Thanks for this, I really like the first graph in particular.
1. Isn’t this a faster doubling time than experts are reporting? If so, do you know why experts disagree with you?
2. What about the fact that reported cases =/= actual cases due to carriers with mild symptoms and inadequate testing? How likely is it that the trajectory of actual cases is different? For example, maybe the actual doubling period is 5 days, and early on there are vastly more actual cases than confirmed cases, but then as testing ramps up confirmed cases comes to be more in line with actual cases (and thus also has a doubling period of about 5 days) Not saying this is likely (it seems unlikely to me) but I’m wondering how unlikely it is.
For 1 I’m not entirely sure but a couple of thoughts:
a) Just a week ago there were only 4 counties outside China which had enough cases to really analyse. One of these is Japan which could have misled any analysis. For Italy, S Korea and Iran there were only ~7-10 days of data. Between all that I think any analysis more than a week old would have really struggled to get an accurate picture of global behaviour.
b) Depending on what you’re trying to predict you might look at different sections of the graph.
I am specifically looking at the early outbreak numbers (highest fractional change per day). This is partly because I think this represents more of a natural understanding of the virus and partly because people are wanting to know what to expect in the US/UK etc.
If instead you look at the highest number of new cases per day section of the graph, then in China this happens from days 18-22. In this section the doubling time is 5-9 days. These numbers are less helpful IMO (in these sections the doubling times are constantly increasing so an exponential model won’t work very well) but it might give a farily good guess for what will happen in the next few days.
Between those 2 factors I can imagine other people getting other results.
Googling quickly I notice that at least one paper get similar results to me!
Agreed that #2 could be a big issue. Rapid increase in confirmed cases could easily be due to rapid increase in testing rather than (such) rapid spread of the virus.
What would the graphs look like if they plotted the number of deaths attributed to COVID-19 rather than the number of confirmed cases? In theory the number of deaths should mostly be a lagged & noisier reflection of the number of cases, with less dependence on testing regimes.
This is a great thought and I’ve added a graph in the appendix.
It seems to confirm that early doubling time is lower than commonly reported.
There is no lag between cases and deaths as would have been expected. Any ideas?
Later on the doubling time goes up faster for cases than for deaths. The final 4 points are from after China started including clinical diagnoses in their statistics. Here I just used the number actually tested which I thought would be fine but it’s possible that the number of tests being carried out decreased which would explain the increase in doubling time. However, even before this change there was a bit of a trend upwards so I’m not entirely sure what to make of it.
Note: On days 1 and 3 there were no additional deaths so the calculated doubling time is infinite. In reality we should just adjust the surrounding points up a bit
Thanks for this, I really like the first graph in particular.
1. Isn’t this a faster doubling time than experts are reporting? If so, do you know why experts disagree with you?
2. What about the fact that reported cases =/= actual cases due to carriers with mild symptoms and inadequate testing? How likely is it that the trajectory of actual cases is different? For example, maybe the actual doubling period is 5 days, and early on there are vastly more actual cases than confirmed cases, but then as testing ramps up confirmed cases comes to be more in line with actual cases (and thus also has a doubling period of about 5 days) Not saying this is likely (it seems unlikely to me) but I’m wondering how unlikely it is.
For 1 I’m not entirely sure but a couple of thoughts:
a) Just a week ago there were only 4 counties outside China which had enough cases to really analyse. One of these is Japan which could have misled any analysis. For Italy, S Korea and Iran there were only ~7-10 days of data. Between all that I think any analysis more than a week old would have really struggled to get an accurate picture of global behaviour.
b) Depending on what you’re trying to predict you might look at different sections of the graph.
I am specifically looking at the early outbreak numbers (highest fractional change per day). This is partly because I think this represents more of a natural understanding of the virus and partly because people are wanting to know what to expect in the US/UK etc.
If instead you look at the highest number of new cases per day section of the graph, then in China this happens from days 18-22. In this section the doubling time is 5-9 days. These numbers are less helpful IMO (in these sections the doubling times are constantly increasing so an exponential model won’t work very well) but it might give a farily good guess for what will happen in the next few days.
Between those 2 factors I can imagine other people getting other results.
Googling quickly I notice that at least one paper get similar results to me!
Agreed that #2 could be a big issue. Rapid increase in confirmed cases could easily be due to rapid increase in testing rather than (such) rapid spread of the virus.
What would the graphs look like if they plotted the number of deaths attributed to COVID-19 rather than the number of confirmed cases? In theory the number of deaths should mostly be a lagged & noisier reflection of the number of cases, with less dependence on testing regimes.
This is a great thought and I’ve added a graph in the appendix.
It seems to confirm that early doubling time is lower than commonly reported.
There is no lag between cases and deaths as would have been expected. Any ideas?
Later on the doubling time goes up faster for cases than for deaths. The final 4 points are from after China started including clinical diagnoses in their statistics. Here I just used the number actually tested which I thought would be fine but it’s possible that the number of tests being carried out decreased which would explain the increase in doubling time. However, even before this change there was a bit of a trend upwards so I’m not entirely sure what to make of it.
Note: On days 1 and 3 there were no additional deaths so the calculated doubling time is infinite. In reality we should just adjust the surrounding points up a bit