Did COVID-19 models get it wrong? Professor Sally Cripps explains
Did COVID-19 models in the United Kingdom and the United States get it wrong?
As part of an online discussion between an international group of researchers, including Stanford University Professor of Medicine John P.A. Ioannidis and Northwestern University statistician, Professor Martin A. Tanner, University of Sydney statistician Professor Sally Cripps discusses the efficacy of the two countries' COVID-19 modelling efforts, with recommendations on how countries – including Australia – can better prepare for uncertainty.
In April, Professor Sally Cripps alongside US colleague, Professor Martin A. Tanner found that over 70 percent of US states had death rates that were inconsistent with IHME predictions. Credit: Centre for Translational Data science.
“COVID-19 models got it wrong in that they failed to forecast with any degree of accuracy daily death counts even when the forecast was for one day in advance. In addition they failed to accurately quantify uncertainty," said Professor Cripps, who is the Director of the University of Sydney Centre for Translational Data Science and the ARC Industrial Transformation Training Centre for Data Analytics for Resources and Environments.
“But that doesn't mean all modelling is bad or ineffective. In fact, what this scenario represents is an opportunity to learn and do better. Modellers have tried to justify their faulty models rather than learn from their mistakes."
At the outbreak of the COVID-19 pandemic, societal lockdown was the sensible thing to do. Australia acted swiftly in that regard, but now we need to find a way out.
“At the outbreak of the COVID-19 pandemic, societal lockdown was the sensible thing to do. Australia acted swiftly in that regard, but now we need to find a way out.
“Though lockdowns minimise COVID-19 deaths, there are many studies which show that lockdowns increase domestic violence, as well as increase deaths from cancer and heart attacks.
“We need to develop models which have multi-criteria functions that allow governments to better understand a measure’s effect across all of our society.
“We also need access to reliable data, and models that are validated and continuously reappraised for their performance in real-time."
As of 26 June 2020, COVID-19 had claimed almost 125,000 lives in the United States and over 43,000 in the United Kingdom.
DISCLOSURE:
Professor Cripps conducted an analysis on the efficacy of COVID-19 modelling efforts as part of an ongoing online discussion with international colleagues, Stanford University Professor of Medicine John P.A. Ioannidis and Northwestern University statistician, Professor Martin A. Tanner. The discussion has been published by the International Journal of Forecasters.