3 ms·
A general question to all. Disclaimer, I'm no epidemiologist and not even a statistician. I may totally misunderstand something here. That said, I copied the da
by throwaway_pdp09 7y ago
A general question to all. Disclaimer, I'm no epidemiologist and not even a statistician. I may totally misunderstand something here. That said, I copied the data from the US, UK and Italy on cases (go to https://en.wikipedia.org/wiki/2019%E2%80%9320_outbreak_of_novel_coronavirus_%282019-nCoV%29 https://en.wikipedia.org/wiki/2019%E2%80%9320_outbreak_of_no... and click on the relevant flag in the table, copy the data off the graph, scrub with regexes).
I dumped this into Tableau, graphed it and put an exponential trend line over the top. The UK and US data dropped almost perfectly onto an exponential (R-squared > 0.99 for both, P value better than 0.0001 for both, whatever those mean). The visual fit was remarkably good.
The italian data very obviously did not fit so well. It clearly was flatter than the best fit exponential (strange, given that covid is totally ablaze there) and it had an R-squared of 0.946, though it's P value was still less than 0.0001.
Can anyone comment? The raw data for italy is below.
date,cases
2020-02-21,20
2020-02-22,79
2020-02-23,150
2020-02-24,227
2020-02-25,320
2020-02-26,445
2020-02-27,650
2020-02-28,888
2020-02-29,1128
2020-03-01,1694
2020-03-02,2036
2020-03-03,2502
2020-03-04,3089
2020-03-05,3858
2020-03-06,4636
2020-03-07,5883
2020-03-08,7375
2020-03-09,9172
2020-03-10,10149
2020-03-11,12462
2020-03-12,15113
2020-03-13,17660
2020-03-14,21157
Edit: for comparison, US data is below.
date,cases
2020-02-26,15
2020-02-27,15
2020-02-28,19
2020-02-29,24
2020-03-01,42
2020-03-02,57
2020-03-03,85
2020-03-04,111
2020-03-05,175
2020-03-06,252
2020-03-07,352
2020-03-08,495
2020-03-09,643
2020-03-10,911
2020-03-11,1182
2020-03-12,1577
2020-03-13,2184
2020-03-14,2628
- maxerickson 7y agoA possible explanation is that their testing can't keep up with the spread. ~13,000 test per day for the last several days, with more positive results yesterday for the same number of tests.
- throwaway_pdp09 7y agoIf true that would apply much more to the US data, where I understand the amount of testing has been extremely low. In any case the data is surely of confirmed cases, so these have got past the testing, no? And I'm not sure if that would produce a flatter-than-exponential curve, which seems to be what it is.
- maxerickson 7y agoIt depends on how many tests are needed. If they are running 13,000 and need 50,000, the growth will look flat. And then for the US, if most places are doing a high percentage of the needed tests, then the test results should reflect the real growth. Given the limitations on testing, we likely won't have a good understanding of this for weeks to come (people on the internet saying they should have been tested is only worth so much).
- tunesmith 7y agoIf we had rough numbers of "percentage of tests that are positive" over time, then that would help judge if non-tested cases are growing faster.
- fanf2 7y agoThere are some good charts at http://nrg.cs.ucl.ac.uk/mjh/covid19/ http://nrg.cs.ucl.ac.uk/mjh/covid19/ which show a change in growth rate in Italy near the start of March, possibly due to the lockdown in Lombardy a week earlier