Vaccines and Cases: A look at US States
In the course of updating my collection of US based dashboards, I noticed on the dashboard “Key Metrics Vax by County” (4th tab), several states seemed to have strong positive correlation between the adult (18+) population vaccinated and new cases per 100K in the last month. For example, below is a plot showing Ohio. Each dot is a county in Ohio, the x-axis is the % of adults fully vaccinated, the y-axis is the number of cases/100K in the last month.
In the case of OH, the correlation is quite strong with an R-squared of 0.52 and a p-value of <0.0001. I decided to look into all of the continental US States to see if Ohio was an anomaly or if this was a pattern elsewhere. In doing a simple regression like this, we typically define “statistically significant” as having a regression equation with a p-value of < 0.05. With that criteria, I label the states as having positive (vax ⬆ , cases ⬆) or negative correlation (vax ⬆ , cases ⬇) and whether the relationship was statistically significant or not. Here are the results:
The states in grey (Oregon, Nebraska, Missouri, and Indiana) did not have updated county-level cases for the last month in the source I use, so I could not evaluate them. The five states in Blue (Wyoming, Kentucky, Delaware, Massachusetts, Vermont), showed negative correlation (higher vaxxed counties had lower case rates), but none of these were statistically significant. The 13 states in orange represents states where positive correlation (higher vaxxed counties had higher case rates), but were not statistically significant. The red states, 26 in total, were states with positive correlation that was statistically significant!
To summarize, of the 44 states evaluated, 39 states showed higher case rates in counties with higher vax rates. 26 of the 39 were statistically significant. 5 states showed lower case rates for higher vaxxed counties, but none of them were statistically significant.
Feel free to ask about a particular state. Per comment below, here is VT (it’s essentially 0 correlation, but technically negative (thus not statistically significant):