Share this page!

Maybe Scrolly?
Michael Levitt Profile picture
Twitter logo
30 May, 9 tweets, 5 min read
I am sure that there will be something that correlates with COVID19 death rate better that the stringency of restrictions.

Finding it is important!

The entire world needs to know what to do the next time there is a serious viral threat.
1/8
 Interesting replies led me to carefully analyze data in @youyanggu GitHub table. Alway more objective when one collects data, another analyses it.

Looking at data so first tidy table in Excel.

Color formatting is red-yellow-green low value to high. Image
2/8
 Rather than make plots of one measure against another, we get the correlation coefficient of all pairs of measures.

Correlation coefficient, CC, of A to B is same as CC of B to A so table is symmetric. Correlation coefficient of A to A is always 1; it is whited out here. Image
3/8
 The measures @youyanggu has collected are diverse. I fix names a bit.

We are most interested correlations of the COVID19 Death Rate “deaths per 100k” with other measures.

In top row we see a problem. High correlation of Death Rate & “Death Rate over Flu Death Rate”

No.... Image
4/8
 It is an artifact: the 2 measures are obviously related & so falsely correlated.

Another pair of measures are dependent. “Mar21 Unemployment” & “Apr21 Unemployment”. Their table entries are very similar for all measures.

We make a new table without two grayed out measures. ImageImageImage
5/8
 Now it gets interesting.
COVID-19 seems not depend on any of these measures.

Strongest is weak 0.359 correlation to Flu death rate.
Less correlation to all other measures including:
•Obesity
•Mar21Unemployment Rate
•Mean Temperature

I assume the Death Rate accurate... Image
6/8
 @youyanggu please check.

The many strong correlations (positive & negative) make sense (if needed can explain).

Lack of correlation of COVID-19 Death Rate to all measures except Flu Death Rate is surprising.

Need more & better measures to explain deaths. Please send them. Image
7/8
 Analysis of the level of the signal in each of 18 measures (use the Standard Deviation of Correlation Coefficient with other 17 measures) shows that Death Rate (Deaths per 100k) has least signal.

For COVID-19 Death Rate highest correlation is to Flu Death Rate, CC=0.359. ImageImageImage
8/8
 Presumably “Flu death rate” is an average over previous years?

Flu death rate is higher if more Obesity.

Flu death rate is lower if greater income, higher Percent 25plus with Bachelors degree, and more been vaccinated for COVID-19. Clearly no causation with COVID-19 dose. Image

• • •

Missing some Tweet in this thread? You can try to force a refresh
　

Keep Current with Michael Levitt

Michael Levitt Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @MLevitt_NP2013

Michael Levitt Profile picture
21 Mar
1/7
 Excess death (E) in any period is the difference between the actual all-cause deaths and those that are expected. Expected deaths in the current year, c, can be calculated in many ways. Easiest is to use the data from a few recent years as a reference (we use, 2017 to 2019).
2/7
 Data can be used in 3 ways to calculate expected deaths.
(1) as average death in the reference years.
(2) as average corrected for the change in total population.
(3) as average for each age band corrected for its population, what we call age-adjusted.

We use 5 age bands.
3/7
 (1) If D(i) is death in reference years i, then expected death in year c is E(c)=average[D(i)].
(2) If P(i) is population; E(c)=P(c)*average[D(i)/P(i)].
(3) If (P(i,j) is population of age band j in year i, D(i,j) the corresponding death; E(c,j)=P(c,j)*average[D(i,j)/P(i,j)]
Read 7 tweets
Michael Levitt Profile picture
11 Mar
Thanks all. Link is ynet.co.il/news/article/B…

Your analyses in reply to @AlexBerenson are wonderful.

I worry that the number of deaths (0, 1 & 2 vaccines) averages 46/day for first 6 weeks. Real Israeli deaths then are 171/day, 3.7X more.

Something is wrong. Please help me.
@ynetnews @adiryanko @Blitz20191 @AlexBerenson @MatanHolzer @dvir_a @boulderfish
Sorry I missed you analyses. Just for a little sanity check. In Israel there are ~45,000 deaths each year (about 124/day). This year expected 47,000/year or 129/day. Where are missing deaths?
·
I am wrong as this is just Covid19 deaths. Sorry...
Read 4 tweets
Michael Levitt Profile picture
27 Jul 20
Excess deaths are related to total all-cause deaths. While subject to recording delays, this number avoids all uncertainty about what caused death. Often reported each week, total death can be converted to excess by subtracting a baseline of the total number of deaths expected.
The baseline can be calculated as an average over earlier years or even simply be the totals for a different year. Deaths can be reported anywhere but a larger population makes numbers more certain. The total death is shown here for Europe (350 million) euromomo.eu/graphs-and-maps.
Notice the winter flu peaks that are large in 2016/17 & 2017/18 but smaller in the next two years. There is also a large peak centered at Week 14 in 2020. This is due to COVID19. The excess death from COVID19 is the total number of deaths above the baseline for the entire period.
Read 6 tweets
Michael Levitt Profile picture
27 May 20
Here is my clearer analysis of the Population Fatality Rate (PFR) related to influential predictions by Ferguson et al. 2020. It use data released by the Chinese CDC on 14Apr20 @ChinaCDCWeekly, not full-text indexed by Google @Google but released in The @guardian on 1Mar20. ImageImageImage
Here is the evidence showing Google still does not full-text index this essential report. The Guardian @guardian finally gave CCDC @chinacdc age-range death numbers on 1 Apr. weekly.chinacdc.cn/en/article/id/…
theguardian.com/world/2020/mar… Vulnerability of OLDER people is not even headlined! ImageImageImage
A perceptive reader will ask for the Verity et al., 2020 IFR. My 25Mar report to UK scientific leaders used that data. After normalization to percent, Verity IFR data is identical (0.6% RMSD) to deaths/Chinese_population in Col. F on Tweet1 Excel. My numbers are unchanged. ImageImageImage
Read 4 tweets
Michael Levitt Profile picture
16 May 20
@EdConwaySky @benton328 @phil_luttazi @KielRobinson @CeeMacBee @ScottGottliebMD @CT_Bergstrom @BuckSexton @nataliexdean @sav_says_ @JamesOKeefeIII @NAChristakis Europe's COVID19 Excess Deaths plateau at 153,006, 15% more than 17/18 Flu with same age range counts. Details follow ImageImageImageImage
EuroMOMO euromomo.eu/graphs-and-maps Excess Deaths from 2020 Week 8 now match reported COVID Deaths @JHUSystems perfectly (better than 2%). In earlier weeks the reported deaths were lower. Not sure why? It allows me to do this in depth analysis & comparison with EuroMOMO influenza. Image
Analysis of Europe's Excess Deaths is hard: EuroMOMO provides beautiful plots; data requires hand-recorded mouse-overs. COVID19 2020, Weeks 08-19 & flu 2018, Weeks 01-16 is relatively easy for all age ranges (totals 153,006 & 111,226). Getting Dec. 2017 flu peak is very tricky. This is a composite with fo...
Read 5 tweets
Michael Levitt Profile picture
7 May 20
"Comparative Analysis of COVID-19". Stanford group of 5 is well-supported by @StanfordMed & @NIH. Three helped part-time: Andrea Scaiewicz @andreascaie, Frédéric Poitevin @fredericpoitev1 , João Rodrigues @jpglmrodrigues . Also Francesco Zonta @Cescoxonta (at @ShanghaiTechUni Image
The app was written in Python by Andrea Scaiewicz advised by João Rodrigues. It is research software allowing comparison of all locations with >50 deaths or >3000 cases (see examples). Response is slow as freely hosted by heroku @heroku. Tested too little, it is likely buggy. ImageImageImageImage
The table of locations is classified by Class score. Columns can be sorted or filtered by a string (eg. "====" with quotation selects worst locations). UNSM is raw data, SMO3 is moderate smoothing & SMO5 is more extreme smoothing.
Read 4 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!

Like this author's thread?

Get emails when @MLevitt_NP2013 has new unrolls!

Check out other threads from @MLevitt_NP2013!

Read all threads by @MLevitt_NP2013

:(