| HN Mirror

http://en.wikipedia.org/wiki/Correlation_does_not_imply_caus...

> Predict is something you can do if something is "correlated"!

This is false. I saw a puddle, so I predict that the puddle will cause rain. See the problem? You can't "predict" until you have a cause-effect relationship.

A correlation is not a cause-effect relationship:

A quote: "The opposite belief, correlation proves causation, is a logical fallacy by which two events that occur together are claimed to have a cause-and-effect relationship. The fallacy is also known as cum hoc ergo propter hoc (Latin for "with this, therefore because of this") and false cause. It is a common fallacy in which it is assumed that, because two things or events occur together, one must be the cause of the other."

> You can see how "predict" is a word for an action you can take if something is correlated.

You may be able to do that. Scientists require evidence.

Evidence of a correlation is all that's necessary to do that, scientist or not. What a correlation tells me is that if I observe A, I am likely to observe B. In the case of rain and puddles, there is a temporal component that seems to be confusing you. Rain now is correlated with puddles for a while in the future. Puddles now may not be correlated with rain now or in the future[1]. Observing puddles now, I can't tell you anything about rain in the future because there is no correlation, not because there is no causation.

[1] If the weather in the area tends to be the same day to day, then I can absolutely predict rain tomorrow based on puddles today, to occasionally be frustrated by weather change or sprinklers.

> there is a temporal component that seems to be confusing you.

I am not confused. Confusing correlation with causation is a very common logical error -- you are not alone.

Please respond to the point of my post, if you are going to respond. "I'm not confused, you are" adds nothing to the discussion - obviously, we both think that. And yes, confusing correlation with causation is a common logical error - one that, I believe, you have become oversensitive to such that you are spotting it where it doesn't exist. This, I thought, was already established before your post here.

> "I'm not confused, you are" adds nothing to the discussion

That might be true if anyone had said it (avoid rhetorical escalation).

> And yes, confusing correlation with causation is a common logical error - one that, I believe, you have become oversensitive to such that you are spotting it where it doesn't exist.

The post that started this thread correctly identifies a case of touting a correlation as though it represents a cause-effect relationship. This is very common in popular science journalism, where data sets that barely represent a correlation are described as though they were cause-effect relationships, and begin to shape public policy well in advance of, and sometimes in the perpetual absence of, any effort to establish a cause-effect link between some A and some B.

The entirety of my objection is to the notion that the word "predicts" implies causation. I didn't read the article closely, and assumed it was talking about correlation primarily because of the title. If the article generally is making stronger claims, by all means object, but object to the right thing. "Predicts" carries no implication of causation. "Poll results predict the election" is making a claim about a correlation. "The weatherman predicts the weather" is making a claim about a correlation. Heck, even "the tarot cards predict the future" is making a (false) claim about a correlation. Correlations are real and useful things that do or don't exist in any given case, and which can be used to make predictions. Where causation is necessary is in predicting the results of a change or intervention. If I publish higher poll numbers, that won't make my candidate win. If I make it rain, that will make there be puddles.

> That might be true if anyone had said it (avoid rhetorical escalation).

You have a habit of denying the obvious that is leading me to believe you are trolling, your profile and post history to the contrary.

You said, word for word, "I am not confused."

You then proceeded to say, "Confusing correlation with causation is a very common logical error -- you are not alone." This is quite clearly a statement that I am confused about correlation vs. causation.

Therefore, semantically, you said "I am not confused, you are" plus the well established point that confusion was related to correlation and causation.

meric 5023 days ago

Correlation is statistical evidence.

You broke your own example by inserting the word "cause" into the sentence. Of course your example is a cause-effect relationship.

Statistical evidence suggesting correlation of two factors A and B are enough for one to say "The presence of A predicts B" as well as "The presence of B predicts A".

Please, don't warp the meaning of the word "predict".

"I predict tomorrow is going to rain" There is no way that sentence suggests a cause-effect relationship.

"This specific color pattern in the image is a great predictor of the presence of rain when the photograph was taken" Neither does this sentence suggests a cause-effect relationship.

No one here is confused between correlation and causation. I'm simply insisting the word "predict" has to do with correlation, not causation.

http://abcnews.go.com/blogs/health/2012/08/27/teenage-mariju...

> Correlation is statistical evidence.

Yes -- evidence of a correlation, not a cause-effect relationship.

> Of course your example is a cause-effect relationship.

Puddles and rain? Yes, but it is only a description, not an explanation. Science requires explanations. Otherwise we open the door to pseudoscience, to people claiming any associations they care to claim.

> Please, don't warp the meaning of the word "predict".

It is you who is doing that -- look at the definitions at the bottom of this post.

Here's another example. I have a cure for the common cold -- I shake a dried gourd over the patient's head until he get better. Sometimes it takes a week, but my treatment always works. The correlation is perfect, therefore I deserve a Nobel Prize for ridding the world of this scourge.

My dried gourd treatment "predicts" that the cold sufferer will get better -- always.

> I'm simply insisting the word "predict" has to do with correlation, not causation.

And you are mistaken. Rain is a predictor for bumper crops, but bumper crops are not a predictor for rain. Teenage driving is a predictor for car crashes, but car crashes are not a predictor for teenage driving.

A recent bogus study found a correlation between marijuana use and lower IQ. But the marijuana use did not predict the IQ drop, it was only correlated with it, and the researchers included this fact in their article. Needless to say, the science journalists ignored the qualifiers in the article and announced that marijuana use predicted a fall in IQ:

Here's another account of the same study that makes a claim in its title that the article body contradicts. Title: "Smoking Pot In Teen Years Lowers IQ Later". A quote from the article: "But those who consistently smoke marijuana may simply make less intellectually stimulating choices at critical points in life."

Here is another bogus study: "Low I.Q. Predicts Heart Disease":

http://well.blogs.nytimes.com/2010/02/10/low-i-q-predicts-he...

Except for the fact that it's only a correlation, and use of the term "predicts" is nonsense. Needless to say, the article doesn't consider that the low IQ might predict the heart disease, not the reverse.

> I'm simply insisting the word "predict" has to do with correlation, not causation.

Yes, and you are mistaken.

http://dictionary.reference.com/browse/predict

"to declare or tell in advance; prophesy; foretell: to predict the weather; to predict the fall of a civilization."

http://www.merriam-webster.com/dictionary/predict

"to declare or indicate in advance; especially : foretell on the basis of observation, experience, or scientific reason"

Q.E.D.

http://en.wikipedia.org/wiki/Prediction#Statistics

meric 5023 days ago

You can't try to refute only some of my points and not others.

I meant to say your example of course implied a cause-effect relationship, since you added the word 'cause' to it. (of course it wasn't a cause-effect relationship, do you think I am stupid?)

>> My dried gourd treatment "predicts" that the cold sufferer will get better -- always.

If your treatment is indeed correlated with a cold sufferer getting better, significantly enough, then yes, your treatment does in fact predicts it, even though it might not have been the cure. This is consistent with the definition of prediction being only related to correlation.

>> Rain is a predictor for bumper crops, but bumper crops are not a predictor for rain. Teenage driving is a predictor for car crashes, but car crashes are not a predictor for teenage driving.

That is correct; since bumper crops were only aided by rain if factors X1, Y1 are true, bumper crops are not a predictor of past rain unless X1, Y1 is true. Car crashes are correlated with teenage driving if factors X2, Y2 are true, car crashes are not a predictor of teenage driving unless X2, Y2 is true.

X1, Y1 = (Crops are water dependent, crops were not also thoroughly irrigated by other means)

X2, Y2 = (Teenager was intoxicated, teenager riding in car with more than 3 other members, all male)

>> But the marijuana use did not predict the IQ drop, it was only correlated with it, and the researchers included this fact in their article.

Yes it did, correlation means you can use the statistics to predict it, even if it wasn't the cause.

Predict means correlation, and correlation means what you think correlation means. You have the meaning of predict wrong. It has as much to do with causation as correlation has. i.e not much (though some nonetheless, since if there is a causation it is likely to be correlated).

"In statistics, prediction is a part of statistical inference. One particular approach to such inference is known as predictive inference, but the prediction can be undertaken within any of the several approaches to statistical inference. Indeed, one description of statistics is that it provides a means of transferring knowledge about a sample of a population to the whole population, and to other related populations, which is not the same as prediction over time."

If you are talking about prediction and correlation in the same sentence, you are in the realm statistics and therefore should abide by its use of language.

You also did not address my two examples of use of predict in laymen sentences showing the word "predict" did not have anything to do with causation.

>> "to declare or tell in advance; prophesy; foretell: to predict the weather; to predict the fall of a civilization."

>> "to declare or indicate in advance; especially : foretell on the basis of observation, experience, or scientific reason"

Neither of those definitions have any semblance of implying causation. Observation, experience, or scientific reason can all be instances of correlation.

http://dictionary.reference.com/browse/predict

> You can't try to refute only some of my points and not others.

Of course I can. I chose only those points where your logical errors were most obvious.

> If your treatment is indeed correlated with a cold sufferer getting better, significantly enough, then yes, your treatment does in fact predicts it, even though it might not have been the cure. This is consistent with the definition of prediction being only related to correlation.

No, that is false. Shall I list the definitions for "predict" again, or will you read them again on your own?

> You also did not address my two examples of use of predict in laymen sentences showing the word "predict" did not have anything to do with causation.

You quoted some laymen who suited your purpose, while I listed the dictionary definitions for the word, the definitions that show that "predict" means to assert an effect based on a cause.

> Neither of those definitions have any semblance of implying causation.

Oh, really?

"to declare or indicate in advance; especially : foretell on the basis of observation, experience, or scientific reason"

A prediction is therefore the use of observations of A to predict B, to show a cause-effect relationship. I see the disappearance of the middle class (A), and on that basis I predict the fall of civilization (B). I see gathering clouds (A), and on that basis I predict rain (B) -- and puddles (C).

> Observation, experience, or scientific reason can all be instances of correlation.

Yes, but it's a false analogy with no bearing on this topic. A prediction forges a link between an observation (A) and an outcome (B). It assumes a cause-effect relationship, one that may not be real, but a word isn't responsible for how people misuse it.

> Indeed, one description of statistics is that it provides a means of transferring knowledge about a sample of a population to the whole population

Yes -- an observation of a small sample (A) is used as the basis for a prediction about the population as a whole (B). Also, remember that "prediction" commonly refers to an assertion about the future (B) based on present observations (A).

"(Verb) to foretell the future; make a prediction."

> Yes it did, correlation means you can use the statistics to predict it, even if it wasn't the cause.

Nonsense. Marijuana use doesn't predict an IQ drop, the study doesn't support that prediction, as the authors were careful to point out, and as the journalists were at pains to ignore.

The marijuana use, and the IQ drop, are only correlated -- one does not predict the other.

meric 5023 days ago

>> A prediction is therefore the use of observations of A to predict B, to show a cause-effect relationship. I see the disappearance of the middle class (A), and on that basis I predict the fall of civilization (B). I see gathering clouds (A), and on that basis I predict rain (B) -- and puddles (C).

You are using the temporal sense of the word "predict", not the cross-sectional sense. Just because you can use the word predict when there is a cause-effect relationship doesn't mean you can't if there isn't. Here is an example illustrating this: I see lots of graffiti in the town (A) and on that basis, predict this town has a high crime rate (B). Notice this prediction was made independent of time.

>> A prediction forges a link between an observation (A) and an outcome (B). It assumes a cause-effect relationship, one that may not be real, but a word isn't responsible for how people misuse it.

A prediction forges a link between an observation (A) and an outcome (B) to explain a correlation relationship, which may or may not be because of a cause-effect relationship.

>> Yes -- an observation of a small sample (A) is used as the basis for a prediction about the population as a whole (B). Also, remember that "prediction" commonly refers to an assertion about the future (B) based on present observations (A).

Let's say I am a statistician and after surveying 10% of the population, found out lower income earners are correlated with a lower IQ. I use this observation as a basis for a prediction about the population as a whole - that lower income earners can predict a lower IQ. Notice again, time is irrelevant.

>> Also, remember that "prediction" commonly refers to an assertion about the future (B) based on present observations (A).

A word may have more than one sense. I am talking about the word prediction as used in statistics.

>> The marijuana use, and the IQ drop, are only correlated -- one does not predict the other.

If they are correlated then you can use the evidence in the sample to make general predictions in the population (independent of time), provided your experiment methodology was valid.