Counterfactual Experiments Are Crucial, but Easy to Misunderstand

Between us, we have more than a century of experience in climate research, literature assessment, and scholarly support for domestic and international efforts to respond to environmental challenges. We have learned the value of rigorous scientific research, even when it challenges conventional wisdom, and of skepticism where it is appropriate. As we watch the response of epidemiologists and public health experts to COVID-19 (e.g., here, here, here and here), we have some idea of the challenges they face. We have seen this play before. Scientists respond to a need to provide information that will save lives; scientists are subjected to political attack for their efforts. We don’t know how the play ends—but we’ve seen enough to know what happened in the second act of the climate change version.

Efforts to understand the behavior of COVID-19 and estimate its future spread began early in 2020. As part of this effort, researchers at Columbia University conducted a counterfactual exercise to answer an important question: What would have happened if nontherapeutic interventions in the United States had started before March 15? According to their calculations, starting only a week earlier, on March 8, could have saved approximately 35,000 U.S. lives and avoided more than 700,000 COVID-19 cases through May 3 (a 55 percent reduction from what happened). Starting interventions another week earlier could have reduced deaths by more than 50,000.

On June 8, Nature published two more counterfactual studies. Solomon Hsiang and colleagues focused on six countries (China, France, Iran, Italy, South Korea and the U.S.) that had imposed travel restrictions, social distancing, event cancellations and lockdown orders. Their calculations, supported by an estimate that COVID-19 cases had doubled roughly every two days starting in mid-January, suggested that as many as 62 million confirmed cases (385,000 in the U.S.) had been prevented or delayed through the first week in April.

In the second Nature study, Seth Flaxman led a group that focused on 11 European countries. They worked with estimated viral “reproduction rates” between three and five; that is, every infected person was expected to infect between three and five other people per unit of time. This number, called the “serial interval”, is estimated for COVID-19 to be roughly four days. Flaxman and his colleagues calculated that 3.1 million deaths (plus or minus 350,000) were avoided through the end of April, but they found that only lockdowns produced statistically significant effects on the number of estimated cases.

Are these high numbers really physically plausible? Yes. The virus is virulent and exponential growth is powerful. Left to its own devices, COVID-19 reproduction in humans increased at a daily rate of nearly 34 percent over the study period. If you were 20 years old and could find a tax-exempt asset that would pay that as an annual return for the next 44 years, then a $1 investment today would allow you to retire with a $3.1 million nest egg at age 65.

All of these results must be judged in their complete and proper contexts. They describe alternative assumptions about the form and timing of a response to COVID-19, leading to different trajectories for cases and deaths attributed to the virus. Each imagined path also involves policy interventions that have other economic and social effects. Ultimately, it is up to decision-makers to consider the implicit tradeoffs between these intertwined impacts, and to make some overall assessment of joint levels of tolerable risk. This is a judgment that they cannot honestly make unless they acknowledge the veracity of what the science is telling them.

We are alarmed that the U.S. president took the Columbia analysis as a personal attack on his handling of the pandemic. “Columbia is a liberal, disgraceful institution,” he asserted. “It’s a disgrace,” he continued, “that Columbia University would do it, playing right to their little group of people that tell them what to do.” The Columbia report was, according to the president, nothing more than a “political hit job.” We are even more dismayed that conservatives equate their feelings about coronavirus models with the “detest” that they feel about climate models.

Let’s return to the play we mentioned above. In climate world, the first act involved doing the science and conducting counterfactual experiments similar to those produced for the virus. Consider, for example, the finding that human activity is the primary cause of observed planetary warming since the beginning of the industrial revolution. This conclusion results from a well-defined set of counterfactual exercises, wherein ensembles of climate models were run with and without greenhouse gas emissions.

In the second act of the climate play, scientists cope with public and political reactions to their findings. We know Act II well. Coronavirus modelers are now living through it. In some countries and in many areas of science, scientific findings are generally accepted, and Act II seems implausible. But in the U.S., science has frequently been dismissed out of hand or ignored—a victim of misinformation campaigns designed by those with personal and/or institutional stakes in the results.

The COVID-19 counterfactuals were not a “disgrace” or “hit job”. They are standard operating procedure—skillful applications of an investigative procedure that is one of the fundamental ways that serious science is performed.

This understanding of the role of science is why we argue that these particular counterfactual studies are so important. They provide rigorously supported insight into the human cost and benefits of decisions that were or were not implemented. The counterfactuals are lessons about the consequences of disregarding warnings that emerge from scientific analysis—including the warning conveyed by the exiting Obama administration, which, based on the best-available science, highlighted the urgency of early, decisive action in the case of a novel virus outbreak anywhere in the world.

The actual numbers of deaths and infections are not the message here. The real news is that they are big and believable, and that ignoring science can be very costly. The blockbuster corollary is that even a little bit of delay (or acceleration) in implementing decisions can matter a lot. It is a profound message that puts climate scientists in the same theater seats with the COVID scientists.

Source link