Note: this page is being updated continuously throughout the day. Check back soon for additional states and countries.
Publisher’s Note: On March we began posting a series of charts by Dr. Douglas G. Frank. We continue that series today, as we will each day until the model proves unsustainable or the crises diminishes. Dr. Frank’s work is part of a long-line of thinkers who look at data, analyze the data into statistics, and use the results to make informed decisions. Our prayer is that it provides comfort and insight.
Dr. Douglas G. Frank
Dr. Frank received a B.A. in Chemistry from Westmont College in Santa Barbara, California. He qualified for his doctorate at the University of California, Santa Barbara before transferring to the University of Cincinnati in 1986 as part of the Ohio Eminent Scholar program. In 1990, he received a Ph.D. in Surface Analytical Chemistry. After graduating, he formed “ADAM Instrument Company, Inc.,” named for the new surface analysis technique he discovered during his graduate studies. The “ADAM” technique brought him international acclaim, and his work was featured in several scientific books and international journals, including cover articles in Science and Naturwissenschaften. He has over 50 scientific publications, and is internationally regarded as an expert in Auger spectroscopy.
Dr. Frank currently serves as the Math and Science Chair at the Schilling School for Gifted Children in Cincinnati, OH. He is also the President of Precision Analytical Instruments.
All words below this line are compiled from Dr. Frank. Other than ordering and organizing the comments and the charts, we (DPH) have not edited any of his comments because that would be out of our area of expertise. The comments were given in a social media context, so they have a friendly, conversational tone.
Dr. Frank’s Commentary on Government COVID-19 Response
Here is how the message goes:
“This is what *could* happen if we don’t implement X.
Ahhhh!!! Everybody do X.”Well, there also *could* be little green men on the back side of the moon.
The scientific and mathematical response to such drivel is:
What is *likely* to happen, based upon the *evidence*?
I’ve had the TV and radio off for over a week. There were just too many *could happen* messages. On all sides. Frankly, it is embarrassing to see my country acting like this. I would be in a really pissy mood by now if I were still listening. (My wife and you guys are keeping me current on the important stuff.)
I think I understand. I think our leaders want to scare us into doing what they think they need us to be doing. (Like scaring your kids to keep them out of the street.) They think they are doing it for our own good.
And while I think I understand this, and appreciate their intentions, I *abhor* this approach. It is dishonest, and in the long run it destroys the credibility of our government officials.
I would much prefer our leaders to tell us the truth, clean, rational, and simple. (Complete with graphs!) They can certainly explain the consequences if we don’t act. I’m good with that. But then they need to trust us to do what is right. And to trust us to demand proper behavior from one another. That is what grown ups do. And that is what moral people do.
Perhaps they are taking this approach because they suspect we are not moral enough to make the right decisions? I pray they are wrong.
Each of us is responsible to do the right thing, or the whole experiment is a mess.
Please pray about it. And get your house in order.
World and USA Model:
Covid-19 “Big US Model Update”
The “Big USA Cases Model” combines all the state models (small peaks underneath) into one BIG peak, then plots the actual total tally of cases in the US (red diamonds) on top of that peak. You can see the daily reports of total cases tracking the big peak for several days.Don’t be *too* impressed by the fit. After all, we updated most of the state models today, and those updates are included in the big peak. So of course the diamonds fall near or on the big peak.Except… and this *really* important… read this slowly: *The big peak is NOT data.* It is the sum of all the state MODELS. In other words, our state models are adding up to match the total cases reported in the US.Our state models are working so well, that when you combine them all, they match the total for the whole country. This is the first day this has happened for the Big Model, and it is easily understood as the *convergence* of all the underlying state models.Our state models are rocking the house, so their sum is rocking the house.This peak is now is the best look we’ve had yet at a “Cases” forecast.
Covid-19 “Quick Look at the US”
Now, if I use my new smoothing algorithm and do a new fit, wow. Some awful pretty stats! Makes sense. The biggest numbers, already past halfway up the peak, new testing protocols are mostly re-stabilized. Some small adjustments. I decreased the final cases tally from 180,000 to 160,000. Also changed the death DOWN to a rate of 0.9%. (Hooray!)
Peak date of this model is 3/27. I like the sound of that.
Several of you told me today that president Trump is claiming that this whole thing is going to be over in early April.
I wonder if he is looking at my graphs?
Covid-19 “US Deaths Tracking”
Just added the data. I have repeatedly said that the deaths graph is the best way to track the progress of the epidemic.
Stickin’ to my guns on that.
Prior to this, we’ve acknowledged that the discrepancy between my initial projections and the data is the situation in New York. So yes, I am working on something (and you are gonna like it). But this graph remains valuable, because it SHOWS us the effect of New York. Why would I want to obscure the graph that illustrates this so perfectly?
I am not changing it today.
Dr. Frank answers the question, “Why will the cases peak rather than continue exponentially?”
State by State (Alphabetical)
Hypothesis/speculation
Reasons the states are now shutting down:
1) They know we have two more weeks to go (i.e. see my graphs);
2) They know the college kids are returning home, potentially spreading secondary infections;
3) They know the spring-break revelers are going home, potentially starting secondary infections.
These are not conspiracy theories, these are reasonable hypothetical explanations. There are more.
Covid-19 “Quick Look at Arkansas”
Just added data. Looks like we’ll be adjusting this one DOWN.
Good news, but also bad news. Arkansas has its first two deaths. God bless them.
Covid-19 “Quick Look at California”
Covid-19 “Quick Look at Colorado”
Once we have data for halfway up the peak, we can usually make a very close prediction. It’s been challenging in lots of states however, because the deaths numbers are so low (noisy) and right in the middle of the most important part for predictions, they changed the testing protocols, messing up the cases data. Maybe the right public health policy, but certainly bad for the scientific method.
Covid-19 “Quick Look at Connecticut”
Just added data. Looks like testing anomaly.
Covid-19 “Quick Look at Florida”
Oooh, baby. Lower cases. We’re saving lives here, folks!
(Not really. The math only describes reality, not controls it. But, oh do I love turning the model down.)
Covid-19 “Quick Look at Georgia
First look, reasonable assumptions. Also a smoothed graph.
Covid-19 “Quick Look at Idaho”
Gotta love those potatoes. And gotta love this graph. First look at Idaho, with typical assumptions. Be safe, Idaho. When all this is over, we all need to patronize our businesses. They are going to need help getting back on their feet.
Covid-19 “Quick Look at Illinois”
Covid-19 “Quick Look at Indiana”
Covid-19 “Quick Look at Iowa”
First death in Iowa.
Model tracking great. Noisy data. Just added points.
Speaking of Fermi calculations and Iowa, I often have my students calculate big numbers. Like how much food it would take to feed the whole world. Turns out, Iowa could feed us all.
We are rich in America. Never forget that.
Covid-19 “Quick Look at Kansas”
I used the new smoothing model, with a good forecast. Kansas’ peak is not narrow. The height (“amplitude”) is low, so there won’t be too much strain on the hospitals, nor much death. So that is good news. But it is looking to be more drawn out. Sorry. The infection(s) must have spread slowly.
Covid-19 “Quick Look at Kentucky”
Kentucky just announced that it had double-counted some cases, and revised their numbers down a tad. Rather than updating the earlier graph, I will leave it in place. However, I took the opportunity to apply my new algorithm to the Kentucky data, and to refine a new prediction. Thank you, Kentucky. It was awful nice of you to adjust your numbers to fit my prediction.
Covid-19 “Quick Look at Louisiana”
Just added data. Tracking. Awesome.
Covid-19 “Quick Look at Maryland”
Covid-19 “Quick Look for Maine”
“Cases” are just dribblin’ along. No deaths. The model is obviously off, which is normal for such small numbers.
Covid-19 “Quick Look at Massachusetts”
Added the data and adjusted up the death rate a sniggle.
Covid-19 “Quick Look at Michigan”
First graph, just added data. This one is going to be fodder for a lot of discussion later. Looks like a dramatic increase in cases… which aligns with analogous increase in deaths. This one is not just a testing artifact.
These folks did not detect their initial infection.
The second graph is my updated projection, *assuming* the new reported cases are correct. Apparently, there has been an issue with reporting.
Covid-19 “Quick Look at Minnesota”
Looks like a testing anomaly threw us off at the start, precisely when it matters most to the forecast. First graph, just the new data. 2nd graph revised.
Covid-19 “Quick Look at Montana”
First stab at it. On track.
Covid-19 “Quick Look at New Hampshire”
Wow were we off today. HOORAY! I’m hoping that this result is real, but it could be just a reporting error or something. So I am just adding the data. No changes.
Covid-19 “Quick Look at New Jersey”
Covid-19 “Quick Look at New York”
Just added data. Changed nothing. Looks like we are overshooting a little bit. Not surprising since we updated it after a massive jump in testing.
Covid-19 “Quick Look at North Carolina”
Covid-19 “Quick Look at Ohio”
Just added data. The model is spooky accurate.
Covid-19 “Quick Look at Oregon”
Covid-19 “Quick Look at Pennsylvania”
Just added data. Model tracking.
Covid-19 “Quick Look at S Carolina”
Covid-19 “Quick Look at Tennessee”
Just added data, model tracking.
Covid-19 “Quick Look at Texas”
Covid-19 “Quick Look at Utah”
Looks like things are going quite well there. Our first stab at a model.
Covid-19 “Quick Look at Vermont”
Updated Vermont using my new smoothing algorithm. Adjusted a bit, but still tracking.
Covid-19 “Quick Look at Washington”
Covid-19 “Quick Look at West Virginia”
Covid-19 “Quick Look at Wisconsin”
Worldwide Models
Covid-19 “Quick Look at Australia”
Covid-19 “Model Update for S Korea”
Looks like the secondary infection in that rest home in S Korea is still playing out. But the cases plot is leveling, so they are on track for it to be over soon. Just added new data. Didn’t change the model at all, leaving it so you can see the secondary infection play itself out relative to the model we set up a month ago.
Covid-19 “Italy Model Update”
Covid-19 “Quick Look at India”
Covid-19 “Model Update for Iran”
I pray that you are correct in your analysis since it is much better than all the projections we see elsewhere and those other projections are being used to drive the behavior and almost complete shut down of this country. I am not educated enough to understand the science behind your models, but I trust you are quite confident of your analysis. However, I have one significant concern about your methodology. Let me explain myself using a football game analogy:
You are predicting the final score of a football game based upon what you see happening on the field. However, you are only watching the second half of the game and you don’t have any idea what happened in the first half. Thus, at the end of the second half, you will know the second half score, but it will only be at that time you will be told what happened in the first half – obviously having a tremendous impact on the final result.
The “first half” represents the lack of testing throughout the country with no idea how many people previous had the virus and either “rode it out at home” or went to see a doctor who may have tested for the flu (with a negative result) and sent the patient home with a simple diagnosis of “undetermined respiratory illness”.
I hope you can tell me I am wrong.
Please post the Texas plot.
Would you please post the various mathematical forms you are using in your regressional analysis.
I forgot to ask.
What software are you using. I though I knew all the gnuplot , etc,etc , but I don’t recognize this app with the fancy shading.
Please see my facebook page, Dr. Frank Models
Regarding the anonymous commenter with the football analogy…what you are suggesting is that we have had perhaps tens or even hundreds of thousands of cases already…that were so mild as to not be worrisome, and were ridden out at home…WHY would you hope you were wrong? If anything, that HELPS the problem instead of hurting it.
The point I was attempting to make was that Dr. Franks models are probably very accurate based upon the information he was using, but he is missing a tremendous amount of information (I.e the “first half”, inconsistent testing procedures from state to state, etc) which essentially makes his projections worthless. He is predicting a US peak of around March 28 with a total of 180,000 cases and 1,400 deaths. Those predictions are about to be blown out of the water over the next week and NYC alone will probably exceed those numbers. Piece and good health to all of us.
“Peace”
When extrapolating to the future, uncertainty is the enemy against accuracy. My background in statistics and reliability begs me to ask for confidence bounds on the extrapolation portion of the models. Please show 50%, 75% , 90% and 95% to show , what I would suspect, is a large variation in the estimates.
Please explain False/Negative test results and what the estimated value could be with the USA tests.