About That Financial Times Graph

We propose a more precise way of aligning and comparing the case trajectories of different countries.

The case graphs produced by the Financial Times (FT) have been widely cited by others in the media. Because of how the data is aligned and presented, it is possible that people are misreading the data and making the wrong conclusions about where we are and where we're headed, particularly in the US.

To summarize the FT graph plots the total number of cases for each country. Because the pandemic hits different countries on different days, the Financial Times shifts each countries' data to the first day that the country exceeds 100 cases. While this approach is more useful for comparing the spread in different countries than simply plotting countries' cases by date, a key flaw is that it doesn't account for the range of populations across the various countries. To correct this, we propose recasting the data in two critical ways:

1. Instead of comparing countries by total cases, we normalize the data to the countries' populations (that is on a per capita basis). While other sources (e.g., Worldometers) normalize to cases per million, we prefer cases per 100,000 for reasons that should be clear from #2.

2. Given the measure of cases per 100,000, we align countries using when each country crosses the threshold of at least 1 case per 100,000. For a country of 10 million people (e.g., Portugal) this would be the same as the 100 case threshold used by the Financial Times. For smaller countries their starting points would be moved sooner, and for larger countries, including those in our large western democracy benchmark group, the starting points would be moved later.

Once we make the above adjustments, what was once a muddle now looks pretty clear:

It also makes sense to apply the new time-shifting threshold to the growth rate curves:

Note that another change that we have made is to change the growth rate to five-day lookback instead of seven days (1 week). This still has the benefit of smoothing out the daily noise and is consistent with the incubation time (3-5 days) of the virus. Specific, the growth rate is calculated as:

N = Number of cases

M = Number of cases five days prior

Growth Rate = N / M

Takeaways:

1. After re-aligning the data with the new approach, the observation from a previous post is even more clear, namely that despite wild variations in their initial case growth rates, the countries seem to converge on the same or closely parallel trajectories after a couple of weeks.

2. Notwithstanding the previous comment, the graphs also show that when a country doesn't control the virus early (specifically the US and Spain), it is difficult to get back to the common trajectory. Even as they flatten their case curves, Spain has an estimated 90% more cases and the US an estimated 40% more cases than they would have had they taken quicker action to apply social distancing to reduce the spread of the virus.

3. Note that the jump in France's cases on day 29 (April 3 for France) was due to the correction of an under-reporting of cases and does not reflect a change to the underlying spread of the disease. Interestingly enough, this brought them more into alignment with the rest of the group.

4. Despite the early disregard of social distancing protocols (since reversed), the UK actually appears from these graphs to be doing a little bit better than its cohort on a per capita basis. It's possible that the disconnect with the public narrative is due to an under-reporting of cases (ala France), but the death rate of total cases (around 11%) is consistent with France and Spain, which suggests that any correction will be small. (Italy of course has a much higher death rate because of the complete and immediate overwhelming of their medical systems.)