This interesting piece manages to ignore the key point. Each of us must choose how much "self protection" behavior to engage in. Do we continue to isolate ourselves? Our propensity to do so hinges on the data we process and how we process it. (1/2) https://www.nytimes.com/2020/07/27/us/coronavirus-data.html?smid=tw-share

In the Chuck Manski tradition, imagine if a researcher could collect a representative sample of 5,000 Americans and survey the same people each week throughout 2020 about their perceived infection risk and death risk that each of them face each week. (2/3)

(3/4) By knowing each surveyed person's location and demographics, the researcher could compare the subjective to the objective risk and see whether they converge over time. By asking each person about their news sources, we can learn about real time Bayesian updating.

Ideally, the research team would also elicit each survey respondent's risk aversion at baseline. Those who are risk lovers and underestimate the virus risk pose the largest dangers to society by under-investing in self protection. How do we correct this externality?

When researchers identify those who are under-investing in protecting themselves and society as a whole from contagion risk, do we nudge them? Do we pay them not to go out? How does applied micro economics inform the real time policy choices over minimizing contagion risk?

The key challenge here in scaling up to society as a whole based on the sample results is whether these "under-investors" (UI) can be identified based solely on their demographics. Targeting incentives is easy if most UI can be identified from a simple rule such as age<25.

Structural IO papers (see BLP) spend plenty of time discussing preference heterogeneity and decomposing such preferences into those components that vary by demographic group and those that vary even within narrow demographic groups. The latter poses the targeting challenge.