Improved estimation of time-varying reproduction numbers at low case incidence and between epidemic waves
Fig 7
COVID-19 transmission in New Zealand.
We compute smoothed and filtered reproduction number estimates, (red) and
(blue) respectively, from the COVID-19 incidence curve for New Zealand (available at [43]) in the left panels. We use EpiFilter with m = 2000, η = 0.1, Rmin = 0.01 and Rmax = 10 with a uniform prior distribution over the grid
. The top of 7A shows conditional mean estimates and 95% credible intervals for
(red) and
(blue). Vertical lines indicate the start and end of lockdown, a major intervention that was employed to halt transmission. The additional ‘future’ information used in smoothing has a notable effect. The bottom of 7A provides smoothed one-step-ahead predictions
(blue, with 95% credible intervals) of the actual reported cases Is (black). The inset gives the estimated probability of Rs ≤ 1. We observe a clear trend of subcritical transmission that eventually seeds a second wave by August. In 7B we compare EpiFilter with EpiEstim (using weekly windows) and APEestim (both with Gam(1, 2) priors) with all left subfigures presenting Rs estimates and right ones providing filtered Is predictions. We observe that both APEestim and EpiEstim lead to largely unusable estimates that mask transmission trends, in sharp contrast to EpiFilter.