Note that there was no LGA dataset released yesterday, so the regional breakdowns are more out of date than usual - three days behind the statewide figures.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Figured out why my Hunter region R_eff calculation had zero uncertainty. I was literally clipping R_eff values to the "sane" assumption that it would be less than 10 in the uncertainty calculation. Some other clipping for the sake of "sanity" was also limiting the central value.
What used to be insane now isn't. Here's what the Hunter looks like with none of this clipping.
The clipping was there for a reason, sometimes small numbers and bad luck does produce stupid R_eff estimates. My code might now say stupid things in those contexts instead. But won't have to worry about that until numbers are smaller.
They're the experts, but I think ATAGI is making the wrong call. There is a persistent bias in health regulation toward inaction in uncertain times, as if regulators are somehow less responsible for the consequences of inaction than of action. 1/🧵
Much in the article emphasises uncertainty. But we need to make judgement call with the information available. Just because of uncertainty doesn't mean the best option is doing nothing. There's no reason to privilege that course of action over any other. 2/🧵
Just as there are considerable uncertainties about the negative consequences of bringing forward vaccines, there are considerable uncertainties about the negative consequences of not doing so as well. 3/🧵
You might have noticed my short-term projections for the NSW outbreak yesterday were pretty nuts. It turns out this projection was unrealistic, due to some approximations in my code that are no longer valid now that growth rates are higher. 1/🧵
So first, apologies. These issues are fixed now and today's projection ought to be more realistic. Secondly, gory details below. There were two problems. 2/🧵
Problem 1. smoothing. Many have noticed my "smoothed cases" line didn't actually go through the middle of the data. This is not incorrect per se, this is what correct smoothing with a symmetric smoothing window looks like when the data has a nonzero second derivative. 3/🧵