Understanding the COVID-19 Forecast

Last Updated on October 28, 2020 by

Answers To Key Questions About Coronavirus Risk…

The COVID-19 coronavirus forecast is a must-read resource from IHME, the Institute for Health Metrics and Evaluation at the University of Washington.

I have been – dare I say? – analyzing their analysis, and would like to share my insights. And this is an appropriate time, because states are shifting from imposing social distancing restrictions to loosening them. The current set of restrictions may be as tight as they will ever get!

I updated the data in this blog on 4/28/2020 in preparation for publication on 4/29/2020.

Note added October 28, 2020: A later blog on COVID-19 Forecast & Data Models gives updated information on the IHME model and describes more recent coronavirus data analyses.

Questions This Blog Answers

The following are questions that bother me, and that this blog answers (to my own satisfaction, at least).

  • Which human activities are the major causes of the virus spreading?
  • Why does the per capita death rate vary immensely from State to State?
  • How do social distancing restrictions affect the ultimate death rate?

There’s a lot of data in this blog. If you’re only interested in the answers, skip down to the end.

COVID Data Is Spotty

Data about the pandemic is uneven, inconsistent and in some cases suspect. IHME’s explanatory article devotes the last two paragraphs of its Discussion section to listing study limitations. IHME has also been criticized for relying heavily on data from Wuhan, China that some believe is inaccurate or biased. Nevertheless, given the many unknowns about the virus, IHME has done a conscientious job of attempting to forecast, and their forecast is the most respectable one currently available.

For the purpose of today’s blog, I am assuming that the IHME forecast is accurate, including its estimate of final death count. One year from now we would have much better data, but one year from now we might all be dead. Therefore for right now, I assert that it’s worth studying the incomplete data that we have.

One strength of the IHME analysis is that it uses data on deaths, not on virus infections. It’s true that when a patient has multiple health problems, it may be difficult to assign a cause of death. Nevertheless, the fact of the death is inescapable. It would be much harder to analyze virus infections, because countries vary widely in the number of people they have tested. And in locations such as the US where test kits are scarce, tests have been limited to seriously ill patients and to persons who were exposed to them.

Which Human Activities Spread COVID-19?

To understand which human activities are most critical to control the virus, I did the following analysis. I listed 13 venues commonly considered as transmission spots. They are shown in Figure 1, along with the entities which they link. Thus, between Homes, important ways that infection spreads are Restaurants & Bars, Schools, Movie Theater, Gyms, and Grocery/Drug Stores. This level of spreading infects the entire City. And between Cities, infections are spread at Concert Venues, Malls/Big Box Stores and Public Transport, leading to infection of a Metro area. And so forth, up to the Global level.

Within one household, if one member gets infected then they all do. Outside the home, the first level of transmission is from home to home, what I call city level.

For example, consider the example of K-12 schools. One child may attend school 5 days a week for 32 weeks a year, for 7 hours per day. This child has 5 x 32 x 7 x 60 = 67,200 minutes per year of exposure. If there are 40 kids in the class, any one of whom may have an infection, that’s 40 x 67,200 = 2,688,000 exposure-minutes per year, which I abbreviate 2688K in the adjoining Figure 1. Of course, any average glosses over a wide range of highs and lows. Fellow compulsives who wish to study my assumptions may read a PDF explanation HERE.

COVID-19 Transmission Venues

Figure 1. COVID-19 Transmission Venues. See adjoining discussion.

            High Risk Activities

There are two important things to grasp from Figure 1:

  • Reading left to right, some venues are orders of magnitude more important than others for spreading the virus. For example, going to restaurants gives you 76 times the exposure to infection compared with grocery shopping! It has to do with how long your group is close to how many other people, and how often people eat out.
  • Reading top to bottom, your city or metro area may have the virus well under control. However, some activities promote national (that is, state-to-state) infection, and others bring the virus to you from anywhere in the world. The U.S. and its states have chosen not to strongly restrict travel. Therefore, our state-level control measures give only limited protection as long as the virus runs rampant abroad.

Yes, I had to apply some judgment. For example, when you spend an hour or more at one location, you may share rest room facilities with strangers, which you likely won’t do during a short trip to the supermarket. When the venue is a restaurant, you’re in a confined space with a hundred people, consuming food that may carry infection from patrons or kitchen staff. I allowed for factors like these in preparing Figure 1.

            The Role of Government Restrictions

COVID-19 Transmission Legend

Figure 2. Legend for Figure 1.

Figure 1 contains the letters A through F in parentheses. They stand for various government restrictions that have been implemented, as explained in Figure 2. We see that specific restrictions target specific virus transmission venues, and that together they potentially address all sources of transmission. However, since the US has a near-absence of type F restrictions, we remain vulnerable to external sources of infection.

Why Do Death Rates Vary So Much From State To State?

One of the remarkable predictions of the IMHE forecast is the state-to-state variation of final death rate per million population. This number ranges from a high of 1230.1 for New York to a low of 17.0 for Hawaii. How can we possibly account for that seventy-to-one range?

COVID-19 Death Data by State

Figure 3. COVID-19 Death Data by State

Figure 3 begins to answer that question. We know so little about COVID-19 that it’s a matter of taste which parameters to consider. The figure includes a few that I considered worth analyzing.

COVID-19 is highly contagious, being spread by coughing, sneezing or even just by speaking. Thus we would expect it to move rapidly through high-density populations.

            Urbanization Spreads Infection

There are various definitions of Urbanization. However, it generally measures what fraction of the population resides in cities. The bottom of Figure 3 shows that there’s a 38% correlation between the Urbanization and Projected Deaths per Million columns. High urbanization means lots of people close together, so more chance for infection to spread.

Some have proposed that poverty is also an important factor. The idea is that poor people have smaller residences holding larger families, have jobs that put them in close contact with other people, and generally have poor healthcare. And as Figure 3 notes, poverty correlates 41% with urbanization.

However, poverty is not a strong explanation for the spread of coronavirus in cities. The correlation between poverty and death rate is only 7%!

However, this is not yet a full answer. Urbanization only ranges from 38.7% (Maine) to 100% (Washington D.C.), which doesn’t in itself justify the 70-to-1 difference between states. But bear with me…

How Does Social Distancing Affect the Final Death Rate?

We can gain another piece of the answer by looking at the social distancing restrictions imposed by the various states, in Figure 4.

COVID-19 State Restrictions

Figure 4. Social distancing restrictions imposed by each State.

As shown, the various State restrictions were imposed beginning March 11 and continuing until April 7. Except for the obvious omission of travel constraints, a majority of states opted for all five of the other restrictions.

What might trigger a governor to announce restrictions in his or her state? The first US coronavirus death that was widely reported was on February 26. (More recent investigation suggests an earlier date of February 6, but that was not known at the time governors were making their decisions.) Therefore, a far-seeing governor might have said, oops, COVID-19 is not a foreign problem, it’s here in the US. So we should try to contain it. Figure 4 shows that California was first to react, although it took them 20.4 days to do so.

Another possible governor attitude might be to wait until his or her state had measurable presence of COVID-19. IHME determined a “threshold” of 0.31 deaths per million population that we may use to normalize “measurable presence” according to state population. According to this measure, West Virginia gets the prize for prompt action, imposing its restrictions an average of 9 days before seeing its first death. In other words, West Virginia looked at the problems in other states, began to see patients in its hospitals with suspected COVID-19, and immediately took action.

            Graphs of Social Distancing Effects

We can gain more insight with x-y plots of the data in Figures 3 and 4. The first plot, Figure 5, posits that state governors might have been triggered to action by the first reported US COVID-19 death on February 26:

Death Rate per Days to Restrict

Figure 5. Final Death Rate per millions Versus Days to Restrict After First US Death.

The second graph, Figure 6, instead assumes that a governor would be more strongly motivated by emerging cases of COVID-19 in his or her own state:

Death Rate per Days to Restrict

Figure 6. Final Death Rate per millions Versus Days to Restrict After State Death Threshold.

What is striking about these two charts is the huge disparity in death rates. Fourteen states are in a class of their own, forecasting 200 or more deaths per million from the virus. Among those states, the data is quite scattered. However, Figure 6 seems to show an upward slope to the fourteen states: in other words, the longer the governor took to act, the higher the final death rate. But we will see that if this effect exists, it’s weak and not a strong indicator.

More important are the states clustered near the bottom, the ones with lower death rates. In this case, there’s an obvious upward slope to the data from left to right, clearly showing that the longer a state took to act, the higher the final death toll the forecast predicts they will experience.

            Social Distancing Affects Final Death Rate

A better way to understand these charts is to sort the data by descending death rate, which we do in Figure 7 below.

Restrictions and Death Rate

Figure 7. Days to State Restrictions and Final Death Rate by State, by Descending Death Rate.

Now, let’s see whether there is a correlation between death rate and how quickly the state imposed social distancing rules.


Figure 8. Correlations Between Days To Restrict and Death Rate.

These results are remarkable! If you include every state, the correlations are negative. That means that the states with highest death rate were ones that acted most quickly!

But the truth comes out when we look more closely. If we leave out the 14 states with death rates of 200 or more per million, we suddenly have positive correlations. These become even stronger when we limit the correlation to states with 100, or 70 deaths per million. In fact, for the 9 states with final death rate below 50 per million, the correlations between death rate and days to impose restrictions are very high!

Low Level Infections in High-Death-Rate States

I will now offer my interpretation of the diverse data shown in Figures 5 and 6.

Those figures suggest that social distancing did not have much effect in the high-death-rate states. A possible explanation of this fact is that these relatively urban states became infected early, likely due to international travel and foreign visitors. By early, I mean January. The infections spread rapidly in restaurants and in public transportation. However, the numbers of viruses transmitted was small. The recipient may have experienced a few more sniffles and coughs than usual, but his or her body was able to combat the disease. And the disease was “invisible” because no one was testing people who had no symptoms.

Within a few weeks, many people in the urban areas became carriers of COVID-19. The real problems emerged when vulnerable people began to encounter mildly infected people, over and over again in the course of the day. Instead of receiving just a few virons which their body could fight off, these vulnerable folks received multiple doses from many encounters. The total amount of infection passed some threshold for the individual and suddenly bloomed out of control, leading to serious disease and death.

Thus, long before governors of these states considered taking action, their citizens were already harboring a huge reservoir of COVID-19 virus. By then social distancing was too late to control its spread, because a large fraction of the population was carrying around the virus, even though they didn’t know it.

            Personal Digression

As it happens, Nola and I may have been among those unwitting hosts of the virus. In January we were vacationing on Maui. Although we had both received flu shots last fall, I came down with flu symptoms. The local urgent care clinic ran a test that confirmed that I had some version of the flu. They wrote a prescription for Tamiflu and gave me a mask to wear for five days so as not to infect other people. Nola, of course, caught it from me. A week later we were both well.

Did we have a weak case of COVID-19 and shake it off? We don’t know, because serological tests are not yet routinely available. However, it’s possible that when we receive tests, we may find that we are overflowing with COVID-19 antibodies. And the residents of all the high-death-rate states (our Michigan is #8) may find that they have the same.

Note added May 19, 2020: Here’s more information:

(1) COVID-19 antibody tests have now become widely available in Michigan, using a high-accuracy test from Abbott Labs. My wife Nola and I got tests and were reported “negative”: that is, we are not part of the hidden population of recovered patients.

(2) On the other hand, random testing in New York City indicates that 20% of all residents have coronavirus antibodies, indicating they already had the disease and recovered from it. This is consistent with research indicating that COVID-19 was circulating in major U.S. cities in early February. These results are consistent with my conclusions about high death rates in urban areas.

Note added June 27, 2020: A New York Times article confirms that, according to epidemiological analysis, the early “hot spots” for COVID-19 in the United States were in fact centers of infection that had not yet evolved into symptomatic cases: “How the Virus Won”.

            The Case of Low-Death-Rate States

This hypothesis, of early low-level infections, could explain another mystery in Figures 5 and 6.

The mystery is the following. Authorities promote social distancing as a way to “flatten the curve” so that COVID-19 infections do not overwhelm medical facilities. However, Figures 5 and 6, and the correlations in Figure 8, show that social distancing also reduces the final death rate. The sooner it is implemented, the lower the death rate.

Why should that be? Well, if the big cities harbored low-level virus infections that people caught and recovered from, that would have primed their immune systems to resist a stronger infection. In less urban areas, social distancing could reduce the intensity of viral infection. Thus people have a chance to catch and recover from a weak case of disease, and become somewhat immune. The result of this “priming” of the immune system could be a reduction in total deaths, and not simply a delay in the deaths.

The Answers to Every Question

We’re now in a position to propose, to hypothesize, answers to the three questions posed at the beginning of today’s blog:

  • Which human activities are the major causes of the virus spreading? Answer: Within a metro area or state, restaurants & bars; concerts & sporting events; and schools & universities. National and global spreading are promoted by driving trips, air travel and foreign visitors.
  • Why does the per capita death rate vary immensely from State to State? Answer: Highly urban states acquired many invisible cases of mildly infected citizens, and those cases spread through the population before anyone recognized the risk. Once those reached some threshold that could overwhelm sensitive people, the deaths were very hard to stop.
  • How do social distancing restrictions affect the ultimate death rate? Answer: For areas that have not yet acquired many invisible cases, social distancing allows a gradual buildup of population resistance as people catch mild infections and recover from them. This reduces the final death rate by conferring some immunity.

I caution you that I am not an epidemiologist. And my conclusions are based on analyzing the IHME forecasts, which themselves contain many uncertainties. But the answers above are what I came up with using the best data available to me.


COVID-19 is a disruptive plague, about which we still know very little. However, I feel confident that a combination of disease testing, antibody testing and vaccination will dispatch it. This will happen more quickly than the fear-mongers think, but not as quickly as the Pollyannas would have us believe.

I encourage you to take heart and stay safe. In addition, guard yourself against “compliance fatigue” that encourages you to take chances!

While we await the enlightenment of COVID-19 understanding, plus release from our home prisons, I hope this analysis provides welcome insight into what is happening around us.

Image credits: All tables and diagrams created by Art Chester





Understanding the COVID-19 Forecast — 19 Comments

  1. Thanks for another great educational read on a sensitive subject Art. I have tried to keep up on the latest scoop ever since “19” started making the news in January, 2020. Your Blog is a wonderful summary of all the salient facts I have read. One of the precautionary things I do, since you mention sun-light has killing power on the virus, is to keep my FACEMASK on the dashboad of my vehicle when driving around. As you know, it is mighty SUNNY and HOT in Texas and I believe I am sanitizing my mask each time I am behind the wheel ! Regards, Joe.

    • Joe, that’s a good idea. It certainly can’t hurt to give the sun a chance to sterilize. Car windows, especially the windshield (which is thicker) partly block UV light, but we know that some UV still makes it through — because in the summer a driver’s left arm tans more than his right one, due to sunlight coming through the window. Thanks for the helpful thought. Disinfecting on the go! – Art

  2. My thanks to Susanna Gordon for alerting me to yet another excellent COVID-19 information site: http://91-divoc.com/pages/covid-visualization/ As she points out, the bottom two charts, which are scaled to population, are especially useful. The buttons below the charts allow switching between logarithmic and linear scales, which as the website points out give different insights. And the pull-down menus below the charts allow choosing which data to see, and allow highlighting a particular country or state of interest.

  3. Outstanding analysis Art! You covered the subject quite well and I hope makes your take on the issue into the COVUS Task Squad.
    We are holding our own in Cedar Hill, TX as we don’t allow Chi-Coms to invade our space ! This virus has been a great concern to me with Kaye being in rehab following her femur break. Extraordinary ground rules relative to not allowing ANY visitors has made a big difference. She is scheduled to be released May 8 after being away from our home 3 weeks. Further rehab will take place at home.
    Hope you and Nola are safe and avoid the bug. Joe

    • Thanks, Joe! And our warmest sympathies to Kaye, who has been enduring not only physical therapy but physical isolation. We’re glad that you both are safe, and that she can continue rehabbing at home where it’s much nicer. Stay safe, and keep away from those giant “murder hornets” (our latest thing to worry about). – Art

  4. Friend Charles South alerted me to a coronavirus forecasting model that seems to be doing a more accurate – although more pessimistic – prediction of deaths by state and by country: https://covid19-projections.com. I particularly like their description of the model and explanation of the reproduction number R that describes how many others each infected person infects (https://covid19-projections.com/about/). If R is less than 1.0, the virus gradually dies out, but if it’s great then 1.0, then…. (The high number they find for New York State suggests an urban density effect in transmission, as this blog concluded.)

  5. Quite interesting analysis, Art – thanks! I’m curious why not include the workplace, which should be somewhat similar to Schools and Universities. Workers, in particular, might tend to go in even when sick, while in many other venues that is likely less so. Another factor to tweak in your analysis is the exposure minutes wherein the attendee is within 6 feet of another person. One doesn’t normally get that close to every attendee. With prudent and common sense use of masks going forward, the 6-ft, unmasked exposure minutes should be far less. BTW – my wife and I also had flu-like symptoms in Dec and Jan, and we wondered the same thing as you … early exposure? Take care, be well, hope the hip recovery is 100% soon.

    • Thanks, Rick!
      Work infection presented me with a dilemma because there’s such a wide range of workplace geometries. But regardless of the traditional degree of exposure that work presents, businesses are now re-thinking how they operate. My son-in-law contractor spent the last couple of days installing plexiglas panels in a car dealership, to partially isolate salesman and customer when sitting across the sales desk from one another. An ad hoc solution to one business owner’s particular case.
      I hope that a combination of business reconfiguration, masks and personal habits will reduce infection rates enough that the virus is not a constant worry, thus things will feel close to normal while we wait for effective vaccination. High-contact businesses like meat packing may find that they can’t operate until their whole staff has caught it and recovered. Ouch!
      But if my suspicion and yours is correct, we may not be terribly far from the date that “everyone” is a recovered patient.
      Thanks for your good wishes. And may you both stay healthy and immune! – Art

  6. This is fascinating and certainly more informative than anything that else that I’ve seen or heard. Thank you, Art, for applying your analytical skills toward illuminating what’s going on in this very puzzling domain.

    • Many thanks, Bill! But I am tempted to quote Mark Twain: “There is something fascinating about science. One gets such wholesale returns of conjecture out of such a trifling investment of fact.” As we well know! – Art

  7. Concerning state-to-state differences, I heard a newscast this evening comparing death rates in Florida to those in New York State. The reporter posited that possible reasons include these characteristics of Florida: (1) More residents living in single-family dwellings; (2) No mass transit (subways) to spread infection; (3) Warmer temperatures, which might suppress COVID-19 virus; and to these I would add (4) More retirees, hence fewer residents spending days close to other people in office settings.

    Although Florida’s “urbanization” measure is less than New York’s, New York City is certainly higher density than most, perhaps all, of Florida. A somewhat different urbanization measure than I used in the blog would directly reflect items (1) and (2) on this list. And it might lead to more than the 38% correlation in today’s blog between urbanization and forecast final death rate.

  8. Thanks Art for an excellent analysis of the available data and for providing an rational perspective on this deadly contagion.

    I welcome your future thoughts as we go forward. Stay safe.

  9. It is sad that our Native American communities (AZ and NM) are experiencing the worst results. Infections, and deaths from Covid 19 are 50% in NM. The problems with hygiene, obesity and diabetes are overwhelming. Almost half of these homes have no running water. Hard to wash hands.

    • George, I agree. The Native Americans suffer from the conditions you mention, and in addition many of them are not near hospital care. Their fatality rate exceeds even that of nursing homes, which are hotbeds of infection and death but for different reasons. Both are deep tragedies and deserve a national effort at improvement. – Art

    • Cheryl, sometimes intuition reveals truths that the data obscures. Yours is apparently in good working order!