• Assessing the Evidence from COVID-19 Deaths: A Q&A with Christopher Knittel

    Recent research by Professor Knittel and his coauthor sheds light on the important question of who is at greater risk of dying from COVID-19. It also raises an even more important question: Why?

    What groups are more likely than others to contract COVID-19, the disease brought on by the novel coronavirus? And what factors tend to put people at greater risk for contracting the virus?

    These questions are the focus of What Does and Does Not Correlate with COVID-19 Death Rates, a study coauthored by MIT’s Christopher Knittel, an applied economist and the George P. Shultz Professor at the Sloan School of Management, and his colleague from MIT, Bora Ozaltun. In this Q&A, Professor Knittel, who is an Analysis Group affiliate, discusses what the data can and can’t tell us about pandemic-related deaths, untangles questions of causation and correlation, and clarifies the relationship between studies like this and policymaking.

    Broadly speaking, how did you conduct the study?

    Christopher R. Knittel - Headshot

    Christopher R. Knittel: George P. Shultz Professor, MIT Sloan School of Management; Director, Center for Energy and Environmental Policy Research, Massachusetts Institute of Technology; Co-Faculty Director, The E2e Project

    We gathered county-level COVID-19 death rates for the period between April 4 and May 27, 2020. Then we used standard statistical tools to measure how strongly certain variables – socioeconomic factors, health variables, commuting modes, and climate and pollution factors – were correlated with death rates. And we analyzed these correlations both within and across states. I should note that we excluded New York City from the analysis, so as not to skew average results for the rest of the nation.

    Throughout the study you’re careful to use the word “correlation,” rather than “causation.”

    Yes, that’s a critical distinction. Our study does not say that the presence of one or another of these factors causes an individual to be more or less likely to die from COVID-19. It only examines the ways in which the death rates and the prevalence of these variables move together, as it were.

    Which of the results struck you as especially important?

    Race was a central factor. When we compared counties both within and across states, we found that those with a higher African American population had correspondingly higher mortality rates. For example, if you live in a county in which African Americans make up 87% of the population (which is the highest share we found), the COVID-19 mortality rate is 10 times higher than a county with no African Americans (a 0% share).

    Those differences are dramatic, and certainly alarming. Do you think that that correlation could be driven by income inequality, rather than race?

    No. One of the strengths of our statistical modeling is that it can control for a variety of other factors. When it comes to the correlation between race and death rate, we controlled for factors such as income disparity, whether or not people had health insurance, diabetes, countywide poverty and smoking rates, and others. So the correlation that we found cannot be driven by those factors.

    Did any other variables emerge as especially important?

    We found a strong, and troubling, relationship between commuting in any form and death rates from COVID-19. For example, the percentage of people who use public transportation varies in the counties we studied between 0% and just under 21%. Our analysis found that a 20.6% increase in public transportation use was linked to a nearly tenfold increase in the mortality rate during this pandemic.

    We also found that all modes of commuting, public transportation and otherwise, were linked to higher rates of COVID-19 deaths. In other words, telecommuters and others who can avoid local travel are at lower risk. If this relationship is causal, it may be because they are better able to practice social distancing.

    Can you draw any conclusions as to what’s driving this correlation?

    Again, while correlations don’t prove causation, these two facts are consistent with a scenario in which the use of public transportation itself is the cause of the higher death rate – for example, because of the increased exposure to the coronavirus by public transit workers. And the interaction of those workers with other people who are commuting to work – by driving, walking, or biking – is a plausible explanation for the higher mortality rate among commuters in general.

    “Our hope is that the analysis we’ve presented can allow policymakers to focus on a narrower set of potential causal links – and, in so doing, improve this state of affairs.”

    –Christopher Knittel

    As we noted, you’re clear that your analysis finds correlations, not causes. But can a study like this still be of use to policymakers who are grappling with what steps to take to make their constituents safer?

    Even though we don’t make causal statements in our study, we think the results can still inform policymaking; indeed, understanding what is driving these correlations is an important public policy directive. If, for example, our thinking about public transportation use is correct, it would point to a greater need to disinfect public transportation systems, and to socially distance while using them. It may also help inform decisions about the pace of reopenings from state to state.

    In addition, let’s recall the correlation we found between race and COVID-19 death rates. A natural instinct among policymakers is to chalk up this link to something else – often, income disparity. By controlling for this factor and many others, we’ve shown that this correlation cannot be driven by these other variables.

    In other words, the reason why African Americans face higher death rates is not because they have higher rates of uninsurance, or poverty, or diabetes, etc. So it necessarily must be some other mechanism – for example, it could be because the quality of their insurance is lower, or the quality of their hospitals is lower, or some other systemic reason.

    Our hope is that the analysis we’ve presented can allow policymakers to focus on a narrower set of potential causal links – and, in so doing, improve this state of affairs. ■