0552: "Correlation"

[markfiend]

I think someone's karma ran over Apples1337's dogma.

There is no goddess but Eris and Malaclypse is her profit. Fnord.

[BlueLaughter]

Causation -> correlation.

Nope, doesn't work that way either. Let's say you have three variables.
A - the force you use to push a cart
B - the combined mass of the cart and it's contents
c - The speed the cart moves.

Now suppose we doubled A and quadrupled B,
Higher A causes higher C, however in this case as A goes up C goes down.

Correlation isn't a measurement of one instance.

[enjoyeverysandwich42]

Face it. Inductive reasoning never "proves" anything. And deductive reasoning doesn't function without inductive reasoning to support it. I don't think anybody here can "prove" that the Laws of Physics won't go haywire one day. That doesn't mean you should live in fear of getting thrown off the earth when gravity shuts down. Am I the only person who shouted "YES!" in between laughter after reading the alt-text?

My biggest problem with the problem of induction is that it's right. Now what do I do? Nothing? Anything? None of the options are good. It's a conundrum. There's no solid argument, IMO, against extreme skepticism.

The only real solution, of course, is to follow the example of Hume - go find something better to do than philosophy.


This strip was funny.


Someone mentioned that they didn't know of the logical definition of "implies". After taking logic twice (a philosophy class and also in discrete math) I sort of forgot about the english definition.

[islesv]

Surprisingly, someone I know in real life whom I know is so much better at math than me (I'm just a high school math teacher, she is a college instructor) did not get the joke.

[Brooklynxman]

anime vs hentai (hentai in red)
http://google.com/trends?q=anime%2C+hen ... all&sort=0

Coincidence? I think not (in some cases, its okay to go out on a limb)

I have proof that correlation ⇏ causation. See Google trends doesn't show a link!

Though correlation seems to be decreasing, but still has a cycle.

What was that a graph of?

And to washington cupid person. Its b/c of presidents day and valentine's day. But the corealation isn't perfect, unlike hentai-anime.

[phlip]

What was that a graph of?

correlation,causation (the search volume line for "causation" when I do it looks different to that picture, but the news volume line matches, so I'll blame Google for being strange).

[reevey]

It may have been the worst way to explain it, but I didn't even get that aspect of the joke until I came here. I was just cackling at the alt-text.

Yes, me too. I have a very vivid picture of Groucho Marx waving his cigar towards a Powerpoint slide, waggling his eyebrows and grinning wildly.

[Iridos]

definately gonna show my math teacher friend this one. Good comic.

well, as long as you don't show this post to your English teacher...
http://www.urbandictionary.com/define.p ... definately

CHRIST

IM AN ATHEIST

Haha... well said... very well indeed.

You cannot actually establish causation.

Of course you can - if you change A while leaving everything else the same and then B changes, you've shown causality. Sure, you might inadvertently and unknowingly also have changed C and this C is the true cause (and in so far you are correct) - but that means the experiment was not set up correctly.

The correlation != causality mostly comes into play when you can't do the experiment - like you find a correlation between heart attacks and obesity. Is obesity increasing the risk of heart attacks? The experiment would be to take two sufficiently groups of similar make-up and make one group obese without changing anything else, then see if they get more heart attacks. Well I guess you see the practical difficulties in running the experiment

As for the computer wire problem, let's say your problem was that you couldn't send an e-mail, and you found that plugging in the ethernet cable made the e-mail problem go away. Every time you unplug the ethernet cable, you cannot send the e-mail, and every time you plug it in, you can send the e-mail. However, it is incorrect to conclude that "plugging in the ethernet cable causes e-mail to work", since, in reality, you can send e-mail without plugging in the ethernet cable, if you have wireless internet access.

Oh yes, you have found causality there, because it's you who pulls the plug - and therefore changes a single parameter.
And yes - for this computer plugging in the ethernet cable "causes e-mail to work". The wireless would be a different thing that can cause the same result (if there's a wireless nearby and the computer can do wireless etc.).

So - it could be that the cable ethernet is indeed not connected and every time he plugs/unplugs the cable he hits a button on the wireless router with his head/knee and such switching it on or off - and then indeed it would be not the cable that "causes his e-mail to work" or not, but the wireless. The point here is, though, that the experiment is not set up correctly.

You can prove causality by changing the parameters of an experiment one by one (and repeating sufficiently often etc. etc.) - it's only when you cannot change single parameters one by one that you cannot see what is cause and what is effect.


I.

[EnderSword]

You cannot actually establish causation.

Of course you can - if you change A while leaving everything else the same and then B changes, you've shown causality. Sure, you might inadvertently and unknowingly also have changed C and this C is the true cause (and in so far you are correct) - but that means the experiment was not set up correctly.

The correlation != causality mostly comes into play when you can't do the experiment

Part of it though is that you can never actually do the experiment. You have no way of knowing if B was going to change at that moment anyway.

If you fire Cue ball A at 8-ball B, it moves, but you can never truly proved that's why it moved, or that it wouldn't have moved without you doing anything to A.
C may not even be necessary, since C simply assumes a different cause.

You can't prove that anything causes anything at all, let alone the a specific Things causes a specific other thing.

[Iridos]

You cannot actually establish causation.

Of course you can - [... if you can do the experiment ...]
The correlation != causality mostly comes into play when you can't do the experiment

Part of it though is that you can never actually do the experiment. You have no way of knowing if B was going to change at that moment anyway.

If you fire Cue ball A at 8-ball B, it moves, but you can never truly proved that's why it moved, or that it wouldn't have moved without you doing anything to A.
C may not even be necessary, since C simply assumes a different cause.

You can't prove that anything causes anything at all, let alone the a specific Things causes a specific other thing.

That's what you repeat experiments for and that's what physics does all the time.
I admit I shouldn't call it 'prove', as this has a distinct meaning in mathematics and you cannot 'prove' it in this sense.... and you are in the middle of "I think, therefore I am" territory (but what if I stop thinking for a moment because I have sex? oops..wrong comic) ...

But people didn't seem to be aware of the difference between correlations in e.g. medical data (like the obesity/heart attack example) and something like your cue balls or Newtons Law, where you can do the experiment and find a way to very accurately describe reality - and in that model there *is* something cause and effect.

We *can* find out what is cause and effect - but only if we can change the single parameters at will - and if that's not the case, there's only a correlation and that does indeed not show at all that one of the correlated events causes the other.

That's actually why the punch line isn't really such a great one - of all people, he should know why he doesn't believe that correlation implies causation anymore, now shouldn't he?


I.

[leveldeaded]

I really want to run a randomized controlled trial to prove that XKCD causes laughter. It could be so simple in a sound proof room, until we started testing with and without the alt text. I was thinking the control would be Garfield, but then we'd want to test Garfield minus Garfield too. Oh dear. So much for simplicity.

[dennisw]

I really want to run a randomized controlled trial to prove that XKCD causes laughter. It could be so simple in a sound proof room, until we started testing with and without the alt text. I was thinking the control would be Garfield, but then we'd want to test Garfield minus Garfield too. Oh dear. So much for simplicity.

I'd think a good control for false positives would be Nancy.

[AMZ]

As someone who loves statistics, I heartily enjoyed this particular comic and sent it to various math teachers, etc. However, I think it has a bit of a flaw (or else a deeper level of humor than has otherwise been acknowledged in this forum). The inference, "You took a statistics class and that caused your learning" is not, to my mind, an example of fallacious thinking of the "correlation implies causation" form. If fallacious, it is an example of "Post hoc ergo propter hoc" thinking. To my mind, correlation requires interval or ordinal data, so that we see "as x gets higher (or lower), y gets higher (or lower)," neither of which is true of the two variables in the comic. The "this happened, then that happened, therefore this caused that" is the classic post hoc fallacy, not the correlation fallacy. But I still laughed out loud.

[EugeneSlipped]

Here's another interesting correlation/causation note I've been hearing. You know those earthquakes and Tsunamis that happened in the last week? There are people crying out that it's because of global warming.
Uuummmmmm yeah.

[aquilo]

As someone who loves statistics, I heartily enjoyed this particular comic and sent it to various math teachers, etc. However, I think it has a bit of a flaw (or else a deeper level of humor than has otherwise been acknowledged in this forum). The inference, "You took a statistics class and that caused your learning" is not, to my mind, an example of fallacious thinking of the "correlation implies causation" form. If fallacious, it is an example of "Post hoc ergo propter hoc" thinking. To my mind, correlation requires interval or ordinal data, so that we see "as x gets higher (or lower), y gets higher (or lower)," neither of which is true of the two variables in the comic. The "this happened, then that happened, therefore this caused that" is the classic post hoc fallacy, not the correlation fallacy. But I still laughed out loud.

I'm going to assume that every flap of butterfly wings changes everything and that post hoc is always ergo propter hoc.

[Voltaire II]

I have to say that it would be funnier if he said 'association', because correlation is the linear relationship between two quantitative variables, and association is just the relationship between two variables. Even a parabolic association could apply in this case; but if someone already said that, pretend that this post doesn't exist.

[aliosha]

Here's one for you:

http://news.bbc.co.uk/1/hi/health/8644016.stm

Research in Archives of Internal Medicine shows those who eat at least a bar every week are more glum than those who only eat chocolate now and again.

But they say they cannot rule out that chocolate may be a cause rather than the cure for being depressed.

AAAAAAARRRRGH!

[mmmcannibalism]

This comic reminds me of the thoroughly well-proven fact that the decline of 18th century style pirate is the direct cause of global warming, as illustrated by this graph:

Lol... I liked the comic, but I liked this post more... this graph makes no sense! It looks like they put the X-axis on the data labels, put the data values along the X-axis, then just made whatever curve they thought looked pretty. I don't know what's worse, the ludicrous statement they were mocking or the patently incorrect way they did so.

the horizontal axis doesn't have to go from zero on the left to positive numbers

also, I love pasta

[vookaloop]

This comic reminds me of the thoroughly well-proven fact that the decline of 18th century style pirate is the direct cause of global warming, as illustrated by this graph:

Lol... I liked the comic, but I liked this post more... this graph makes no sense! It looks like they put the X-axis on the data labels, put the data values along the X-axis, then just made whatever curve they thought looked pretty. I don't know what's worse, the ludicrous statement they were mocking or the patently incorrect way they did so.

the horizontal axis doesn't have to go from zero on the left to positive numbers

also, I love pasta

True, but it's still nonsense -- took me a while to see why, keep looking

[mmmcannibalism]

Spoiler:

This comic reminds me of the thoroughly well-proven fact that the decline of 18th century style pirate is the direct cause of global warming, as illustrated by this graph:

Lol... I liked the comic, but I liked this post more... this graph makes no sense! It looks like they put the X-axis on the data labels, put the data values along the X-axis, then just made whatever curve they thought looked pretty. I don't know what's worse, the ludicrous statement they were mocking or the patently incorrect way they did so.

the horizontal axis doesn't have to go from zero on the left to positive numbers

also, I love pasta

True, but it's still nonsense -- took me a while to see why, keep looking

Are those temperatures too low?

[Elvish Pillager]

The scale on the X axis isn't even monotonic...

[gypkap]

On Prometheus's graph, you could replace the X Axis label "Number of Pirates" with "Furlongs per Fortnight," and the graph would make just as much sense. I like.

[vookaloop]

Spoiler:

This comic reminds me of the thoroughly well-proven fact that the decline of 18th century style pirate is the direct cause of global warming, as illustrated by this graph:

Lol... I liked the comic, but I liked this post more... this graph makes no sense! It looks like they put the X-axis on the data labels, put the data values along the X-axis, then just made whatever curve they thought looked pretty. I don't know what's worse, the ludicrous statement they were mocking or the patently incorrect way they did so.

the horizontal axis doesn't have to go from zero on the left to positive numbers

also, I love pasta

True, but it's still nonsense -- took me a while to see why, keep looking

Are those temperatures too low?

The only thing I see that breaks the graph is that the x-axis goes from 35K to 45K then back down to 20K therefore there are 2 different temperatures indicated for all the numbers between 35K and 45K.

edit: On second thought maybe it's not broken. This represents the same data but makes a little more sense.

[mmmcannibalism]

Oh I see it now, one of the data points conflicts with the claim that pirates directly effect the temperature since their number went up along with temperature.

[phlip]

How about this:

[hatten]

Shouldn't the temperature go down again with the pirate parties, the pirate bay, and all illegal downloaders (eg pirates)

[mmmcannibalism]

Shouldn't the temperature go down again with the pirate parties, the pirate bay, and all illegal downloaders (eg pirates)

Only jolly seafarers who sail the oceans with cutlasses in hand are true pirates; downloaders are just nerds and the somalians are just guys with ak47's

R amen

Also, I love my avatar right now.

[BrianB]

Was reminded of xkcd with this morning's Dilbert comic:

[addams]

This is a Great one to bring out of the depths of time.

Thank you.

[VectorZero]

Banner outside a church on my way to a new job.

[jyrbain]

i actually very recently got into an argument about this in a chat. turns out, the actual statistical definition of "implies" is different from the day to day usage, and it does actually mean "proves", which is what my argument was, that it does imply causation, but doesnt prove it.

Thats why Im very careful with the word implies. I use "infers", where most people say "implies", also, "indicates" is abit stronger than infers, or "seems to indicate" if you're really not sure, but have a good guess.

[Pfhorrest]

Thats why Im very careful with the word implies. I use "infers", where most people say "implies", also, "indicates" is abit stronger than infers, or "seems to indicate" if you're really not sure, but have a good guess.

"Infer" is something that a reasoner does from one proposition to another, not something one proposition does to another. That is, you can infer from one proposition to another, but a proposition cannot infer another proposition. "Suggests" may be a better word for you to use there if you want to avoid "implies".

Also, "imply" doesn't really mean "prove" in any sense. Correlation cannot prove causation because causation cannot be proven, only disproven, and perhaps suggested. "Entails" would entail "proves", but still isn't quite the same thing, as it is a nontrival matter to prove that one proposition entails another; a proof is a step-by-step demonstration that one (set of) proposition(s) entails another. Entailment is necessary implication, and would be even harder to prove than implication; even if you could show that one proposition was never true when the other was false, which would show that the latter implied the former, to show that the latter entailed the former you would somehow need to show that it could not possibly be any other way, which is pretty much only doable if any other way can be shown to entail a logical contradiction.

[Quizatzhaderac]

Causation -> correlation.

Nope, doesn't work that way either. Let's say you have three variables.
A - the force you use to push a cart
B - the combined mass of the cart and it's contents
c - The speed the cart moves.

Now suppose we doubled A and quadrupled B,
Higher A causes higher C, however in this case as A goes up C goes down.

Correlation isn't a measurement of one instance.

I could give you any number of data points you like, if you don't control other variables correlation isn't essential to variables.
If you were doing an experiment of pushing a cart and measuring it's speed, you could still get a negative correlation if I'm there messing with the masses each time.

[phlip]

But if you're changing the masses in response to how much force people are applying, then you've changed causation too - more force indirectly causes greater mass, causes less acceleration, and the correlation will bear that out.

And if you're not changing the masses in response to how much force people are applying, then there shouldn't be any correlation between the two (for a well-designed experiment), so the law of large numbers should prevail, and there'll still be a positive correlation between force and acceleration. Your interference will just be a source of noise.

[djonasty]

The FDA likes it:

PPT on mandatory reporting of clinical trials. (Slide 25).
http://prsinfo.clinicaltrials.gov/webinars/module1/resources/Overview_Handouts.pdf