A
study recently published in PLOS One by a team of researchers from University of Louisville, Johns Hopkins University, and the U.S. Department of Defense Joint Artificial Intelligence Center looked at the relationship between temperature and COVID-19. They found 61% of the variation in confirmed COVID-19 cases could be explained by variation in the daily low temperature. That is called an "effect size" and it's a pretty high figure for research of this type.
The research is based on a theory that SARS-CoV-2 virus particles persist longer in colder and dryer environments than in warm, humid ones. It turns out that this is the latest in a slew of studies looking at this question (search "temperature and coronavirus transmission"). There is wild variability in their results, with some researchers concluding there no relationship, others saying there is a modest relationship, with others, like the current one, concluding there is a big relationship.
Because I live in a place with extreme temperatures and I remember what happened last year, I am skeptical of their findings. I decided to have a look at the relationship for the Phoenix, Arizona area during the second quarter of last year.
I got Phoenix low temperature data from Weather Underground, and historical data on new cases per 100K, 7-day average (for Maricopa County), from the ASU Biodesign Dashboard. Here is a plot of cases and low temperatures from March 31 to July 31 last year:
We can see from this that there is clearly no relationship at all from the end of March to the beginning of June. There is a very low number of cases and the number is almost constant, while the low temperature increases steadily.
Then starting on June 6, the cases start to increase along with the temperatures and continue to do so for a month. Here is a scatterplot of the values in that range, including a trendline:
As we can see, the trendline shows an
increase in cases with an increase in low temperatures. I did a regression analysis to get an effect size statistic comparable to that presented in the study. It shows that variation in temperature explains 58.6% of the variance in cases. In other words, these data show
the exact opposite of study with nearly
the same effect size.
To make matters even more confusing, after July 7, the top chart shows cases starting to decrease while temperatures increase. I did not compute the effect size here, but it is probably in line with that presented in the study.
So in this one dataset, we have evidence for no relationship, the opposite relationship to that shown in the study, and the same relationship shown in the study. This is probably why there is so much variation between studies published on this matter.
My conclusion that there is probably no relationship of significance between temperature and COVID-19. What relationships exist are probably due to a
third-variable problem. The classic example of this taught in statistics classes is that there is a huge correlation between ice cream sales and swimming pool drownings. Though we might be tempted to conclude that ice cream causes drownings, that relationship exists only because both are related to a third unobserved variable, temperature.
It's probably something like that in the present case. We know that transmission is increased when people gather in poorly ventilated areas. While cold weather drives people inside in cold climates, hot weather drives them inside in hot climates. That's a likely explanation for the June-July increase in Arizona (and the relationship opposite to the study) in the graph above.
