Much has been made of the modelling now underlying Daniel Andrews’ COVID roadmap — after all, in the words of the Premier himself, this is “one of the most comprehensive modelling exercises that we have ever seen in the state, arguably nationally, and a piece of work that has international significance as well”.
The Premier has also been fond of reminding reporters and others who ask any kind of questions about the modelling exercise that the whole thing was run on a ‘supercomputer’ — as if that alone puts the outputs above reproach or debate. Modelling is useless if it can’t be interrogated or replicated. Anyone claiming otherwise is either moronic or malicious.
Here’s the thing, though: Dan Andrews is either a liar or genuinely ignorant as to the construction and operation of the model that is now guaranteeing Victoria will be under effective lockdown for at least the next two months. How can we prove this? Easy — download the model from source code and run it yourself. Who knows, with a bit of parameter fiddling (or, if you prefer, careful calibration of assumptions) you might just be able to create a future where Victoria doesn’t commit economic suicide and ruin Christmas for millions.
Step One: Find the peer-reviewed article that discusses the construction of the COVID roadmap model simulation. Find the link to the Github (code) repository in the article and download all the source files to your laptop.
Step Two: Download NetLogo. As the website helpfully explains, NetLogo is a ‘multi-agent programmable modelling environment’. What does this mean exactly? At the core of the Victorian COVID model is the notion that citizens (i.e. agents) interact with each other and in the process run the risk of contracting or spreading COVID. NetLogo is a handy piece of software that makes it easy to create virus modelling scenarios. Oh, and it’s free (unless you choose to make a donation when you download it).
Step Three: Unzip the archive file you downloaded from Github and find any of the NetLogo files (extension .nlogo) and run them in NetLogo on your desktop. Congratulations! You have now successfully “modelled” the pandemic. Here’s what you should see:
There are some caveats to this, firstly that the instructions given confirm that “models relating to the current Victorian situation are not included in the repository, but they can be considered analogous in many ways to those that are”.
So you can’t replicate the lockdown modelling exactly from a single file, but if you read the journal article carefully enough you should be able to find all the parameters used for the roadmap scenario and adjust the generic model to match. Run it 1,000 times, tabulate your results and voila — you have just performed a highly scientific act, the attempt to replicate someone’s results.
Of course, two things are apparent: even if you have the precise source code used to run the lockdown scenario you wouldn’t be able to replicate its results exactly anyway. Why is this? The model uses stochasticity (or, in layman’s terms, randomness) to simulate certain interactions in the spread of COVID. That is, each time the model runs it simulates thousands of interactions using randomly generated numbers as inputs, ergo, you cannot replicate its results exactly as the probability of the same random numbers being generated in your version are ludicrously small.
Secondly, it should be clear that a supercomputer is not required to run the model. Did Melbourne University run it on a supercomputer anyway? They sure did. One has to wonder, did Dan Andrews know this? The author of the model, Jason Thompson, helpfully points out that it was ‘originally built for use by anyone on a laptop’.
One final observation — the Victorian government provides a somewhat useless summary PDF on the modelling outcomes here. Until late yesterday, the PDF included a live link to the journal article providing the Github repository used for the model. Today, the PDF includes no such link. Is it possible that the emperor fears that all will see him — or perhaps, his model — naked?
The author is a former statistician. A pseudonym has been used to protect their privacy.
Got something to add? Join the discussion and comment below.