Dear Data Guru and Covid data experts,
As we are going deeper into the Outbreaklocation data (US mainly & at State level), we found some data behaviours looked very interesting or even counter-intuitive:
The (Apple, Google )mobility data & PlaceIQ Exposure shown strong negative correlation to the daily new case at some periods, even considering lags. Attached is a screenshot example for Florida and Apple driving metric, but we found these kinds of cases for all most all mobility & PlaceIQ Exposure metrics at almost all states.
Our current hypotheses are:
Mobility curves looked like the “shock and recover” curves, as people’s mobilities were driven by what they heard from news about the number of the new cases: 1) before April, got ‘Shocked’, then mobility reduced immediately. In later months, as the recent cases number decreasing, people feel relieved or less afraid, so the mobility is recovering
Although we thought the restriction on mobility would have a positive lag effect on new cases (less mobility, then later less new cases), these (Apple, Google )mobility data & PlaceIQ Exposure are not the right data as they are too high-level, rather passive reaction then predictors. So what do you think? Instead of dropping them from the model, any transformation ideas to make use of them?
Another interesting finding is: the LEX data seem to be very making-sense.
We can find a lot of exciting potentials predictors in Lex, for example, for Florida again, LEX Visited Alabama are showing very strong lagging positive correlation to new cases in Florida, looks like if there were more Florida people visited Alabama, then 10 to 14 days later, there were more new cases. (many other simliar examples also found, e.g. Califonia vs Nevada and Arizona)
Of course, would that just be correlation or real causality, we need to verify. But anyone who is local and living in these regions, can provide us with some first-hand and real-world insights?