The model forecasts the number of future confirmed cases and deaths as reported by the Johns Hopkins University (JHU) Coronavirus Resource Center dashboard. We understand that the number of confirmed cases and deaths is an underestimate for the actual number of COVID-19 cases and deaths. However, the confirmed counts of cases and deaths as presented on the JHU dashboard provide measurements that we can validate the model forecasts against.
The model generates a confirmed cases/deaths forecast for any U.S. state that has at least one confirmed COVID-19 case/death and any country with at least 100 confirmed COVID-19 cases and 20 deaths. U.S. states with fewer than 100 confirmed cases and/or 10 confirmed deaths may be unstable and should become more stable as the number of confirmed cases/deaths increases. We are working to relax the conditions under which forecasts for countries are made.
The model produces forecasts, not projections; meaning it does not explicitly model the effects of interventions or other "what-if" scenarios. We distinguish forecasts as attempts to predict what will happen versus projections as attempts to describe what would happen given certain hypotheses. The model does, however, assume that interventions will be implemented and continue to be upheld in the future, resulting in an overall decrease in the growth rate of COVID-19. The model also allows for the possibility that the transmission rate may go up. See COVID-19 Growth Rate for more details.
The model forecasts are probabilistic. There is a high degree of uncertainty in future trajectories, given the possibilities of changing intervention strategies, changing case definitions, and changing rates of testing. The model does not explicitly model these different possible futures; doing so directly would be near impossible. Instead, the model integrates over those potential futures.
We intend to update forecasts each Monday and Thursday, incorporating the latest data into the model.
We acknowledge that every model, including ours, is an abstraction of reality and has its strengths and weaknesses. We present our model's state-by-state estimates in hope that multiple model predictions in the field will collectively provide more robust situational assessments and encourage collaborative efforts such as the development of multi-model ensembles.