Correlation vs causation once again
July 27, 2017 4:22 PM Subscribe
This time taught by xkcd cartoons "Hill for the data scientist: an xkcd story"
Oh man, I just had an actual run in with that last one last week.
ANOVA, people, it's called an ANOVA.
posted by telepanda at 5:53 PM on July 27, 2017 [2 favorites]
ANOVA, people, it's called an ANOVA.
posted by telepanda at 5:53 PM on July 27, 2017 [2 favorites]
So is this further evidence that there is an XKCD for everything, or not?
posted by Mr.Encyclopedia at 6:19 PM on July 27, 2017 [3 favorites]
posted by Mr.Encyclopedia at 6:19 PM on July 27, 2017 [3 favorites]
So is this further evidence that there is an XKCD for everything, or not?
Just so we don't rush them, who's first to get a research grant for this?
posted by Samizdata at 6:40 PM on July 27, 2017 [3 favorites]
Just so we don't rush them, who's first to get a research grant for this?
posted by Samizdata at 6:40 PM on July 27, 2017 [3 favorites]
So is this further evidence that there is an XKCD for everything, or not?
From what I learned in statistics class, maybe.
(Significance is one of the best XKCDs ever; pleased to see it here.)
posted by TedW at 7:34 PM on July 27, 2017
From what I learned in statistics class, maybe.
(Significance is one of the best XKCDs ever; pleased to see it here.)
posted by TedW at 7:34 PM on July 27, 2017
Good idea & execution. I love the comic they use for consistency. As the alt-text says, how could you choose avoiding a little pain over understanding a magic lightning machine?
The obligatory whining about my personal pet peeves:
If we found it perfectly plausible, we may not need statistics to show the relationship.: No, no, no! Plausibility just means it makes sense. It isn't evidence it's true. Very often we can construct plausible stories that tell us opposite things. Something totally implausible should not be believed, but something totally plausible isn't necessarily true or even probable.
Specificity: No complaints with her text on this one but the XKCD is pointing to a totally different flaw in thinking that drives me crazy: Curing something in a test tube is relatively easy (as is doing something at a small scale, inefficiently, uneconomically, etc.) but these things get reported breathlessly. I wish it weren't hijacked for a different concept but maybe she couldn't find a better one.
Also a bonus complaint not directly related to the article but I'm not sure the phrase "correlation doesn't imply causation" has done more good than harm. I see it used to glibly discard studies who's conclusions someone doesn't like with a tone of smug superiority from someone who hasn't engaged at all with the research. It's definitely a favorite to trot out for every study that shows minority or women get paid less or have fewer opportunities.
posted by mark k at 8:54 PM on July 27, 2017 [9 favorites]
The obligatory whining about my personal pet peeves:
If we found it perfectly plausible, we may not need statistics to show the relationship.: No, no, no! Plausibility just means it makes sense. It isn't evidence it's true. Very often we can construct plausible stories that tell us opposite things. Something totally implausible should not be believed, but something totally plausible isn't necessarily true or even probable.
Specificity: No complaints with her text on this one but the XKCD is pointing to a totally different flaw in thinking that drives me crazy: Curing something in a test tube is relatively easy (as is doing something at a small scale, inefficiently, uneconomically, etc.) but these things get reported breathlessly. I wish it weren't hijacked for a different concept but maybe she couldn't find a better one.
Also a bonus complaint not directly related to the article but I'm not sure the phrase "correlation doesn't imply causation" has done more good than harm. I see it used to glibly discard studies who's conclusions someone doesn't like with a tone of smug superiority from someone who hasn't engaged at all with the research. It's definitely a favorite to trot out for every study that shows minority or women get paid less or have fewer opportunities.
posted by mark k at 8:54 PM on July 27, 2017 [9 favorites]
I'm avoiding green jelly beans from now on.
posted by chapps at 11:01 PM on July 27, 2017 [1 favorite]
posted by chapps at 11:01 PM on July 27, 2017 [1 favorite]
My favorite xkcd for illustrating causation versus correlation is Conditional Risk.
posted by Jonathan Livengood at 11:13 PM on July 27, 2017 [5 favorites]
posted by Jonathan Livengood at 11:13 PM on July 27, 2017 [5 favorites]
Hell yeah, I'll buy a handgun, if I can kill cancer with it!
But of course you can! If you don't care much for the patient, that is.
posted by Laotic at 3:23 AM on July 28, 2017 [1 favorite]
But of course you can! If you don't care much for the patient, that is.
posted by Laotic at 3:23 AM on July 28, 2017 [1 favorite]
Also a bonus complaint not directly related to the article but I'm not sure the phrase "correlation doesn't imply causation" has done more good than harm. I see it used to glibly discard studies who's conclusions someone doesn't like with a tone of smug superiority from someone who hasn't engaged at all with the research. It's definitely a favorite to trot out for every study that shows minority or women get paid less or have fewer opportunities.
Often by redditors who have no idea what it means to condition on a variable.
Its like they are missing a third piece of the puzzle. Correlation between x and y, when all other potential causes of y are conditioned on, does imply causation. The hard part is inducing that conditional independence.
posted by MisantropicPainforest at 5:30 AM on July 28, 2017 [5 favorites]
Often by redditors who have no idea what it means to condition on a variable.
Its like they are missing a third piece of the puzzle. Correlation between x and y, when all other potential causes of y are conditioned on, does imply causation. The hard part is inducing that conditional independence.
posted by MisantropicPainforest at 5:30 AM on July 28, 2017 [5 favorites]
I can't wait to use the handgun one the next time someone tells me that bee venom (or insert your hipster alternative therapy here) kills HIV.
posted by Sophie1 at 6:08 AM on July 28, 2017
posted by Sophie1 at 6:08 AM on July 28, 2017
Correlation between x and y, when all other potential causes of y are conditioned on, does imply causation. The hard part is inducing that conditional independence.
Yeahbut the list of all other potential causes of y is infinitely long.
I still favor the broad approach taken by King/Keohane/Verba -- you're never going to prove causality. The best you can hope for is to show that lots of observable relationships in the real world are consistent with your causal story and very few are inconsistent with it. So you should think really hard about what you would see if your causal story is correct beyond the most immediate and obvious implications of your theory.
Which probably aren't even implications of your theory so much as the thing you wanted to create a theory about -- if your theory is "X causes Y because Reasons," your theory is the "because Reasons" part, not the "X causes Y" part.
posted by ROU_Xenophobe at 6:14 AM on July 28, 2017 [3 favorites]
Yeahbut the list of all other potential causes of y is infinitely long.
I still favor the broad approach taken by King/Keohane/Verba -- you're never going to prove causality. The best you can hope for is to show that lots of observable relationships in the real world are consistent with your causal story and very few are inconsistent with it. So you should think really hard about what you would see if your causal story is correct beyond the most immediate and obvious implications of your theory.
Which probably aren't even implications of your theory so much as the thing you wanted to create a theory about -- if your theory is "X causes Y because Reasons," your theory is the "because Reasons" part, not the "X causes Y" part.
posted by ROU_Xenophobe at 6:14 AM on July 28, 2017 [3 favorites]
Also a bonus complaint not directly related to the article but I'm not sure the phrase "correlation doesn't imply causation" has done more good than harm. I see it used to glibly discard studies who's conclusions someone doesn't like with a tone of smug superiority from someone who hasn't engaged at all with the research.
This is the same problem that creationists try to use against the "Theory" of Evolution: when used by logicians, "imply" doesn't have the wiggle room that it does when normal people use it.
And yeah, if I had a nickel for every time I've replied, "You know, correlation doesn't disprove causation...", I'd have a shitload of nickels.
posted by Etrigan at 6:46 AM on July 28, 2017 [4 favorites]
This is the same problem that creationists try to use against the "Theory" of Evolution: when used by logicians, "imply" doesn't have the wiggle room that it does when normal people use it.
And yeah, if I had a nickel for every time I've replied, "You know, correlation doesn't disprove causation...", I'd have a shitload of nickels.
posted by Etrigan at 6:46 AM on July 28, 2017 [4 favorites]
As an autism specialist who does a lot of trainings on Autism 101 and "Autism and Science and Not Science" I am definitely going to include some XKCD to further get the point across on various things. A well-made comic may be more effective than my saying another thousand words.
posted by ITravelMontana at 4:03 PM on July 28, 2017
posted by ITravelMontana at 4:03 PM on July 28, 2017
I see it used to glibly discard studies who's conclusions someone doesn't like with a tone of smug superiority from someone who hasn't engaged at all with the research.
And yeah, if I had a nickel for every time I've replied, "You know, correlation doesn't disprove causation...", I'd have a shitload of nickels.
By far the most important and nuanced line in that XKCD comic is, "Well, maybe."
posted by straight at 4:25 PM on July 28, 2017
And yeah, if I had a nickel for every time I've replied, "You know, correlation doesn't disprove causation...", I'd have a shitload of nickels.
By far the most important and nuanced line in that XKCD comic is, "Well, maybe."
posted by straight at 4:25 PM on July 28, 2017
« Older A sewer becomes a frying pan, gas meters turn into... | 110 N.F.L. Brains Newer »
This thread has been archived and is closed to new comments
posted by Oyéah at 5:28 PM on July 27, 2017 [1 favorite]