The United States Just Took the Marshmallow Test. And Failed.
Do you remember the Marshmallow Test?
It’s a famous experiment run by Stanford psychologist Walter Mischel to test young children’s ability to delay gratification. The researchers placed kids in a room with a treat and gave them a choice: they could eat the treat, or they could wait fifteen minutes, at which time they could exchange it for a more desirable treat. It’s amusing to watch the kids as they struggle to muster the required will power.
Approximately one-third of the kids in the study were able to make it the full fifteen minutes. The self-control necessary to wait for the better treat turns out to be predictive of future success. The toddlers who succeeded at the Marshmallow Test scored higher on their SATs as teenagers.
As Paul Krugman has pointed out, we had our own national version of the Marshmallow Test this spring. And, sadly, we made it about halfway, then we gave up and gobbled the damn thing down.
By rushing to re-open the economy before fully suppressing transmission of the coronavirus, we chose the “treat” of a partial return to normalcy in May and June over the opportunity for a more complete and permanent re-opening that would have been possible around July 1. As a result, we will be dealing with more sickness, more death, and a semi-functioning economy for as long as it takes to produce and distribute an effective vaccine.
The Impact of Lockdowns on Infection Rates
The dramatic effect of stringent lockdowns and shelter-in-place orders on reducing the spread of the virus was evident by the middle of April. The tall curve on the graph below shows the projected number of cases when I first ran my projections on March 27 (The data covers the 85 U.S. counties initially hardest hit by the virus). The short curve shows the revised projections, using the exact same methodology, on April 16.
By mid-April, it was clear that the Effective Reproduction Rate was falling sharply in most of the country and was already below one in many places. Continuing on that path would have lead to the full suppression of the virus in 2–3 months. At that time, when I extrapolated the existing trends, I forecast for June 30 approximately eight daily new transmissions in Miami-Dade, four in Los Angeles, two in Dallas, and 49 in New York, to give a few examples.
When I again updated the analysis in early May, the trend toward suppressing the virus had accelerated. Had we stayed the course, just about all the largest cities in the U.S. would be close to COVID-free by now.
For those who doubt this was a sustainable path, take a look at the outcomes in New Zealand, Finland, Norway, Latvia, Ireland, China, Denmark, Hungary, Greece, Netherlands, Belgium, South Korea, Croatia, Switzerland, Austria, Sweden, Italy, Australia, Canada, and Germany. All these countries are adding less than 500 new cases a day, compared to the 50,000 and growing in the U.S.
But to earn this very desirable “treat,” we had to forgo the immediate temptation to re-open the economy before the end of June. Unfortunately, the temptation proved too great. As you can see from the sampling below showing when various states let their stay at home orders lapse, over a 45 day period this spring most of the country reopened:
· Colorado, April 26. (Closed bars again July 1)
· Georgia, April 30.
· Alabama, April 30.
· Texas, April 30. (Closed bars again July 3)
· Florida, May 4. (Closed bars again June 26)
· South Carolina, May 4.
· Nevada, May 9.
· Wisconsin, May 13.
· Arizona, May 15. (Closed bars, gyms, movie theaters June 29)
· Louisiana, May 15.
· Massachusetts, May 18.
· New York, May 28.
· Illinois, May 29.
· Ohio, May 29.
· Michigan, June 1. (Closed bars again July 1)
· Pennsylvania, June 4.
· New Jersey, June 9.
Why the Impact of Reopening Was Not Initially Apparent
In hindsight, it’s obvious that the Effective Reproduction Rate has increased as a result of these openings. For the first month or so after the reopenings began, however, the impact stayed hidden. It was easy to believe in early June that the reopenings didn’t change much. Let’s explore why.
The delay in recognizing the impact of reopening stems from two complimentary factors: 1) the pattern of geometric growth, and 2) time lags in testing results.
First, geometric growth. This simple graph shows what geometric growth looks like. It starts slowly, then ramps up fast.
When the Reproduction Rate is greater than one, transmissions of the virus resemble this graph. A simple hypothetical example will demonstrate how this works. Assume there is approximately one week between “generations” of transmission. So if 100 people are infected on day 1, the people who they infect will, on average, become infected on day 8.
Imagine a county in which the level of transmission is stable — the Effective Reproduction Rate is one. Each week, one hundred people are infected, and they infect another one hundred people the next week. Suppose on May 1, certain local restrictions are lifted, and those changes increase the Effective Reproduction Rate to 1.3. In this example, by May 8, there will be 130 new infections. Each of those people will infect 1.3 more, so that by May 15, there will be 169 new infections, and by May 22, 220 new infections.
In addition, we need to consider the time lags in testing results. The time between infection to symptom onset averages five days. Once symptoms appear, someone needs to become sick enough to decide to be tested. Then they need to wait for their test results. All told, it takes roughly two weeks between an infection and an official addition to the case count.
So the official case counts will look like this:
In this example, it’s easy to see why, even in late May, one might imagine that reopening had little effect. Especially because, in the real world, the case counts are never consistently 100. They bounce around in random patterns, and it takes a while to discern a trend. By the time the trend becomes obvious, it’s too late to stop it. As I’ll show in some charts below, this pattern is exactly what’s happened in quite a few places. (The clues were there earlier, as I pointed out in my last article in May, but it was hard to see them amidst the continuing decline in overall case counts.)
Geographic Differences Are the New Norm
Early on in the pandemic, every county I tracked followed the same basic pattern: a monotonically decreasing growth rate in daily new cases. In some counties the growth rate declined faster than average, while in others it declined slower. For example, in Eagle County, Colorado, the rate of growth in cases was consistently below the median from March through May.
In contrast, in New York City, the growth rate was well above the median. But in both New York and in Eagle, the growth rate steadily declined.
Two factors accounted for most of the difference between those counties whose growth rates were above the median and those whose were below it: population size/density and weather. Because populous, highly dense areas offered more opportunity for transmission, it was natural that cities like New York and Chicago would see more rapid growth than smaller cities like Milwaukee or St. Louis.
We didn’t know at the time, but it’s clear now, that weather also plays a major role. Because most transmission occurs indoors, southern cities, where in March and April the weather is conducive to outdoor activity, saw slower growth. The hidden impact of weather made a number of southern governors overly self-congratulatory about their states success in mitigating transmission. Their overconfidence came back to bite them, as they misjudged the risks of reopening their economies too soon.
Back in March and April, there was considerable speculation about whether COVID-19 would recede over the summer as typically happens with the flu. The focus then was on the virus’ ability to survive in hot, humid climates. As it turns out, it’s not the virus’ behavior that matters — it’s people’s. When people spend more time indoors, they spread the virus more. As we’ve moved into June, the weather factor has reversed itself: people in the north are spending more time outdoors, while people in the south retreat into the air conditioning.
The combination of the changing effect of weather and the more aggressive reopening in southern states has led to a sharp divergence in outcomes across the country. While cases counts have flattened or are ticking up quite slowly in most northern cities, case counts have skyrocketed across the south. The charts below reflect the dynamic demonstrated in the hypothetical example above. Note how the southern counties, Miami-Dade, FL and Maricopa, AZ, have seen sharp increases since early June while the northern one, Cook, IL, has not.
It’s clear that across the south the pattern will get worse before it gets better. While some governors have begun to re-impose certain restrictions, it will be a while before those restrictions begin to re-flatten the curve. And the reversals may not go far enough. It will be a few weeks before we know whether closing bars and certain other specific venues will be sufficient or full re-imposition of stay-at-home orders will be necessary.
What’s Next?
I don’t know. The methodology I’ve used to make my forecasts until now relied on the assumption that the pattern of transmission rates across the U.S. would follow similar paths, even if the growth rates from county to county might be sharply different. That assumption proved true enough for the first three months, but it’s no longer accurate. From now on, we can expect to see certain counties on the decline while others, which had previously been declining, have flare-ups. Those patterns will be impacted by changes in weather and government regulations, and perhaps by cultural shifts (mask wearing, for example) as well.
But it is safe to say that, at this point, all of the country has reopened to some degree. No governors or states are pursuing the stringent lockdowns that would be required to fully suppress the virus. So it will be with us, with transmission rates likely growing and shrinking in continuous waves, until an effective vaccine is widely available.
