Do Lockdowns Work?

"Shut it down. Shut it down now!"
--Telco operator (Die Hard)

Interesting study of lockdown effectiveness. The authors considered US states that have ordered mandatory lockdowns (I count 45 state data points on the graph). The independent variable was 'Days to Shutdown,' or the number of days after state COVID death rate hit 1 per million that businesses were ordered to close. The lower the number, the faster the time till shutdown. Some Days to Shutdown are negative because some states issued mandatory lockdowns before death counts had ticked above the 1/million threshold.

The dependent variable was Deaths per Million at 21 Days, or COVID death rate per million 21 days after the initial 1 per million threshold was met.

If lockdowns have been a major factor in preventing COVID deaths, then we should expect to see a positive relationship between X and Y. In other words, the longer it took states to lockdown, the greater the death rate should be.

That's not what we see, however. The slope of the regression line is shown to be slightly negative, with a reported R-squared of about 5%. The authors do not report the statistical significance of their regression results.

The authors suggest that lack of relationship here indicates other variables at work. They report that a regression analysis of per capita death rates vs state population density, for instance, produced a correlation coefficient of 0.44.

European countries were added for context (blue dots). Once again, no relationship between time to lockdown and COVID death rate is apparent.

Although the authors do not reflect on limitations of their study, I can think of a couple off-hand. The data are not distinguished by degree of lockdown, although measures have varied by jurisdiction. For example, Sweden (which the authors do discuss) has implemented relatively light lockdown measures compared to other places. It is interesting that states instituting some of the more draconian measures (e.g., NY, NJ, MI, MA) fare worse on the chart.

Another issue involves why the choice of 21 days after the touching one fatality per million to measure COVID deaths? The authors do not explain how they arrived at the three week reference point. Review of daily fatalities in many jurisdictions does seem to suggest peaks after 3-4 weeks, thus lending some intuitive appeal to the 21 day period. However, supporting rationale here would strengthen the analysis.

Even with those limitations, this study presents some interesting findings. These findings suggest that forced lockdowns are not primary factors in preventing COVID-related fatalities.