These models are not basic statistical time series analysis. Time series models are run without context and aim to predict a curve given past history (seasonality and trends). The models used here are mechanistic. They simulate the underlying physical properties of the epidemic. These include things like known incubation periods, mixing rates, infectivity of the virus, etc.
Imagine trying to forecast tomorrow’s daytime temperature. A time series model would simply look at the list of temperatures over the last week and project that trend forward. This would basically be like fitting a trendline in Microsoft Excel or on a graph of points in 9th grade algebra. But these trends may not continue. A mechanistic model would look at high and low pressure regions, wind patterns, and cold fronts and then simulate the entire region’s climate.
The mechanistic models created by UVA are far more computationally intensive than time-series models, but often generate more realistic results. More importantly, unlike time-series analysis, a mechanistic model allows us to project the impact of hypothetical changes. What if!?
- What if… we could increase mask usage by 20%?
- What if… we could double the number of children 5-11 who were vaccinated?
- What if… we could get another round of boosters for the elderly?
- What if… a new variant arose that was 30% more infectious?
Since we understand the mechanics behind the epidemic, we can estimate the impact of these changes by varying parameters in the model. These models also allow us to compare the impacts of competing interventions. If you only have sufficient funds to provide extra boosters for the elderly, or extra masks in schools, which option is better? By comparing the outcome of these two scenarios we can make an educated guess as to which strategy will save more lives. Human behavior is impossible to predict, but a range of scenarios gives us a good idea of what may be coming and how best to respond to it.
Remember, local experts such as health directors and public health physicians often have a better feel for which scenarios may be most likely for their locales, and we do not expect any one scenario to perfectly predict the epidemic. These models are meant as tools to aid them and the public.