No uncertain terms

Three words—forecast, projection, and prediction—in scientific terms are similar, but they have distinct implications in specific contexts. I find it concerning when I read research papers, announcements by government agencies and modelers, or popular science coverage that use the terms incorrectly. Such misuse may cause misinterpretations. And in the worst-case scenario, the correct meanings may be dismissed and the incorrect meanings enforced. It is the responsibility of scientists to correctly and appropriately use scientific terms and to interpret and communicate them with caution.“Forecasts” of the COVID-19 pandemic have been offered by agencies, institutions, and teams around the world. As this issue of Physics Today goes to press, the COVID-19 Forecasting and Mathematical Modeling webpage, via the US Centers for Disease Control and Prevention (CDC), each week provides four-week “forecasts” for COVID-19 hospitalizations and deaths.11. For COVID-19 forecasts for new hospitalizations and new and total deaths in the US, see https://www.cdc.gov/coronavirus/2019-ncov/science/forecasting/mathematical-modeling.html. (Case forecasts have not been posted since December 2021.) The results presented include both the ensemble forecasts and the independent ones that the ensemble numbers are based on.Obviously, huge uncertainties are associated with those forecasts. The CDC’s hospitalization and death forecast pages state that “models make various assumptions about the levels of social distancing and other interventions, which may not reflect recent changes in behavior.” Thus the possible scenarios may not necessarily be probable because of unpredictable factors such as national policies and human behaviors.22. J. Cepelewicz, “The hard lessons of modeling the coronavirus pandemic,” Quanta Magazine, 28 January 2021. The CDC case forecast page states, “While they have been among the most reliable forecasts in performance over time, even the ensemble forecasts have not reliably predicted rapid changes in the trends of reported cases, hospitalizations, and deaths. They should not be relied upon for making decisions about the possibility or timing of rapid changes in trends.”To forecast is to calculate some future events or conditions, usually as a result of study and analysis of available pertinent data. A forecasted event is a probable occurrence. The term is frequently used in reference to the weather—which is forecasted on the basis of correlated meteorological observations. If a weather forecast shows that it is going to snow tomorrow, that means snow is a rather probable weather condition for the next day. And whether it will snow tomorrow does not in the least depend on how humans behave or politicians debate.The way the CDC and many groups use the term “forecast” may cause confusion among the public, policymakers, and decision makers, leaving the wrong impression that a COVID-19 forecast is comparable to a weather forecast. In fact, “projection” is a more appropriate term to use with COVID-19 data. A projection offers only a conditional possible response that depends on the validity of the assumed future scenarios.To help explain the distinction between projection, forecast, and prediction, consider three sample sentences:1. The weather forecast by the Bureau of Meteorology shows that it is going to snow tomorrow. (The forecast is a probable occurrence.)2. The team’s projection shows that the world population will rise to over 11 billion by 2100. (The projection tells conditional possibilities.)3. Astronomers can make accurate predictions about when an eclipse is going to occur. (The prediction is an inference with certainty.)I have seen the term “projection” used correctly by COVID-19 modelers. A good example is modeling work from October 2021 reporting projections of how population contacts during the end-of-year holiday in Mexico City would potentially affect future pandemic outcomes, including infections, deaths, and hospitalizations.33. F. Alarid-Escudero et al., MDM Policy Pract. 6, 238146832110492 (2021), doi: https://doi.org/10.1177/23814683211049249. Some may argue that the term “prediction” should be used for COVID-19 modeling. But that would not be appropriate because a prediction implies an inference with certainty and does not convey the conditional possibilities implied by projection.Confusion between those two terms is by no means rare among scientists working in climate science, as revealed in a survey.44. D. Bray, H. von Storch, Sci. Commun. 30, 534 (2009). https://doi.org/10.1177/1075547009333698 Unfortunately, in a recent paper reporting the relationship between coastal carbon sequestration and climate change, the authors state that they “go beyond recent soil C stock estimates to reveal global tidal wetland C accumulation and predict changes under relative sea level rise, temperature and precipitation.”55. F. Wang et al., Natl. Sci. Rev. 8, nwaa296 (2021). https://doi.org/10.1093/nsr/nwaa296 But there are large uncertainties regarding the assumptions underlying the scenarios, such as unforeseen socioeconomic and technological conditions and uncertain global population growth in the coming decades. So the term “project” would be more appropriate here because the authors are talking about a conditional possible response that depends on the validity of the assumed future scenarios.I strongly recommend careful use of forecast, prediction, and projection in the reporting of science, particularly with regard to climate change and COVID-19. The public, decision makers, and policymakers will gradually get used to the uncertainties associated with projections.ReferencesSection:ChooseTop of pageReferences <<1. For COVID-19 forecasts for new hospitalizations and new and total deaths in the US, see https://www.cdc.gov/coronavirus/2019-ncov/science/forecasting/mathematical-modeling.html. Google Scholar2. J. Cepelewicz, “The hard lessons of modeling the coronavirus pandemic,” Quanta Magazine, 28 January 2021. Google Scholar3. F. Alarid-Escudero et al., MDM Policy Pract. 6, 238146832110492 (2021), doi: https://doi.org/10.1177/23814683211049249. Google ScholarCrossref4. D. Bray, H. von Storch, Sci. Commun. 30, 534 (2009). https://doi.org/10.1177/1075547009333698, Google ScholarCrossref5. F. Wang et al., Natl. Sci. Rev. 8, nwaa296 (2021). https://doi.org/10.1093/nsr/nwaa296, Google ScholarCrossref© 2023 American Institute of Physics.