A model for improvement

From forecasting cases of COVID-19 to predicting future interest rate rises, statistical models have a big impact on our daily lives. But how can we gauge their reliability of this analysis? Associate Professor David Frazier has the answer.

A/Prof Frazier says for better or worse, the world runs on models.

“For example, throughout the pandemic, epidemiological and statistical models have been used to forecast the spread of COVID-19 and the likely hospitalisation rates,” he says.

But the Australian Research Council DECRA Fellow warns these models are not “magical black boxes”.

“Models are fabricated constructs that simply take in inputs and spit out a range of answers, which are subsequently interpreted and used by policy-makers,” he says.

“With the volume and easy availability of data sources, researchers in academia and industry are pushing the boundaries of modelling and statistical techniques.

“Figuring out when we can, and more importantly, when we cannot, rely on the statistical analysis obtained from complex models is a big motivation for me.”

A/Prof Frazier has dedicated years of research to testing the accuracy of the statistical procedures used to model data across a wide range of disciplines from infectious disease to finance.

“Any model and statistical procedure can be broken down into a set of assumptions, which allow us to put structure on the problem,” he says.

“Using mathematics, this structure can be leveraged to develop results that allow us to infer when the output of this statistical workflow is reliable.”

His research has been published widely in prestigious journals including Journal of the Royal Statistical Society: Series B, Biometrika, Journal of Econometrics, and Journal of Computational and Graphical Statistics.

"Any model and statistical procedure can be broken down into a set of assumptions, which allow us to put structure on the problem."

The Dean's Award for Excellence in Research recipient is quick to acknowledge the critical role his network of collaborators has played in his success.

“I’m very grateful that the committee deemed my research important enough and impactful enough to be recognised with this prestigious award, and I would like to credit my many collaborators for the instrumental role they played in the success of our work,” he says.

“A senior academic once told me a very valuable piece of advice: there are two types of successful academics, those that are absolutely brilliant, and those that are good co-authors.

“My academic successes are a direct consequence of me having great co-authors, and knowing how to be a good co-author.”

“Using mathematics, this structure can be leveraged to develop results that allow us to infer when the output of this statistical workflow is reliable.”

His research has been published widely in prestigious journals including Journal of the Royal Statistical Society: Series B, Biometrika, Journal of Econometrics, and Journal of Computational and Graphical Statistics.

The Dean's Award for Excellence in Research recipient is quick to acknowledge the critical role his network of collaborators has played in his success.

“I’m very grateful that the committee deemed my research important enough and impactful enough to be recognised with this prestigious award, and I would like to credit my many collaborators for the instrumental role they played in the success of our work,” he says.

“A senior academic once told me a very valuable piece of advice: there are two types of successful academics, those that are absolutely brilliant, and those that are good co-authors.

“My academic successes are a direct consequence of me having great co-authors, and knowing how to be a good co-author.”