Notes

May 5

Today we introduced a new analysis for the Curve Tracker.

  • Rather than fit each country and state to a reference curve, we are creating a unique curve for each region.
  • This approach allows us now to separately consider each country and remove some of the reporting biases we have seen in the data. 
  • The graphs on the home page show the curves for peak and total deaths. The shaded area indicates uncertainty.

For more information, see our FAQ page.

May 2

  • We have 13 peaks confirmed, with projections one week before the peak giving a mean error of 6% in deaths, and 0 days.
  • Belgium, Turkey, France (hospital), Michigan, and California are following the reference curve closely, with 7-day variance of about 5% or less.

May 1

  • We are working on a new analysis that will be posted soon. Stay tuned!
  • We have added India, Poland and Romania to our analysis.
  • We have 12 peaks confirmed, with projections one week before the peak giving a mean error of 7% in deaths, and 0.2 days.
  • Belgium, Turkey, the UK, and California are following the reference curve closely, with 7-day variance of about 5% or less.

April 30

  • We are working on a new analysis that will be posted soon. Stay tuned!

April 29

  • We have 13 peaks confirmed, with projections one week before the peak giving a mean error of 0% in deaths, and 0.5 days.
  • Germany, Netherlands, UK, New York, and New Jersey are following the reference curve closely, with 7-day variance of 5% or less.

April 28

  • We have 13 peaks confirmed, with projections one week before the peak giving a mean error of 3% in deaths, and 0.6 days.
  • Germany, Turkey, UK, Louisiana, and New Jersey are following the reference curve closely, with 7-day variance of 5% or less.
  • Spain and Italy are now showing an improved reduction in death rates.

April 27

  • Spain has overtaken Italy in having the lowest rate of increase outside of China, though both are improving considerably more slowly than we saw in China. This might be due to the under-reporting of deaths in China, as indicated by the delayed extra deaths recently included in China’s total.
  • Most other countries are currently closely following the mean curve, though NY and MA are improving faster and Brazil more slowly,
  • We have 11 peaks, with projections one week before the peak giving a mean error of 9% in deaths, and 1.3 days.

April 26

  • We have 11 peaks confirmed, with projections one week before the peak giving a mean error of 9% in deaths, and 1.3 days.

April 25

  • US and US State data has been updated
  • We now have 12 peaks, with projections one week before the peak giving a mean error of 3% in deaths, and 0.6 days.

April 24

  • We now have 11 peaks, with projections one week before the peak giving a mean error of 7% in deaths, and 0.7 days.
  • Italy’s slowing does not appear to be further deteriorating. France is catching up with both Spain and Italy.
  • U.S. and state data has not yet been updated, but new data are coming soon.

April 23

  • The US has peaked, but reporting inconsistencies may well be playing a role, with a steep drop in a single day. Hence the US figures should be treated with caution.
  • We now have 11 peaks, with projections one week before the peak giving a mean error of 7% in deaths, and 0.7 days in date. 
  • While Italy and Spain still show slowing improvement, France, Switzerland, Netherlands and Germany are now falling with the mean curve.

April 22

  • Pennsylvania has been removed from the tracker for today because of data consistency issues.
  • Portugal, France hospital, UK, and NJ are following the reference curve closely, with 7-day variance of 5% or less.
  • For the nine countries that have peaked, the mean error rate for peak deaths was 9% and the mean error rate for peak date was 1.5 days.

April 21

  • It has been seven days since the U.S. changed their counts to include probable COVID deaths. Now have the first weekly average based on the new counts. Total deaths in the US are now projected as 92,000, of which 57,000 are in the states we are including in our projections.
  • As countries get further along in the epidemic, we are seeing a slowing in the improvement of their death rates, veering away from China’s trajectory. Italy and Spain, which are furthest along, are currently at 2.3% and 2.5% daily death rate increases respectively. France and Switzerland, who have the next lowest increase in death rates of 3.7% and 3.9%, are also seeing signs of a slowing in improvement. This raises the question as to whether these countries will see death rates fall to effectively zero as was achieved in the deaths reported in China.

April 20

  • Ireland and Mexico have been added to the tracker.
  • US and US state data has been updated using the new basis of confirmed and probable COVID-19 deaths. As a result the projected US deaths have increased from 63,000 to 85,000.
  • We now have seen ten peaks in the number of deaths. Projections made a week before the projected peak are 2% above the actual peak and 0.8 days away, on average. 

April 19

  • France, the UK, and the US are following the reference curve closely, with 7-day variance from the curve of 5% of less. 
  • For the nine countries and states that have peaked, our mean error for peak deaths is 1.05 (5% high) and our mean error for time of peak is 0.9 days.
  • Currently, the US and state predictions do not include data after April 14, when probable deaths were added. When we have enough of the new US data or find a more consistent data source, more current data will be included for the US and states. 

April 18

  • We now have two curves for France, both with and without care home deaths, with total deaths of 15,000 in hospital, and 28,000 in hospital and care homes.
  • For the eight countries and states that have peaked, our mean error for peak deaths is 1.09 (9% high) and our mean error for time of peak is 2.2 days
  • Spain is now only one day behind Italy.

April 17

  • We are removing data from certain countries due to data consistency issues. As well as the change of basis for US data, we have changes for Switzerland and China as well:
    • US and US state data removed after 14/04/2020 as deaths from COVID probable cause were added
    • China data removed after 16/04/2020 as additional deaths added for that day. This has no effect on the analysis as other countries have not processed as far.
    • Switzerland data removed after 16/04/2020 as it appears to be misreported by ECDC.
    • Also, French data is now restricted to deaths in the hospital (Source: https://dashboard.covid19.data.gouv.fr)
  • Death rates for Belgium and Sweden have started rising again, indicating they have not yet reached their peak. For the 7 countries that have confirmed peaks, the mean error rate for peak deaths was 10% and the mean error rate for peak date was 1.6 days.

April 16

  • The US and individual states are not included in this update because they have started including “probable deaths,” making the numbers inconsistent before and after April 14. We will be looking for a source of consistent numbers so the US can be included in future updates. 
  • The UK, Portugal and Belgium now have declining death rates and can be compared to projections of the peak. The projections, made on average a week before the peak, by deaths and date of the peak were:
    • UK: 2% high, 4 days early 
    • Portugal 10% low, 3 days late
    • Belgium 5% high, 5 days early
  • For the 9 countries that have confirmed peaks, the mean error rate for peak deaths was 6% and the mean error rate for peak date was 2.3 days. 
  • Italy’s drop in the fractional growth has been recently noticeably slowing compared to China. The other countries just behind, Spain, Switzerland, France and the Netherlands, are now catching up. This is worth keeping an eye on as it might indicate a second outbreak in Italy or under-reporting of deaths in China.

April 15

  • Pennsylvania has been added to the tracker.
  • Italy, Spain, Switzerland, France, the Netherlands, and Sweden, have confirmed peaks. For these countries, the mean error rate for peak deaths was 6% and the mean error rate for peak date was 2.5 days.
  • Peaks for Germany, Portugal, Louisiana, and California have not yet been confirmed.
  • France, the UK, Germany, Turkey, NY, the US, Brazil, NJ, and MA are following the reference curve closely, with a 7-day variance of 5% or less.

April 14

  • Italy, Spain, Switzerland, France, the Netherlands, Louisiana, Sweden, and Germany have confirmed peaks. For these countries, the mean error rate for peak deaths was 5% and the mean error rate for peak date was 2.5 days. The US, Belgium and Turkey are expected to peak next.
  • The projections of cumulative deaths are stabilizing as time goes on and there is increasing convergence on the reference curve.
  • Germany, Netherlands, Turkey, US, UK, Louisiana, and Michigan and are all following the curve quite closely, with 7-day variance of 5% or less.

April 13

  • Germany, Netherlands, Turkey, US, Canada, UK, NY, and NJ are all following the curve quite closely, with 7-day variance of 5% or less.
  • Projected total deaths for Spain, the Netherlands, the UK, and NY are slightly lower than one week ago. Projected total deaths for Italy and Sweden are higher than one week ago. France and Germany have the same number of projected deaths as one week ago.
  • For the countries that have reached their peak, our predictions of the number of deaths at the peak have been within 2% on average, and predictions of when peak deaths occur have been within 2.2 days on average.

April 12

  • Connecticut and Portugal are now included in the tracker
  • Eight countries have reached their peak, the mean error rate for these countries (estimated vs actual number of peak deaths) is 0.8%
  • The peak date is now marked as a dot on the graph of the projections
  • France, Germany, Netherlands, UK, Canada, NY, and USA are all following the curve quite closely, with variance of 5% or less from the curve over the past 7 days.

April 11

  • Massachusetts (MA) and Canada now have sufficient statistics for inclusion, so we are now providing projections for 19 states and countries.
  • Six of these 19 regions have now a peak count that can be compared to projections: Italy, Spain, Switzerland, Netherlands, France and Louisiana. The average error between all the projections taken between 21 and 7 days before the peak and the actual peak is now 2% for these regions. The same error is also found for projections taken 14 to 7 and 21 to 14 days before the peak, showing the low error is not dependent on the window chosen to average the projections.  
  • The peak counts and dates reported in the composite figure refer to projected data (if the region has not hit its peak) and historical data (if the region has previously hit its peak).

April 10

  • We now have five countries that have reached (at least locally) peak. The projections from 3 to 1 week before the peak had an average error of 7%.
  • The USA, UK, France, and NY are now following the mean rate trajectory extremely closely, with an average variance of just 3% over the last week. 
  • As more countries progress over the peak, the shape of the mean rate curve is being progressively smoothed, as any issues with a data point from any one country is averaged out.
  • The composite of all the rates now shows the end point so you can see more easily that the algorithm is fitting that point to the mean curve, and how far down the curve each country is.

April 9

  • Countries or US states/cities with more than 450 deaths are included when there is a weeks worth of data with cumulative deaths over 100. Iran is excluded due to the very low variance on the data reported. Additions to the tracker include: USA, Belgium, Brazil, Turkey and Louisiana.
  • The variance in the rates between an individual region and the mean curve is now noted for the previous week. This gives an indication of how well the mean curve is being followed.
  • The estimates are based on following from the nearest match to the mean curve from the present day onwards. If the curve is being followed, the date of match should increase by one day per day.

April 8

  • Italy, France, and Germany are still following the reference curve closely.
  • Compared to yesterday, total death projections for the Netherlands and Sweden are higher, while the number of the total deaths projected for Italy is lower. 
  • Total death projections for France, Germany, Spain, Switzerland, and the UK are the same as yesterday.
  • Among U.S. states we are tracking, total death projections for California and NJ are higher than yesterday, while Michigan has fewer projected deaths, and New York’s projection has stayed the same.

April 7

  • We have added an additional country (Switzerland) and three additional US states (New Jersey, Michigan, and California) to the tracker.
  • Italy, France, Germany, and Sweden continue to follow the reference curve closely.
  • Compared to yesterday, projections of total deaths decreased slightly for all countries (except Sweden, which stayed the same).
  • As of today, Italy, Spain, UK, and NY are showing similar numbers of total deaths projected (18-21,000) although their peaks are projected at different times.