My god, yo'd have to be an eternal optimist to draw a straight line through some of those data
What data patterns can lie behind a correlation coefficient? https://figshare.com/articles/What_data_patterns_can_lie_behind_a_correlation_coefficient_blog_post_/6945335/1 …pic.twitter.com/gVSBxLIZMK
-
-
-
Well, check out real examples here:https://www.frontiersin.org/articles/10.3389/fnhum.2012.00119/full …
End of conversation
New conversation -
-
-
Is there any argument for not essentially always using Spearman’s rho as long as the relationship is monotonic? As far as I can see, issues would only remain for only two or three of those examples.
-
Using Spearman by default would help. But it's still sensitive to bivariate outliers, so more robust estimators are sometimes needed.
-
Always showing scatters + Spearman would head off the vast majority of issues & allow for evaluation if other estimators are required.
-
Agreed. Even the language that most scientists use to describe their correlations—"X was positively associated with Y"—suggests that they are making a hypothesis about monotonicity, not linearity. Not sure why Pearson is the convention and not Spearman.
End of conversation
New conversation -
-
-
Counterpoint: r=0.5 does a decent job of explaining the general trend in each situation and requires 1/50th of the data to present.
-
That seems... not right. I mean, what trend is it explaining in (3) or (4) or (11) or (12)?
- End of conversation
New conversation -
-
-
Important point. This one is also pretty cute https://www.autodeskresearch.com/publications/samestats …pic.twitter.com/Q5zunko4Lt
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@cMadan https://twitter.com/robustgar/status/1027133166849544192 …@philippajane5 May come in handy . . . .#stats
-
Thanks! :)
-
No - thank you!
End of conversation
New conversation -
-
-
*All* reported correlations should be accompanied by a scatter plot, at least as supplementary data.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Brilliant. Needs fat tailed x & residuals. Then again, you can't see that in an xy plot.
-
here you go (t(1) distributions for both x and resid)pic.twitter.com/Itjonn3p1R
End of conversation
New conversation -
-
-
The lower part of the image (not visible in mobile preview) is particularly nice/scary. And the dinosaur, of course. Always plot your data.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
no dinosaur?
- 1 more reply
New conversation -
-
-
I might print this out
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.