News for hackers. Stuff that matters.

A comment on $5 device accurately tests for breast cancer in under 5 seconds. Parent comment

Re:$5 device accurately tests for breast cancer in under 5 s... by ethanbond on Friday February 16, 2024 @12:25PM (#39400083)
If you're really good at statistics, you can just intuit what seems like a sufficient N. This is actually a powerful statistical method because you can vary your intuition depending on whether you want to believe the study was well-powered or not, enabling you to easily discard Bad Evidence and include Good Evidence.

Few understand this.

[ Parent | on HN ]
Re:$5 device accurately tests for breast cancer in under 5 s... by Spivak on Friday February 16, 2024 @01:07PM (#39400713)
You don't intuit it, you calculate it. If you claim to be able to predict coin flips with a specific accuracy I can give you the exact number of trails N you need to be x% confident.

If you claim 90% accuracy and I want to be 95% confident with the standard alpha of 0.05 you need 38 trials.

[ Parent | on HN ]
Re:$5 device accurately tests for breast cancer in under 5 s... by jncfhnb on Friday February 16, 2024 @12:51PM (#39400466)
No, that’s utter nonsense. Statistical power is well defined and the required N to reliably detect an effect is a function of its effect size, which is the thing that wannabe statisticians don’t understand.
[ Parent | on HN ]
Re:$5 device accurately tests for breast cancer in under 5 s... by lazyasciiart on Friday February 16, 2024 @02:00PM (#39401466)
> you can vary your intuition depending on whether you want to believe the study

I suspect it is deliberate nonsense.

[ Parent | on HN ]
Re:$5 device accurately tests for breast cancer in under 5 s... by solardev on Friday February 16, 2024 @02:14PM (#39401647)
> I suspect it is deliberate nonsense.

But is that your intuition talking?

[ Parent | on HN ]