There’s an interesting new paper today that, using cell-phone mobility data, aims to separate voluntary social distancing from the effects of government policy. It asks: Did people leave their homes less, travel less, or mingle with others less after new policies went into effect, relative to people in places without those policies?
The upshot is that “state and local government policy and informational events induced changes in mobility on top of what appears to be a much larger response across all states to the prevailing knowledge and events at both national and international levels.” That seems about right to me: People responded strongly on their own, but government policy reinforced (and likely prolonged) this trend.
The paper provides a lot of very detailed analysis of specific kinds of policies and local news events, including school closings, emergency declarations, stay-at-home orders, and the earliest reports of nearby cases and deaths. Interestingly, emergency declarations often have a bigger effect than stay-at-home orders, and policies implemented at the county level are often more powerful than those implemented statewide.
To take just one example, the authors look at “mixing,” a measure of “the concentration of devices in particular locations.” At the state level, emergency declarations reduced this measure a whopping 45 percent after 20 days, but none of the other policies had a measurable effect. Yet when the researchers look instead at county-level policies, they get “the largest effects found in our analysis, indicating that first cases lead to an 18% decline [in mixing] seven days out, that school closures and SAH laws lead to an 18% decline, and first death reports lead to an 8% decline.”
Maybe there’s something to this, or maybe there’s a hiccup somewhere. As I wrote last week, because these policies all went into effect very quickly throughout the country, I’m skeptical we can really separate out their independent effects.
To be fair, the authors note this problem and make some compromises in their analyses to account for it:
The first COVID-19 case in a state is easily set apart in timing from the other policies, as is the first COVID-19 death. Emergency Declarations also appear separate. However, School Closures, Gatherings Restrictions, and Restaurant/Business Closings are likely too closely related to be separately identified. Thus, we follow School Closures, knowing that to some degree, the effect of the two other policies may be reflected in those results. Similarly, there is a close correlation between activity on non-essential business closures and SAH policies, although there is more policy activity in SAH laws; we select to follow the latter, as it essentially implies businesses would close too.
But even with these adjustments, they’re trying to suss out the effects of several policies at once — and analyzing them separately at the state and county level! So, take the results with a grain of salt.