Page 1 of 1

T's instead of error bars

Posted: Tue Dec 06, 2022 2:19 pm
by shpalman

Re: T's instead of error bars

Posted: Tue Dec 06, 2022 2:34 pm
by IvanV
Thank you. I think I have never seen something that comes quite so blatantly and obviously into the category of cargo cult science. I remember first coming across the term on this forum, and having to think carefully about what it might mean. For all my interest in pseudoscience, I had never seen that term before, and presumed the poster must have made it up. I was very impressed when I realised what a nice distinction it was. But it was apparently first used by Richard Feynmann, and deserves to be better known.

Re: T's instead of error bars

Posted: Tue Dec 06, 2022 3:55 pm
by monkey
I wanted to do a joke about a t-test, but I read the article and it had already been done.

Re: T's instead of error bars

Posted: Tue Dec 06, 2022 11:28 pm
by Bird on a Fire
Thoroughly depressing.

Using a bar chart to plot a distribution of data is f.cking stupid. Plot the actual data, or at least a violin or box plot or something.

Re: T's instead of error bars

Posted: Wed Dec 07, 2022 12:01 am
by dyqik
Bird on a Fire wrote:
Tue Dec 06, 2022 11:28 pm
Thoroughly depressing.

Using a bar chart to plot a distribution of data is f.cking stupid. Plot the actual data, or at least a violin or box plot or something.
Violin plots are best, because they display the actual distribution, and inevitably look a bit Georgia O'Keeffe.

Re: T's instead of error bars

Posted: Wed Dec 07, 2022 12:13 am
by Bird on a Fire
Might start calling them O'Keeffe plots

Re: T's instead of error bars

Posted: Fri Dec 23, 2022 4:35 pm
by nekomatic

Re: T's instead of error bars

Posted: Sun Jan 01, 2023 8:02 pm
by jimbob
dyqik wrote:
Wed Dec 07, 2022 12:01 am
Bird on a Fire wrote:
Tue Dec 06, 2022 11:28 pm
Thoroughly depressing.

Using a bar chart to plot a distribution of data is f.cking stupid. Plot the actual data, or at least a violin or box plot or something.
Violin plots are best, because they display the actual distribution, and inevitably look a bit Georgia O'Keeffe.
I tend to prefer normalised probability plots, but yes.

I have converted about half my colleagues to those.

Re: T's instead of error bars

Posted: Mon Jan 02, 2023 12:20 pm
by Bird on a Fire
There is an R package for everything, now including correctly fitting Ts as error bars, via the wonderfully named geom_terrorbar() function:

https://mival.netlify.app/blog/2023/01/ ... terrorbar/