**This is a follow-up on a previous post about Covid-19 and innumeracy**

So, what’s a model?

I was not much into model cars or planes as a kid, but I made a few. I had a phase that included extra glue, rubber cement, paints, and brushes. Done well, models are shiny and attractive. Expensive model cars had rubber wheels, and hoods, doors, and trunks that opened and closed. If you popped up the hood, you might even see a tiny, chrome engine glued into a void. Spiffy, especially for a child.

Still, those models were not cars. They lacked functioning engines and electrical wiring. They were also microscopic in comparison to what General Motors was churning out by the thousands. Only a person who had never seen a real car would mistake a model car for an Oldsmobile. Model cars, planes, or sailing ships are only pale approximations of the real thing.

Statistical models are not all that different. They are data-driven efforts to portray complex realities. Instead of plastic, snap-in parts, they are made of variables that must be measured and defined. They are constructed representations of a complicated process like voting or buying a product. Like any construction, they bear the marks of their creators. Choices must be made. Assumptions are necessary. Every model, even the most robust ones, are imperfect representations. There are good models and bad models. There are no perfect models.

These models run on data. If they are trying to project the future, they use past events and information to predict what could happen in the next few days, weeks, or months. If the data is skewed, or the variables are poorly realized, the model will not be very predictive. As more data becomes available, and as the time horizon gets smaller, these models are more effective, but still flawed if they are built on a poor foundation.

There is nothing revolutionary about what I’ve written. This is pretty normal stuff for people who are trained in data analysis. The problem, of course, is that most of our population–including politicians and journalists–have little to no understanding of these models. This helps explain the sometimes rollercoaster reporting on early models and the sneers now aimed at their revisions. It also explains how politicians have reacted differently to the same models.

The University of Washington’s Institute for Health Metrics and Evaluation (IHME) forecast the swath Covid-19 would cut through the United States of America. More than 240,000 Americans were likely to die given current conditions, the IHME warned. Imperial College (London) catalyzed the United Kingdom when it projected more than 500,000 deaths in the U.K. and more than 2,000,000 in the United States.

Those models were built on some clear assumptions. Early versions assumed static policy and behavior in the U.S. and U.K., thereby creating a worst-case scenario. They also used data from Asia and Western Europe and assumed some commonalities across continents and regions. These assumptions were reasonable, but not bulletproof. The models also provided ranges of probable outcomes, going from something that looked more like the seasonal flu to a catastrophe with millions dead.

Politicians and journalists reacted to the models. Both groups were incentivized to focus on the highest projections. The White House provided careful charts that carried the whiff of certainty. Governors across the nation relied on the same data to justify orders that sounded unthinkable then but feel normal now. Social distancing. Essential businesses only. Lockdown. Shelter in place. Safer at home.

Other politicians, like in Sweden, Florida, and New York City (early) took a different approach. They decided to function as normal, to the degree possible, either under the belief that the low end of the ranges were more reasonable, that herd immunity could be achieved quickly, or other factors, like economic vitality, was not worth sacrificing for a “bad flu” season. These may turn out to be poor choices, and they were driven by either competing values or perhaps a misunderstanding of what could happen.* This is why it is vital for politicians, who cannot be experts in everything, to surround themselves with the right people. They must also have the capacity to listen to those experts and make decisions based on information.**

At the most fundamental level, emphasizing the biggest numbers was completely defensible. Even if the outcomes were unlikely, we, the people, needed to grapple with what could happen if we did not adapt to a new regimen. Critics are already arguing that outcomes far below the projections are proof that governments were too hasty and overreacted. These criticisms assume that similar performances could have been possible with less drastic measures. There is no evidence this is true. In fact, the evidence points in the other direction.

As I sit here in Ohio, on April 10, Good Friday, we are starting to plateau. If the trends hold, we should have sufficient beds and ventilators to handle whatever comes our way. Cases and deaths are also well below the worst expectations. Nothing is finished. There could be more waves that land before a vaccine or treatment are available, but Gov. DeWine’s actions have likely saved hundreds or thousands of lives.

Ohio did not become New York City, and we hope it does not. The initial national projections were made in light of the possibility that it could have. There are still emerging hot spots in the United States, and we may have more urban centers that come close to New York. When you think of it from that perspective, the early numbers seem more feasible. But not only did governments react, people did as well. Models can assume social distancing, but they cannot always know how effective it will be. They can make educated guesses based on past information. That is all they can do.

The models are continually being revised in light of new data. As social distancing and sheltering in place exercise a stronger impact, they should be revised down. This is all unsurprising. Again, these models are reacting to changing circumstances. This does not make them wrong. It makes them adaptive. Unless there is evidence that IMHE and Imperial College were acting in bad faith, or used methods inappropriate for the discipline, they don’t deserve condemnation.

Now, we are beginning to hear that government policies should be revised since projections are going down. Perhaps, the reasoning goes, it is time to “open up” the economy. Since cases and deaths are beginning to decline in most areas, the “cure” of social distancing could be worse than the economic diseases of unemployment and failed small businesses. I have a lot of sympathy for these arguments. I don’t think we should assume they are being made for the wrong reasons (only out of political gain, for example). I still think this is the wrong conclusion to draw based on what we know, which may be flawed. We are doing well, and projections are going down, because of what we have been doing during the past several weeks. Now is not the time to stop. We should expect some strict limitations for at least a few more weeks.

From what I can tell, as a pure amateur, for us to return to normal, one of two things will need to happen. 1. We develop or identify a treatment or vaccine that is both widely effective and available. 2. Testing has to be more widespread than it is now and we will have to be willing to identify and isolate people who test positive, maybe even through governmental means.

I have no idea when those could happen, but talk of returning to “normal” before they do seems premature.

*I think Pres. Trump fits into this category of politicians who focused too much on economic costs up front or did not understand the possibilities. His failures in these areas were minimized in some cases by aggressive governors who had the power to act before Trump in key ways. The President’s biggest failure will most likely be in the areas more fully in his hands–ramping up testing through federal agencies or by rallying corporations able to help, using his power to command production of personal protective equipment (PPE) or ventilators much sooner, and to purchase and distribute resources to state and local governments instead of forcing them to compete in a market against each other, thereby driving up prices unreasonably during a crisis. We still have much to learn about all these potential actions, so it is too early to draw hard conclusions.

**There is an easy research design sitting out there. It will try to explain the factors that lead some states to wait while others were aggressive. Was it a function of ideology? Political experience? Tenure in office? Quality of expertise surrounding the decision-makers? I am sure there are dozens of political scientists already constructing that dataset.