Models Are Not a Panacea

Models Just Don’t Work.

More to the point, models with inputs with a high degree of uncertainty do not generate strongly correlative outcomes—i.e. garbage in equates to garbage out.

There is no better example of the garbage in/garbage out phenomenon than Neil Ferguson’s model that predicted 2.2 million Americans along with 500,000 British citizens would perish due to the Coronavirus.

Ironically, the Imperial College of London epidemiologist was never going to get it right. The reason, at the onset of the Coronavirus there were simply too many unknowns associated with the virus.

In fact, Mr. Ferguson jumped the gun—trying to get out ahead of the crowd.

The trouble with trying to make predictions prior to fully understanding the fundamental inputs and/or their interrelations, the resulting models typically represent worse case scenarios.

This has been the case with the Coronavirus.

With so many unknowns, regardless of what model has been applied—it was always going to be a guessing game as to how many deaths would result from the Coronavirus outbreak.

Even several months in—the death toll remains anyone’s guess.

A distinction should be made between models and the inputs.

The models in and of themselves may be tried and true, i.e. they may be models that have proven valid over time; however, for a model to actually be useful we must have an understanding of the correlations between variables as well as their importance—causal significance.

(Note: Models are mathematical representations of what we know.)

In modeling it’s called the weighting of a variable.

What multiplier/coefficient do we assign a particular variable?

Furthermore, is there a correlation between variables—i.e. does one variable impact another variable.

At the onset, even today, much of this knowledge is still unknown or at least not known with any degree of certainty.

For example—social distancing; what significance and thus weighting should be assigned to social distancing? (Read: The Dubious Causality of Social Distancing)

Is it the main driver in the control of the spread of the virus—does it play a significant role in the ultimate death toll or has its importance been overblown?

Is it any wonder the death toll for America has morphed from Mr. Ferguson’s initial 2.2 million to several hundred thousand to less than 100,000 only to be revised to approximately 60,000 in the first week of April—and now resides at around 150,000.

Perhaps the bigger question: Given just how inaccurate the predictive capabilities of models have been—why bother?

What drives our fixation with modeling?

It can be summed up in one word—uncertainty.

Humans don’t like living with uncertainty—even if that means relying on wholly inaccurate models.

Unfortunately, we can’t know the impact of COVID-19 without knowing the various inputs, their interrelationships and causality.

And because we don’t have a firm grasp on the inputs, there is no way we can, with any degree of certitude, predict how the Coronavirus outbreak will impact America or the rest of the world.

Still, we become fixated with models, which is especially true in times of crisis.

Unfortunately, you simply can’t predict what you don’t know.