This is my attempt to recreate the model from Imperial College London's COVID-19 Report 26. It can be found here.
The model treats the reproductive number as a smooth function of mobility:
R_t = R_0 exp(-b*M)
... where M
represents mobility data (from Google) that's in [-1, 0]
. In the stan model I've written,
I've made a small chamge to this to make post-lockdown behaviour not be exactly the same as pre-lockdown behaviour.
R_t = R_0 exp(b_0*M - s(t)*b_1)
... where b_1
is in [0, 100]
and represents my belief that mobility post-lockdown will correspond to a smaller
reproductive rate than pre-lockdown mobility of the same level. This model resulted in sensible predictions in some
tests that I ran. s(t)
is a smooth step function that steps up during
April 2020.
The observation model is roughly:
D_t ~ NegBin2(R_t^eff * D_t^eff, d)
... where R_t^eff
is the effective reproductive rate after accounting for the infection-to-death distribution and
D_t^eff
is the effective number of previous deaths that lead to more deaths in the current time period (this figure
accounts for the serial interval distribution - the waiting time between infector and infectee deaths). Refer to the
model doc for details as my understanding can very well be wrong.
Please take results with a grain of salt as they might not be accurate. If you see an issue, please feel free to raise an issue or a PR.