News for hackers. Stuff that matters.
Re:All models are wrong, but some are completely wrong by smitty1e on Sunday April 05, 2020 @02:11PM (#22787250)
One feels that, irrespective of the models, the data in the covid-19 case may be unusually bad.

It may be time to add a third error category[1]

I. False positive

II. False negative

III. Deliberately skewed off the map for propaganda reasons.

[1] https://en.m.wikipedia.org/wiki/Type_I_and_type_II_errors#Ty...

Re:All models are wrong, but some are completely wrong by YZF on Sunday April 05, 2020 @02:44PM (#22787497)
This is driving me crazy ;) the poor quality of data.

You'd want health organizations around the world to be publishing every possible detail (anonymized) so that the disease can be better understood. Yet three months in, with over a million cases worldwide, we still have experts disagreeing about things like asymptomatic transmission, use of masks, droplets vs. aerosol, how much distance one should stand from another, viability on surfaces, etc. etc. Even for treatment options rather than insisting on randomized double blind trials start by using the natural experiments that are already happening.

We should have the data to answer a lot of these questions (or at least draw out some probability distributions), or at least someone has it. This stuff is going to be critical in informing exit strategies.

Re:All models are wrong, but some are completely wrong by magoghm on Sunday April 05, 2020 @02:47PM (#22787518)
My experience with data is: you never get good data. Always, the closer you look at it the more problems you find in how it was collected.
Re:All models are wrong, but some are completely wrong by thaumasiotes on Sunday April 05, 2020 @06:02PM (#22789104)
> You'd want health organizations around the world to be publishing every possible detail (anonymized)

If you're publishing every possible detail, then your data isn't anonymized. Anonymization consists of the removal of almost all of the details.

Re:All models are wrong, but some are completely wrong by smitty1e on Sunday April 05, 2020 @06:52PM (#22789407)
Or even if the data are reasonably tidy, they might expose more when juxtaposed with another dataset.