Currently, the underascertainment estimation function propagates uncertainty in the real-time CFR, but not the baseline CFR used to estimate how many cases are being ascertained (under assumed all deaths are known).
A better (but more computationally intensive) approach would be to instead sample both from the uncertainty in the real-time CFR estimate and the baseline CFR (perhaps also including uncertainty in known outcomes, see issue #154 ). This could then return a posterior that better reflects the uncertainties across all inputs.
If the literature only includes a mean and 95% estimate of baseline CFR, then extraction functions in {epiparameter} could be used to convert this into a normal distribution that could be sampled from.
Currently, the underascertainment estimation function propagates uncertainty in the real-time CFR, but not the baseline CFR used to estimate how many cases are being ascertained (under assumed all deaths are known).
A better (but more computationally intensive) approach would be to instead sample both from the uncertainty in the real-time CFR estimate and the baseline CFR (perhaps also including uncertainty in known outcomes, see issue #154 ). This could then return a posterior that better reflects the uncertainties across all inputs.
If the literature only includes a mean and 95% estimate of baseline CFR, then extraction functions in {epiparameter} could be used to convert this into a normal distribution that could be sampled from.