Brain wave signals during sleep show how risk-seeking people may be

medicalxpress.com

20 points·rajnathani··8 comments

The graph in the summary showing "SWA in the right PFC" is almost comically unconvincing.

Remove the two or three most extreme data points and the effect disappears completely. Even removing a single one of those extreme data points is probably enough to toss the significance of the fit and be fully consistent with zero correlation.

But it says p=0.004 so it must be a good result!!!!!

The full article has 5 of these graphs. I'm staring at them and I still can't believe the tiny p values. Is the data of the graph available? Can someone repeat those graphs and explain the small p values?

[My guess is that the slope is caused by the cluster at (SWA=150%, Risk=2m) in the graph.]

You can correlate those brain waves with anything really: predisposed to eat oranges to climb hills, likes color blue. Even infinitely small p won't save those studies.

How many constellations can you find on the data points?

Running a linear correlation over them must be a practical joke.

I'm not knowledgeable about statistics and studies like this in general; do you mean this is a "green jelly beans cause acne" [1] situation, where you test for many hypotheses and one of them turns out to be significant?

[1]: https://xkcd.com/882/

Exactly, if you test many many things against each other you will always find correlations between them by chance. That's why you always need to start with a "serious" hypothesis.

I have some serious doubts about the conclusions of this paper - if think studies like this are extremely dangerous to the field in general.

Reply on news.ycombinator.com