We have a new paper published in EPJ Data Science, on quantifying the economic impact of disasters on businesses using human mobility data via Bayesian structural time series model.
Although the research is not exactly about COVID, I think the method is applicable to COVID research, so please allow me to share here.
I think there are many possible next steps to this work, please let me know your comments / suggestions, or any questions.
These results are very interesting, and seem like a great use of the data! In application to COVID, the sticking point as always for natural-experiment designs is picking a control group (or control group candidate set, as in BSTS, or frequentist synthetic control for that matter). Nobody’s really untreated during COVID, so identification relies on timing variation, which is rough given that things are constantly changing and treatment lag timing is unknown! But it would be interesting to think through where the design maybe could apply. Estimating rebound speeds for a lot of things would be valuable.
@Nick_H-K_Seattle_University Thank you for your message! I’m sorry I missed your message and for the late reply.
You make a great point; COVID is affecting literally the entire world and we’re not able to get the control group unlike regional events (e.g., natural disasters). Estimating rebound speeds would be definitely interesting; looking at differences between industries, regions, and socio-demographic and economic groups could be valuable information for policy making!