The effective reproduction number (R t ), a crucial concept in epidemiology, can be used to detect temporal changes in disease transmission and evaluate the effectiveness of intervention policies. However, different calculation methods lead to varying R t estimates. This diversity of available approaches naturally raises the question of which method provides the most reliable estimates in practice. Therefore, identifying the optimal method for R t estimation is necessary, requiring a comprehensive evaluation of each method. In practice, the “ground truth” R t is difficult to obtain, which makes it challenging to directly compare the performance of different estimation methods. To address this issue, this study designs numerical experiments with a preset R t to compare five R t calculation methods—the time dependent (TD) method, the sequential Bayesian (SB) method, the new time-varying (NT) method, the Bayesian latent variable (BLV) method and the Bayesian data assimilation (BDA) method—across four metrics: correlation, trend similarity, dynamic time warping (DTW) distance, and confidence deviation. Finally, we integrate these metrics using the distance between indices of simulation and observation (DISO) framework to derive an overall evaluation, under which the TD method performs optimally among the considered approaches. In addition, we apply the five methods to COVID-19 and influenza A data for a qualitative comparative analysis.
Shi et al. (Tue,) studied this question.