This is a low-stakes post.
Massachusetts has been releasing county-level coronavirus case counts, which I paired with data from the US Census to look for patterns. I actually didn’t end up finding anything particularly interesting, but some of the graphs are nice, so I thought I’d share.
On case growth
A few days ago, it looked like the growth of covid cases in Massachusetts might be flattening. But as of yesterday, it seems like that’s not quite the case across the board. Here are the total case counts per 1,000 residents of each county county since March 15:
And here are cases per 1,000 residents on April 7, with the geometric growth rate of cases over the last week indicated by color:
Berkshire, Barnstable, and Franklin counties have the lowest case growth rates, ranging from 6.2% to 7.3% on average per day over the last week. These counties have some common characteristics:
- They’re geographically remote;
- They are the only MA counties to have experienced population decline over the last decade;
- They have the highest non-Hispanic white populations per capita and the least foreign residents;
- They have the greatest proportion of residents over 65 (at least 22% in each case!);
- Franklin and Berkshire counties have the lowest population densities, at 102 and 141 people per square mile, respectively.
To me, the above is consistent with the idea that economic activity is a vector for the spread of coronavirus (not literally, but it gets people in contact which causes person-person transmission).
Plymouth, Hampden, and Bristol are the counties with the fastest-growing case counts, each of them averaging an increase of over 11% daily over the last week. These counties don’t have much in common, so I’m having trouble putting together a potential unifying narrative.
Race and population density as correlates
It’s starting to look like black Americans might be more susceptible to coronavirus than other racial and ethnic groups. At first glance, that appears to show up in county-level data. But upon closer examination, that doesn’t appear to be the case — first because population density and the percentage of black residents are collinear, and population per square mile has a higher correlation coefficient; and second because Suffolk county (Boston) is influencing the linear relationship in both cases. Adjusted R-squared drops heavily if we exclude Suffolk county from the data set. (Race and population density were the best predictors I found of cases per 1,000 residents.)
This isn’t to say race and its many correlates aren’t good predictors. I think it speaks more to the (severe) limits of the data set I’m working with. If I have time, I may try to build a city-level data set. If anyone knows of one (or something better), link me!