During Minnesota’s extended experience under Gov. Tim Walz’s statewide stay-at-home order, I decided to statistically evaluate the effects of such orders (or their absence) in the 48 continental states. This was especially interesting since Minnesota has been nearly enveloped by four states (the Dakotas, Nebraska and Iowa) that are among the eight without a statewide stay-at-home order.
Per capita rates of COVID-19 deaths and active cases were correlated with various factors that might predict those rates: population density (people per square mile), minority population percentage, median age, average winter temperature, the existence or absence of a stay-at-home order, and the total days of the order.
To account for possible benefits from an activist government, total tax burden was added.
The simplest model for active cases — using only well-known factors — showed that most (56%) of the interstate variation was due to population density, followed by median age and winter temperature. Density is obvious, as it interferes with social distancing; older and warmer populations have fewer active cases.
The strongest model explained 66% of the differences and found that density, age structure and tax burden were the best predictors.
The tax burden was statistically significant with what we will imprecisely call a 99.95% confidence level. Higher taxes were correlated with a higher active case rate. This was true whether or not we excluded the outlier states of Delaware (low taxes but many cases) and Utah (moderate taxes but few cases). This strong correlation persisted even after removing those states (Connecticut, New Jersey, Rhode Island and New York) with high taxes, high density and many active cases.
The best predictor of deaths per million explained 71% of the interstate variability — which is impressive in biostatistics. It simply combined a state’s population density, African-American population percentage, and the interaction of density and temperature (high density was less deadly in warmer states).
Neither tax burden nor median age were factors in predicting death rates.
Importantly, in none of the models was the presence of a statewide stay-at-home order statistically significant. It never helped or hurt.
It is often said that correlation does not necessarily imply causation. But the absence of correlation is strong evidence of a lack of causation.
These results suggest that there is no evidence of benefit when a governor issues a stay-at-home order. Social distancing should reduce (or at least delay) COVID-19 cases and deaths. Why, then, does a stay-at-home order not seem to help?
The primary reason may be that the amount of social distancing achieved is largely determined by population density. The models show that density already explains the majority of the explainable state-to-state differences in active cases and deaths. Where someone lives — think Wyoming vs. New York — has much more influence on social distancing than a directive from the government.
Another large factor is that voluntary social distancing of rational people began before most stay-at-home orders were issued. Seat-belt use provides an excellent analogy. Most of the readers of this piece probably buckle up whether they are seated in the front or back. Some 96% of educated Americans always buckle up. Yet you can be pulled over for failing to wear your belt in the front seat in only 68% of the states; for the rear seat this drops to 46%.
The flip side is the widespread disregard for the orders themselves. A study from the University of Maryland found that only 35% of Americans are personally observing their stay-at-home orders and that these orders changed behavior only in 4.3% of people. Thus, it is hardly surprising that these orders are failing to affect the active case or death rates.
Do not confuse a simple personal stay-at-home order with the shutdowns of businesses — restaurants, bars, professional sports, etc. Compliance with such orders is high, as any defiance is immediately obvious. It is highly plausible that shutdowns of large gatherings and crowded venues made a positive difference — but that is not what was studied here. The question analyzed here was very simple: Does it help to order citizens to stay home (except for essential travel)?
These results are based on a single snapshot in time (May 2) and consider only statewide data. It is entirely possible that different effects will appear with later data or with detailed data from the 3,000 counties of the U.S. Perhaps the more nuanced county-level approach being initiated by Florida and other states will demonstrate some benefit.
Meanwhile, however, the present data show no benefit from statewide stay-home orders. Politicians arguing that they are following the science should be asked to produce the science.
Mark W. Kroll is an adjunct professor of biomedical engineering at the University of Minnesota. The views expressed here are solely his own.