So the model on which the model on which the Victorian government apparently based its policy was based —

yeah, it's messy —

uses the following code to simulate transmission events.
Now, this is an agent-based model in a graphic environment (NetLogo) with a programming language that is pretty close to readable. I've done beginner training in it and I understand probably 95% of what they're doing. Code is taken from the GitHub.
Simuls are simulated agents – in old-school logo, 'turtles.'

If there are 2+ simuls in a block of virtual space, they can interact. This snippet handles the likelihood of Covid-19 being passed on when they interact.

Notice the word RANDOM all through the code snippet.
1. Simuls move at random through the virtual landscape, but as people, our movements are intensely non-random. We follow the same route to the same job/study and interact with mostly the same people most days.

In epi terms, this setup assumes mass action principle applies.
It uses repulsion at the encounter level to try and model the way in which different population segments do and don't interact, e.g. essential workers, students, household members, etc. But that's a really hack-y way of handling small world network structure in a model.
2. For every encounter, it picks a random value for each agent based on population means for each variable, such as infectivity. That is, to put it mildly, not realistic. Covid-19 has very time-specific infectivity and an agent-based model should track that per agent over 'time.'
With those two random factors – random mixing and random, per-encounter infectivity – you've got a model setup that is *going to favour lockdown* because lockdown applies, subject to some % compliance, to the whole population and all the encounters.
But in Victoria's actual outbreak, infections are not spread out across the whole population or happening during random encounters in public places — they're concentrated in particular sites (workplaces) and groups (defined largely by exposure to work sites/types).
You can follow @engagedpractx.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: