Dynamic Estimation of Epidemiological Parameters of COVID-19 Outbreak and Effects of Interventions on Its Spread

Hongzhe Zhang; Xiaohang Zhao; Kexin Yin; Yiren Yan; Wei Qian; Bintong Chen; Xiao Fang

With the aforementioned framework, we need to estimate parameters Θ H and Θ f . Instead of using the frequentest approaches (such as maximum likelihood estimation or MLE), we develop an Bayesian approach for our problem because of the following considerations. First, the Bayesian approach allows us to incorporate existing knowledge on COVID-19 to give a guided estimation of Θ H through well-informed prior selection, while the MLE ap-proach would have to largely ignore the valuable information from prior literature. Second, the posterior distribution, given our proposed modeling strategy and prior, has clear interpretation and can provide straightforward uncertainty quantification. To our knowledge, the MLE approach for our specified model settings has no well-developed inference theory for the estimators. Third, from a practical perspective, our Bayesian sampling scheme (described in the subsection of Parameter Estimation) for the posterior distributions is straightforward to derive and implement, while the MLE estimator is more computationally involved and difficult to obtain.