Data Tell Us? When Will It End?

[Discussion is dated 4-18-2020]  If you look at the U.S. Totals, we appear to be past the peak of infection rate, hospitalization rate, and death rate. According to Farr’s Law (which has never been wrong since statistical scientist William Farr discovered it in 1840), it will be over in 3-4 weeks.

Here’s a good discussion of Farr’s Law and its application to the COVID-19 epidemic from the Centre for Evidence-Based Medicine at the University of Oxford:  cebm.net/covid-19/covid-19-william-farrs-way-out-of-the-pandemic/.

[Discussion is dated 4-18-2020]  If you look at the states that were hit the hardest, you can also see that by Farr’s Law, this will end in 3-4 weeks. For states like Rhode Island (RI) where the start of the outbreak was delayed (until folks from NY brought it in), it will take 5-6 weeks for it to end.

High Resolution Image

The image shown below is scaled to fit within the WordPress limit of 2500 pixels in width, so it is only 150dpi.  Click on the picture to open a 600dpi high-resolution PNG image file in a new tab.

If using a computer, click again to zoom in. Then click yet again to zoom back out. Or use keyboard [Ctrl +] and [Ctrl –] to zoom in and out.

If using a phone or tablet, use reverse pinch gesture to zoom in and pinch gesture to zoom out.