Modeling panic in Austin

We do not have the data science skills to answer a question that troubles us, to wit the extent to which epidemiological models have over-estimated risk of Covid-19 and thereby influenced suppressionist responses from politicians. We can, however, peek at a couple of proximate examples and ask a few questions.

In Austin, the celebrated University of Texas epidemiologists have effectively driven the local policy response to the pandemic, because what politician in this town can say “I don’t believe everything UT scientists say?” Not gonna happen, no way, no how. We do not always, or even frequently, agree with Austin’s Mayor Steve Adler, but our unqualified commitment to intellectual honesty compels us to feel his pain. In this “we believe science and definitely Hook ’em” town, Adler’s got to do what the UT epidemiologists recommend whether or not he privately worries that they might be wrong.

The main page for the UT projections is here. Anybody can scroll around and look at current projections for cases, hospitalizations, and deaths from Covid-19 over time. There also seems to be the archived data necessary to see past projections, but that requires the aforementioned data science skills we do not have in our otherwise vast armamentarium.

We do, however, have a few examples of historical predictions that make us worry that the UT disease forecasters have pushed Mayor Adler and Texas Governor Abbott toward pandemic response policies that were and are more restrictive than necessary or healthy.

On March 24 the team at UT presented this report to the Austin City Council and other learned local officials. On page four one finds a summary of the forecast for the key metrics we were all worried about in late March lined up with the policy response:

The “Austin-Round Rock MCA” includes Travis, Williamson, Hays, Caldwell, and Bastrop counties in central Texas. As of August 17, the aggregate deaths attributed to Covid-19 in those counties over five months was 581 (link) out of a population of roughly 2.2 million (for those of you keeping track at home, that is roughly 9 days of all-cause deaths for a typical US population of 2.2 million). Given that the death estimates in the bottom row of the table above are median projections at various levels of social distancing since March 23, one might conclude that the five counties in question have done a rather extraordinary job of social distancing, having reduced aggregate human interaction by between 75% and 90%.

Then again, one might be an idiot.

We do not believe that outside of a few highly compliant affluent neighborhoods in Austin itself the regional population has reduced its interaction by anything close to 75%, much less 90%, in the last five months. Any late-spring tour of Williamson County (“Wilco” is the second largest in the metro area with more than 500,000 residents) revealed virtually no mask game until the mandate and very little regard for the hazards of the disease or the case counts over which the media obsess. This is not just our impression, but the observation of our astute scientist daughter, who lives in Wilco and has been known to vent to her father on the topic.

Then there is Austin proper. Beginning in late May and for weeks thereafter, Austin hosted massive public demonstrations, which were masked up but not much distanced and naturally flowed in to the reopened bars. Grabbing a few beers after screaming at cops in 95 degrees is, actually, completely understandable and predictable. We were there. We saw no end of cute young people with neo-Marxist signs and colorful hair packing in to our favorite craft breweries after (actually) mostly peaceful demonstrations.

And never mind the insouciant youth in the parks. Egads.

Yet we are to believe that Austin and its environs have been 75-90% socially distant?

Not a chance on God’s green earth.

Now, you might say we should cut the UT team a break, insofar as nobody knew much about SARS-CoV-2 in the third week of March. Fair enough. So how has the UT team done more recently?

On August 10, we captured UT’s death projections for the State of Texas on August 31.

As of August 10, the UT epidemiologists projected that 23,460 Texans will have died from Covid-19 by August 31, within a range of 20,665 to 26,726.

Oops.

To be completely fair, the actual August 31 number will probably rise by 100-200 as the death certificates for the last few days trickle in, but UT still will have overshot its three week forecast for Texas by 80%.

What about more recently? We also grabbed an August 18 screenshot.

Once the UT team got within a two week window, they did hit the forecast range. Giving them full credit for late death certificates they hoisted their chin above “Min Projected” 12,602 deaths. We cannot resist noting, however, that the UT scientists gave themselves a massive 64% difference between that the high end of their range (20,672) and the low end (12,602) a mere 13 days out. Plenty of us dashboard doomscrollers could have hit the side of that barn without all that UT computing power.

Believe science, indeed.

Now, notwithstanding the fun we have had with this post, the serious point is not to castigate these epidemiologists or epidemiologists in general. No doubt they are working hard to improve their forecasts, and indeed they are contributing importantly to the policy debate. However, the demands from the media and the political left to “believe science” ought not actually mean “believe models.” Because many models, including the models much relied upon to inform pandemic response policies, are wildly wrong even when developed by the actual best people.

Models of anything, unleavened with judgment forged in honest and open debate, are not “science” at all. Unfortunately, we have of late had very little honest and open debate, and that is a huge problem.