Discover and read the best of Twitter Threads about #PredictionWasCorrect
Most recents (6)
New forecast of Florida covid deaths: there will be 80-110 deaths per day on 05 Sept
It's based on applying the age-stratified Case Fatality Ratio to the age of every know case documented in the state's line list
The model code is open source: github.com/mbevand/florid…
It's based on applying the age-stratified Case Fatality Ratio to the age of every know case documented in the state's line list
The model code is open source: github.com/mbevand/florid…
Well, this is going to cause my forecast to underestimate deaths: Florida has a backlog of 1200 suspected covid deaths. A large chunk of them will be reported in the coming days/weeks.
My #PredictionWasCorrect: 105 deaths per day on 05 Sept. Deaths fell exactly in the relatively narrow range that the model predicted (80-110)
For comparison @youyanggu's (excellent) model predicted a much wider range of about 80-190 deaths per day
For comparison @youyanggu's (excellent) model predicted a much wider range of about 80-190 deaths per day
Here is my new forecast of COVID-19 deaths in Florida (updated every 4 days at github.com/mbevand/florid…)
Deaths are still expected to peak around August 6 at 160-190 deaths per day.
Case Fatality Ratio has stabilized at 1.4%
Mean time from onset-to-death is 19.2 days
1/N
Deaths are still expected to peak around August 6 at 160-190 deaths per day.
Case Fatality Ratio has stabilized at 1.4%
Mean time from onset-to-death is 19.2 days
1/N
My first 4 forecasts (permanently archived here: github.com/mbevand/florid… ) have all been remarkably accurate.
2/N
2/N
My #PredictionWasCorrect:
1. deaths peaked on August 4 (7d SMA) at 185/day (I predicted 160-190)
2. deaths then decreased to 130/day as of 21 Aug (I predicted 120-160)
1. deaths peaked on August 4 (7d SMA) at 185/day (I predicted 160-190)
2. deaths then decreased to 130/day as of 21 Aug (I predicted 120-160)
New forecast: deaths will peak in Florida around August 6 between 160-190 deaths/day.
CFR increased from 1.2 to 1.5% because the median age is higher
My previous forecasts have proven very accurate. See open-source modeling code & doc on GitHub: github.com/mbevand/florid…
1/N
CFR increased from 1.2 to 1.5% because the median age is higher
My previous forecasts have proven very accurate. See open-source modeling code & doc on GitHub: github.com/mbevand/florid…
1/N
In addition to charting deaths by reported date (thick black solid curve), I also chart deaths by exact date of death ("deaths by day", thin black solid curve)
Many don't understand that "deaths by day" is always higher than "deaths by date reported" in periods of increase
2/N
Many don't understand that "deaths by day" is always higher than "deaths by date reported" in periods of increase
2/N
On average it takes Florida about 10 days to report 85% of deaths in the "deaths by day" dataset (see my other analysis: github.com/mbevand/florid…).
So my forecast charts only show "deaths by day" up to 10 days before the present day, so as to not be misleading.
3/N
So my forecast charts only show "deaths by day" up to 10 days before the present day, so as to not be misleading.
3/N
By August 12 Florida will report 150-200 covid-19 deaths per day.
My forecasting model is open source and has proven accurate: github.com/mbevand/florid…
This forecast update (2020-07-26) has some major changes:
1/N
My forecasting model is open source and has proven accurate: github.com/mbevand/florid…
This forecast update (2020-07-26) has some major changes:
1/N
I scrapped model 1 (redundant with model 2) and model 3 (was the least accurate of the 5 models).
All my curves are now charted with a simple moving average instead of *centered* moving average. Less confusing.
I now produce a BEST GUESS estimate, pretty tight interval.
2/N
All my curves are now charted with a simple moving average instead of *centered* moving average. Less confusing.
I now produce a BEST GUESS estimate, pretty tight interval.
2/N
Notice how has an extremely wide confidence interval (dotted pink lines): by August 12 it estimates between 70 and 250 deaths/day. Whereas my best guess interval is 150-200.
So how does the best guess estimate work?
3/N
So how does the best guess estimate work?
3/N
Ouch, Japan's second wave of COVID-19 is starting to look brutal.
(The impact on deaths is imminent, because the mean time between onset of symptoms and death is 18 days.)
(The impact on deaths is imminent, because the mean time between onset of symptoms and death is 18 days.)
Of course, "brutal" is relative. Look at the y axis: what's considered brutal in Japan is 500 cases per day.
For comparison, in the US we have 140 times more cases per day...🤔
For comparison, in the US we have 140 times more cases per day...🤔
My #PredictionWasCorrect — deaths in Japan have risen since 18 July:
I see many non-experts misinterpret COVID-19 hospitalization figures
Here's the thing: raw figures for the last 1-2 weeks almost always show a PERMANENT DIP even when hospitalizations are INCREASING
Case in point, my county, San Diego (charts by @SDCountyHHSA):
1/N
Here's the thing: raw figures for the last 1-2 weeks almost always show a PERMANENT DIP even when hospitalizations are INCREASING
Case in point, my county, San Diego (charts by @SDCountyHHSA):
1/N
The animation above shows how the hospitalization chart from the San Diego County Health & Human Services Agency changes over time (current chart: sandiegocounty.gov/content/dam/sd…)
I didn't modify the charts except by adding the red text "Peak hospitalizations"
2/N
I didn't modify the charts except by adding the red text "Peak hospitalizations"
2/N
The animation shows the chart produced as of 5/8, 5/11, 5/12, 5/17, 5/20, & 5/23
We know that, in May, peak hospitalizations occurred on 5/5 but this day isn't the highest peak until the chart of 5/20, 15 days later!
It is because data for the last 1-2 weeks is incomplete
3/N
We know that, in May, peak hospitalizations occurred on 5/5 but this day isn't the highest peak until the chart of 5/20, 15 days later!
It is because data for the last 1-2 weeks is incomplete
3/N
