Don't Count on the Model Prediction for Coronavirus Deaths

For the first time in decades, Americans who hear the word "model" are more likely to visualize a graph than a woman on a runway.  Now, in the era of the coronavirus, we all are morbidly fixated on the projections that the experts are making regarding the number of people who will contract the virus and the number of them who can be expected to die.

On March 31 at the daily White House briefing, we heard from Drs. Fauci and Birx that the most credible model anticipates a final outcome of 100,000–200,000 American deaths due to the virus.  Subsequent discussion repeatedly stressed that the actual number might be much higher or much lower depending on whether the social distancing guidelines are followed.  The more these two highly respected scientists discussed the matter, the more evident it became that the model cannot be relied on to provide assurance about how the pandemic will play out.  This is not a failure on the part of these two credible scientists; it is a failure of the model.

When one does not know the current level of infection in the population, when little information is available about how quickly the virus can be transmitted from one person to the next, when we remain unsure of whether asymptomatic corona carriers are as contagious as those with symptoms, when nobody seems to know how long the average asymptomatic carrier remains in that state, when we are unable to determine the actual mortality rate among the afflicted — when basic pieces of the puzzle such as these have yet to be inserted into the bigger picture, it is unreasonable to expect this particular model to predict accurately.

The American public seems to have been persuaded that statistical models are examples of solid science and therefore deserve to be treated with respect.  This is a dangerous mindset.  Nobody, absolutely nobody — neither scientist nor layperson — should place confidence in the predictive power of any model that relies on assumptions that have not been thoroughly tested.

It is different when models (e.g., mathematical equations) have already proven themselves to be highly predictive.  The model that describes planetary orbits around the sun, for example, is based on established principles in physics that have been tested repeatedly and have been confirmed by the unfailing accuracy of the model predictions.  In this instance, the model is on a solid foundation — and, indeed, the very purpose of the model is not so much to predict an unknown future as it is to confirm the scientific understanding of how gravity works.

But here we are with a demographic model that is unproven, that may or may not end up yielding an accurate prediction of the future.  If this same model had already given accurate predictions for other pandemics, then it might deserve some respect, but it hasn't done that.  In fact, given science's ignorance regarding the attributes and the dynamics of the coronavirus and its transmission pathways, it is obvious that the model could not have been constructed based on an understanding of them.

Instead, the model must use patterns of infection and death that have been tabulated since the epidemic started 3–4 months ago.  Basically, the model has to presume that what happened somewhere else and at some earlier time will happen here in the whole of the United States in the immediate future. 

It is as if you have a dozen black boxes that are generating coronavirus, each to a greater or lesser extent and all of which vary in their output.  They all generate little coronavirus at first, then geometrically more, followed by a diminution over time.  We don't know how these boxes work to generate the virus; neither do we know in advance how much and how quickly each box will emit its coronavirus output.  But even though the black boxes turn out different total amounts of coronavirus and over different lengths of time, they all do so in that same pattern: at first very slowly, then with increasing rapidity, and ultimately tapering off down to zero.  The S-shaped curve of the cumulative coronavirus output is common to all. 

It is not fair to judge the scientific reputability of the coronavirus model that Drs. Fauci and Birx discussed.  We are in a pandemic that demands a response, and the response requires data or information upon which both tactics and strategy can be based.  The model is the best that can be done under the circumstance.  The problem is that people tend to presume that the model is capable of definitive predictions regarding the future of the pandemic.  It is not and cannot hope to be.

It is worth noting that this demographic model projecting how the coronavirus pandemic will play itself out in the United States probably has a relatively small number of key dynamics that — once well understood — could make the modeling job pretty straightforward.  The same cannot be said regarding the models for global warming.  Atmospheric dynamics are far more complex than the pathways of virus dissemination, and some of those dynamics (like cloud formation and dissipation) are not understood at all.  The global warming models are even less capable of accurate prediction than the pandemic model.  We should view them all skeptically.

For the first time in decades, Americans who hear the word "model" are more likely to visualize a graph than a woman on a runway.  Now, in the era of the coronavirus, we all are morbidly fixated on the projections that the experts are making regarding the number of people who will contract the virus and the number of them who can be expected to die.

On March 31 at the daily White House briefing, we heard from Drs. Fauci and Birx that the most credible model anticipates a final outcome of 100,000–200,000 American deaths due to the virus.  Subsequent discussion repeatedly stressed that the actual number might be much higher or much lower depending on whether the social distancing guidelines are followed.  The more these two highly respected scientists discussed the matter, the more evident it became that the model cannot be relied on to provide assurance about how the pandemic will play out.  This is not a failure on the part of these two credible scientists; it is a failure of the model.

When one does not know the current level of infection in the population, when little information is available about how quickly the virus can be transmitted from one person to the next, when we remain unsure of whether asymptomatic corona carriers are as contagious as those with symptoms, when nobody seems to know how long the average asymptomatic carrier remains in that state, when we are unable to determine the actual mortality rate among the afflicted — when basic pieces of the puzzle such as these have yet to be inserted into the bigger picture, it is unreasonable to expect this particular model to predict accurately.

The American public seems to have been persuaded that statistical models are examples of solid science and therefore deserve to be treated with respect.  This is a dangerous mindset.  Nobody, absolutely nobody — neither scientist nor layperson — should place confidence in the predictive power of any model that relies on assumptions that have not been thoroughly tested.

It is different when models (e.g., mathematical equations) have already proven themselves to be highly predictive.  The model that describes planetary orbits around the sun, for example, is based on established principles in physics that have been tested repeatedly and have been confirmed by the unfailing accuracy of the model predictions.  In this instance, the model is on a solid foundation — and, indeed, the very purpose of the model is not so much to predict an unknown future as it is to confirm the scientific understanding of how gravity works.

But here we are with a demographic model that is unproven, that may or may not end up yielding an accurate prediction of the future.  If this same model had already given accurate predictions for other pandemics, then it might deserve some respect, but it hasn't done that.  In fact, given science's ignorance regarding the attributes and the dynamics of the coronavirus and its transmission pathways, it is obvious that the model could not have been constructed based on an understanding of them.

Instead, the model must use patterns of infection and death that have been tabulated since the epidemic started 3–4 months ago.  Basically, the model has to presume that what happened somewhere else and at some earlier time will happen here in the whole of the United States in the immediate future. 

It is as if you have a dozen black boxes that are generating coronavirus, each to a greater or lesser extent and all of which vary in their output.  They all generate little coronavirus at first, then geometrically more, followed by a diminution over time.  We don't know how these boxes work to generate the virus; neither do we know in advance how much and how quickly each box will emit its coronavirus output.  But even though the black boxes turn out different total amounts of coronavirus and over different lengths of time, they all do so in that same pattern: at first very slowly, then with increasing rapidity, and ultimately tapering off down to zero.  The S-shaped curve of the cumulative coronavirus output is common to all. 

It is not fair to judge the scientific reputability of the coronavirus model that Drs. Fauci and Birx discussed.  We are in a pandemic that demands a response, and the response requires data or information upon which both tactics and strategy can be based.  The model is the best that can be done under the circumstance.  The problem is that people tend to presume that the model is capable of definitive predictions regarding the future of the pandemic.  It is not and cannot hope to be.

It is worth noting that this demographic model projecting how the coronavirus pandemic will play itself out in the United States probably has a relatively small number of key dynamics that — once well understood — could make the modeling job pretty straightforward.  The same cannot be said regarding the models for global warming.  Atmospheric dynamics are far more complex than the pathways of virus dissemination, and some of those dynamics (like cloud formation and dissipation) are not understood at all.  The global warming models are even less capable of accurate prediction than the pandemic model.  We should view them all skeptically.