Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
New Contributor II
111 Views

CV Mach 2

CaptureB.PNG

0 Kudos
46 Replies
Highlighted
New Contributor II
67 Views

Jim: Thanks for the comment.  I am looking for the harmonics (poor term) under the linear regression in log-linear frame of reference.  

We passed 1000 a day earlier than I expected a week ago, the slope on the underlying linear regression analysis is upward -- very slowly.

HIME - University of Washington estimates 2500 peak in the middle of April -- they I think are wrong -- we should hit 2000 by Sunday 

John

0 Kudos
Highlighted
New Contributor II
67 Views

Capture.PNG

0 Kudos
Highlighted
New Contributor II
67 Views

Last picture is a FFT from some corrupted data -- I thought you would all like to see what random data in a Fortran program can produce. 

0 Kudos
Highlighted
Honored Contributor I
67 Views

Nichols, John wrote:

.. from some corrupted data .. 

Dunno about corrupted data, but one can all say the information on Covid-19 cases and fatalities is all preliminary.  It's all too early to read much into the trends.  The state of NY with highest number of cases in US is approaching about 75 fatalities a day that are attributed to this epidemic.  Recently this state has seen about 300 deaths a day due to health-related causes - cardiovascular, lower respiratory, flu, diabetes, etc.  Now many of these are now attributed to this virus because the patients need treatment as-is and they are picking up the infection in the course of seeking treatment itself.  Going to a medical center/hospital now greatly increases one's chances of Covid-19 infection.

So from a healthcare stats point-of-view, it will take a long time to properly measure, validate, and understand the data.  There's little point in "number crunching" until there is better clarity.  The current information is only useful to educate everyone to take EXTREME PRECAUTION and adopt all the safety measures with this virus, for one's own sake and especially for everyone else's health.  One clearly does not need Fortran nor FFT for that.

0 Kudos
Highlighted
New Contributor II
67 Views

I appreciate your point of view. 

Interestingly, as I sit here doing some Fortran and FFT - after being asked by some health statisticians to help - and they are friends so I decided to help, I was listening to Dr Brix in the daily briefing - she mentioned the log curve and that it is not bending.  This is the curve I place on this site each day as it is of direct interest. Each point of the graph at the moment represents 1000 deaths.  By Sunday it will represent 2000 deaths   But there are 202 countries and so doing the analysis each day in R or EXCEL is burdensome, a Fortran program allows for daily quick updates - I will send it to the people who need it for free.  If no-one uses it -- then I will have wasted some time - but this was interesting and I get to converse with some interesting chaps. .  

The HIME University of Washington said from Monday that the curve would be linear from Monday, it is not it is exponential still.  This is the White House data point 

Dr. Brix concluded today that the massive economic pain is not working as well as they thought, I could have told you that on Monday - the curve is not bending yet.  

-- the reference standard is the Italian data and the Chinese Data. The zip file shows all of the major fatality count countries.  If you look you can see the falsity in the Chinese data, it is being human manipulated on a daily basis -- obvious - compare it to the others.  The Italian data is also wrong because you cannot count all those dead.  

We will never have complete data -- but the death data at the moment is the best pointer to how long  we are in lock down  and the death toll in the USA-- I predict from the data that we will be locked down for most of May. We will know by Sunday if the HIME estimated median peak of 2400 is accurate and I am willing to bet a small sum that they are wrong.  Very wrong..

But I appreciate your point.  I just disagree, which is the beauty of academe - we can be friends and disagree. 

 

 

 

 

0 Kudos
Highlighted
67 Views

Where are you getting your updating .csv files from? The site link posted on your earlier thread (my post) has stale data.

Jim Dempsey

0 Kudos
Highlighted
New Contributor II
67 Views

I get the data from the 

https://www.ecdc.europa.eu/en/publications-data/download-todays-data-geographic-distribution-covid-1...

I started talking to them yesterday -- about some mistakes in the US data. 

I need to check the FFT and move onto modelling. 

 

0 Kudos
Highlighted
New Contributor II
67 Views

Capture.PNG

0 Kudos
Highlighted
New Contributor II
67 Views

Slight drop yesterday -- but that has happened before

 

0 Kudos
Highlighted
New Contributor II
67 Views

The ODE solver from NR provides a simple Fortran program to do the SIR model. 

A couple of interesting papers -- 

If you turn the plain vanilla SIR model loose in the USA with nominal values one ends up with 14 million dead in 80 days. Clearly not correct but interesting. 

0 Kudos
Highlighted
New Contributor II
67 Views

Dear All:

The SIR model can be adapted in a number of ways -- but it lacks a method for estimating the number of deaths ---  the R is all those who are no longer susceptible, but is there not  a subset Rd = alpha * (I )who die and  Rr = (1-alpha)(I)  who recover.   Rd + Re = R 

Rd is the only measured quantity in reality --  S we do not know --- in reality - some people will never be exposed 

Or am I missing something?

John

 

0 Kudos
Highlighted
Black Belt
67 Views

From the point of view of the virus, both dead and no-longer-susceptible humans are effectively dead -- no longer useful in its drive to propagate, multiply and prosper, don't you think?

0 Kudos
Highlighted
New Contributor II
67 Views

mecej4 (Blackbelt) wrote:

From the point of view of the virus, both dead and no-longer-susceptible humans are effectively dead -- no longer useful in its drive to propagate, multiply and prosper, don't you think?

This is an absolute first time, but I disagree with you -- not about the model but about the statistical analysis.  I have no interest in the model per se,  I am interested in the data we can calibrate the model against and the quality of the data.  

Data quality is a huge issue in all of the analysis I do - whether it is earthquakes -- fatalities, bridge monitoring or brick bond failure.  

Let us assume there are 5 levels of data quality -- 1 is poor and 5 is perfect.  The death data might be a 4 - because it has to be undercounted -- there are people in the US who live in backwaters that will have died and it is attributed to something else.  The number of infected people -- we will never know accurately so maybe consider it a 2. 

So taking the best data how do you and I fix the SIR model so it gives me a daily estimated death count -- I do not beleive it is strictly linear?

SUBROUTINE derivs(x,y,dydx)
      INTEGER nrhs
      REAL x,y(*),dydx(*)
      COMMON nrhs
      nrhs=nrhs+1
      dydx(1)=-0.5*y(1)*y(2)
      dydx(2)=(0.5*y(1)-0.2)*y(2)
      dydx(3)=0.2*y(2)
      write(*,'(1x,f10.4,2x,2f14.6)') dydx(1),dydx(2),dydx(3)
      return
      END

 

This is the standard model -- with the factors from COV-19 as presented in the limited literature.  I need y(4) that is the old Y(3) - living recovered YN(3).  

In this case I would really appreciate your opinion, I respect your opinion deeply. 

I have looked at a fair amount of literature thru our library -- the models are good but they are really variants on this simple model or the code is not available - for obvious commercial reasons.  The Imperial College Model is probably the best along with the WHO.  

But the travel data from Wuhan does not match the results coming out now - that is obvious. 

Really appreciate your thoughts on this matter. 

John

0 Kudos
Highlighted
Black Belt
67 Views

My comment was about the model, as was yours, when you wrote: "The SIR model can be adapted in a number of ways -- but it lacks a method for estimating the number of deaths."

You can read details about the SIR model (and other models with elaborations) at, for example, https://en.wikipedia.org/wiki/Compartmental_models_in_epidemiology#The_SIR_model_is_dynamic_in_three_senses . Humans who are incapable of propagating the virus, whether because they are dead or whether because they are not carriers are, effectively, the same in the dynamics of the virus. If you want a model in which deaths (from COVID2, not other causes such as cancer, car crashes, murder etc.) and acquisition of immunity are to be distinguished, the SIR model is not enough. The same Wikipedia article describes a few extensions of SIR. 

In some descriptions of SIR, the name of the compartment is "removed". "Removed" may be by death or by recovery from the infection. 

There are quite a few papers on more elaborate models. See, for example, the 7-variable model in https://www.researchgate.net/publication/339323999_Epidemic_analysis_of_COVID-19_in_China_by_dynamic...

You have probably heard of the Lorenz equations from fluid mechanics. That system has only 3 ODEs and 3 parameters, but one can easily spend months investigating that system. There are several excellent software packages for studying equilibria, bifurcations and so on (AUTO07, HOMCONT, LOCBIF, MATCONT). Even in the chaotic regime of the Lorenz equations, unstable periodic orbits of great beauty may be found. It is often more profitable to study the mutual relations between the dependent variables than the influence of time on each of them.

There is a somewhat similar problem regarding hospital bills for COVID-19 patients without insurance. The US govt. just announced that it will pay for treatment for just this virus for uninsured patients. However, if other ailments are found and treated at the same time, the patient will have to pay for everything except the COVID-19 part. How does one separate the two portions of the bill?

0 Kudos
Highlighted
New Contributor II
67 Views

mecej4 (Blackbelt) wrote:

My comment was about the model, as was yours, when you wrote: "The SIR model can be adapted in a number of ways -- but it lacks a method for estimating the number of deaths."

You can read details about the SIR model (and other models with elaborations) at, for example, https://en.wikipedia.org/wiki/Compartmental_models_in_epidemiology#The_S... . Humans who are incapable of propagating the virus, whether because they are dead or whether because they are not carriers are, effectively, the same in the dynamics of the virus. If you want a model in which deaths (from COVID2, not other causes such as cancer, car crashes, murder etc.) and acquisition of immunity are to be distinguished, the SIR model is not enough. The same Wikipedia article describes a few extensions of SIR. 

In some descriptions of SIR, the name of the compartment is "removed". "Removed" may be by death or by recovery from the infection. 

There are quite a few papers on more elaborate models. See, for example, the 7-variable model in https://www.researchgate.net/publication/339323999_Epidemic_analysis_of_...

You have probably heard of the Lorenz equations from fluid mechanics. That system has only 3 ODEs and 3 parameters, but one can easily spend months investigating that system. There are several excellent software packages for studying equilibria, bifurcations and so on (AUTO07, HOMCONT, LOCBIF, MATCONT). Even in the chaotic regime of the Lorenz equations, unstable periodic orbits of great beauty may be found. It is often more profitable to study the mutual relations between the dependent variables than the influence of time on each of them.

There is a somewhat similar problem regarding hospital bills for COVID-19 patients without insurance. The US govt. just announced that it will pay for treatment for just this virus for uninsured patients. However, if other ailments are found and treated at the same time, the patient will have to pay for everything except the COVID-19 part. How does one separate the two portions of the bill?

1. The research gate article is perfect - thank you -- 

2. Having worked for many Insurance Companies -- you will get a 100% of the bill and can then argue -- and most will lose because they do not have the insurance co (Fed Gov) deep background.  -- I have written many letters for people arguing with Insurance Co, my standard fee was a bottle of Scotch until my Step Mother yelled at me for making my father drink to much. To understand what I am talking about read MAsh goes to Maine and Dr Doggy Moore. 

Perfect thanks 

John

3. 

0 Kudos
Highlighted
New Contributor II
67 Views

Capture.PNG

0 Kudos
Highlighted
New Contributor II
67 Views

Question - is the data for the 4th at the 1000 line - a bend over in the line or a harmonic.  

0 Kudos
Highlighted
67 Views

This is probably a naive observation.

The chart in #17 looks linear, even with discarding the first 15 days or so, which may not have sufficient contributing numbers to be meaningful.

Past this point (15 days), a number of practices have been implemented in an attempt to slow the progression. Ask you self: Why does each attempt at mitigating the spread appear to have no effect? Social distancing, stay at home, wash hands, no touch face, wear face mask, etc.... each of these suggestions have come in at different points in time. Why no change in slope?

Jim Dempsey

0 Kudos
Highlighted
New Contributor II
67 Views

your question cannot be answered today -- but it is the key question -- is this a harmonic or a slope change - I suspect harmonic 

0 Kudos