The IHME coronavirus model keeps being wrong. Why are we still listening to it?

Maybe the others tried and decided not to publish their model because they couln't make it good enough.

The least worst model doesn't necessarily pass the threshold of usefulness. If everybody had said a more honest "we don't know, the data is not good enough" then maybe there would have been more effort into getting better data.

The witch hunt analogy is interesting. Witches and witchdoctors made magical claims, like being able to see the future by reading tea leaves. In primitive societies they had enormous influence even though they couldn't do what they claimed. Eventually they were displaced by religions whose priests were not expected to have such powers.

Epidemiologists cannot predict diseases. This is by now an established fact. They have never managed to do it - the history of the field is one of constant failure combined with an apparent inability to improve.

I think the back and forth on this field comes from a belief that if epidemiologists aren't funded and don't model they can't get better. So we shouldn't criticise because they're only trying their best.

But that assumes that improvement is only possible through deployment. That isn't the case. Epidemiology should be defunded by governments completely - the private sector is much better suited to this work. The field has a natural home in the insurance sector, perhaps under a different name. Currently insurance firms don't sell pandemic insurance - probably because models aren't good enough for deployment. Their predictions are always for huge pandemics that don't happen, which would result in unusably expensive policies.

https://fivethirtyeight.com/videos/how-one-modeler-is-trying...

"The nature of SEIR models is they do tend to always show this huge increase up to the point of saturation where most people get infected and very often that has not historically happened"

(for "very often not happened" read "never happened")

The private sector is the right place for this sort of work because it has executives. One purpose of the insurance executive is to decide whether or not a new policy should be sold, and whether the firm understands the risks sufficiently. They make the decision about whether a model should "go to production" or not.

Academia lacks executives. Nobody exists who can tell academics: your output is useless and will therefore not be used. Instead that job falls to politicians, who appear systematically incapable of pushing back on people they perceive as "scientists" (although modellers aren't scientists, as I've argued in prior comments). Executives aren't like that - in fact the frequent lack of impact data scientists have was the topic of a recent complaint article posted to HN!

We're now witnessing the severe problems the lack of executives creates. Models that aren't anywhere near ready for production are nonetheless deployed for policymaking purposes, leading to disaster when their outputs are wildly wrong. Indeed academics are in effect rewarded for deploying inaccurate models. In the UK, the government diverted resources into building field hospitals for COVID that aren't required yet now has a massive backlog of cancer and other patients whose important surgeries were cancelled in the model driven expectation of a surge that never materialised. A significant fraction of excess death in the UK now doesn't mention COVID on the death certificate at all, and is due to people avoiding hospitals because they think they're overloaded - the media tends to present models as fact, so people can't separate predictions from reality. The same story repeats throughout the world, but the UK seems to be amongst the hardest hit.