What are we looking at?
I posted some of these graphs in a comment and I was asked to diary them, so here goes.
Google has made world-wide mobility data collected from cell phones available for download — it’s about 22MB. In the United States, the data set is broken down to county level. The MIT Elections Lab dataset also has voting data in presidential elections down to the county level.
This makes it possible to compare changes in mobility by Democratic voting counties to Republican voting counties in response to the COVID-19 pandemic.
Time Spent At Home
The question naturally arises, is this because Trump voters behave differently, or live in different kinds of counties? Apparently, both:
Here I’ve used the National Center for Healthcare Statistics county classification to separate rural (“noncore” and “micropolitan”) counties from urban counties (containing city of >50,000 and/or suburbs).
You can see that time spent at home has gone up for everyone, but much less so for rural counties. There is a big disparity between urban counties, but rural counties are actually pretty similar to each other. Note the limitation of the Google data: it tells us percent change over baseline, but not absolute amounts. I suspect this rural/urban disparity reflects a much higher at-home baseline for rural counties.
Time Spent at Work
Everyone is spending less time at work; Democratic voting counties more so. This may reflect the fact that Republican-voting counties are more likely to be rural:
Discretionary Shopping and Recreation
This figure does not include grocery or pharmacy, so it’s a measure of time spent partaking in discretionary shopping and recreation.
As you can see, Republican counties have returned almost to their baseline levels of shopping and recreation. This is even more striking when you look at urban and suburban counties vs rural counties. Here urban/suburban counties are solid, rural counties are dashed:
This suggests a little closer look at Republican county behavior by degree of urbanization:
Basically Republican suburbs and tiny (10,000 population) to medium cities all behave similarly and have returned to around 90% of baseline shopping. Major Republican cities have a more pronounced reduction, but not quite as big as their Democratic counterparts.
Compare to Democratic voting counties:
Democratic counties on a class by class basis engage in less discretionary shopping and recreation than their Republican counterparts. Also the change in mobility is a pretty straightforward function of county type: the more urbanized, the less a Democratic county engages in discretionary shopping.
Time Spent at Outdoor Parks
So Republican counties are dealing with the lockdown by doing more discretionary shopping and recreation; what about outdoor recreation? Google provides data on time spent at parks:
Everyone was going to the parks more before the pandemic took off. Democratic park usage initially dropped more than Republican usage in response to the pandemic, but has since risen to outstrip Republican use. Republican county park usage varied less. Note this is *county-wide* and does not necessarily mean Repu blicans and Democrats as individuals acted differently; it may reflect differences in park policies.
Finally we can break this down by rural vs. urban:
Grocery Shopping and Pharmacy
Everybody has to shop for groceries. You can see a stocking-up effect across the board as the pandemic takes off, followed by an across the board drop in grocery shopping time after the shutdowns (around March 23 in most states). Very quickly Republican county grocery shopping returned near-baseline levels. Non-rural shopping is still significantly reduced in urban/suburban Democratic counties.
Caveats
The differences shown here don’t necessarily apply to individual Republicans vs. Democrats *within* a county; I’m simply comparing Trump-voting counties to Clinton-voting counties. The differences between those counties will reflect both individual behavior and local ordinances and conditions.
Also, the raw data from Google is expressed in %change from baseline. Adding averages up is essentially meaningless, so instead I’ve averaged the rates. This is somewhat mathematically dubious; it doesn't tell you what the true average behavior of everyone in, say, all Republican-voting small metro counties, because it ignores differences in population population. However, I do think this gives a rough sense for how different kids of counties’ behaviors have changed, particularly when we look at different *kinds* of counties (e.g. comparing small metro areas to other small metro areas).
If there are any counties of particular interest I can readily generate these kinds of visualizations for individual counties, which is a much sounder approach.