Many different types of dashboards have been developed to help us understand COVID-19 trends and/or to help inform policy decisions. Typically, dashboards include combinations of the following features.
Data visualizations are visual presentations, such as maps or graphs, of COVID-19 data (ex. cases, hospitalizations, deaths). Visualizations may present existing data (ex. total counts) in different ways (ex. by race or ethnicity), but do not attempt to fit a statistical model to the data or make predictions beyond the observed data.
Examples include the Gillings COVID-19 Dashboard, the NC DHHS Dashboard, and the numerous county and university/college based dashboards linked on this page.
Forecasting models may use multiple sources of data to simulate what may happen under different situations or scenarios. These types of models take data from the past and use it to predict how the future may turn out. However, some forecasting models do not take into consideration how the virus spreads, and therefore may not accurately predict the long-term effects of the virus in a variety of scenarios. On the other hand, models that incorporate disease transmission mechanisms (that is, mechanistic models) do account for how the virus spreads (i.e. the mechanism), and thus, with sufficient information, can be used to make predictions about how the virus may spread under different hypothetical scenarios, such as under different policies that promote social distancing or mask-wearing. It is important to understand that we cannot analyze future scenarios without incorporating a mechanistic understanding of the virus (and other factors impacting virus transmission) into the model.
Different tools have been developed to help businesses, policymakers, or other decision makers deal with the uncertainty that the COVID-19 pandemic brings. There are many types of decision tools: some are in the form of risk calculators based on estimated probabilities of infection given a certain group size and location, while others are more proprietary algorithms that may be based on data but have many built-in assumptions about the data and how it can or should inform decisions.
More sophisticated risk calculators report an actual percent risk that can be interpreted. For example, the Event Risk tool developed by NC COVID-19 calculates the percent chance that, at a 25-person event, at least one attendee has COVID-19 and is currently infectious. Other tools may have subjectively ranked levels of risk. For example, the tool COVID Can I Do It gives a 1-5 rating (low to high) risk for different types of activities. In any of these cases, estimations of risk require an accurate understanding of the true burden of infection in a population. In reality, we may only have an approximate understanding of the true burden of infection. Although scientists can chart trends in case counts, it is challenging to know the true burden of infection in a population because not all cases are identified. Moreover, the dynamics of the pandemic can change quickly, and the data that are available to us may not always be completely up to date. Previous work has found that models fit to confirmed COVID-19 case counts may not be reliable predictors of the actual prevalence of COVID-19. Despite these limitations in their ability to perfectly map the COVID-19 disease burden, risk calculators are helpful and tend to be easily understood by the public.
Other decision tools may be informed by data, but give answers that are not based on probabilities or other interpretable units. These tools usually ask a series of questions with yes/no response options. Such tools may have hidden or biased assumptions that have been built into the tool. This means that the output from these tools may be highly subjective. Such subjective built-in values lead to decisions that could be biased depending on the priorities, beliefs, and assumptions of the tool creator. Although decision tools are approachable and often designed to be easy to interpret, their recommendations should be interpreted with caution due to the underlying biases outlined on this page.