Killing people is the best solution.

@Robert Gentel,

Robert Gentel wrote:

Drewdad wrote:
There can be no conclusion made regarding the existence or the direction of a cause and effect relationship only from the fact that A and B are correlated.

This is an inaccurate statement.

Nope. Correlation by itself gives you no information about cause and effect.

Robert Gentel wrote:

In most scenarios if there is no correlation there is no possible causation.

Agreed.

Robert Gentel wrote:

So the mere existence of a correlation allows you to conclude that a causal relationship is possible.

"Possible" is another of those slippery words. If there is no causal relationship, then a causal relationship is not "possible", even if there is a correlation.

Robert Gentel wrote:

This can be useful to the process of elimination.

Agreed.

Robert Gentel wrote:

For example, allergies are hard to pin down to an allergen. The best way to do this would be skin tests or blood tests but something that can help narrow it down is correlation.

If I never ate peanuts, then it can be eliminated as a cause. If I always took aspirin before the allergic reaction that would be the first thing I'd want to test.

Agreed.

Robert Gentel wrote:

There is a relationship between correlation and causation DrewDad,

Agreed. Causation requires correlation.

Robert Gentel wrote:

it's not nearly as strong as most people treat it and that's why there's a lot of "correlation does not equal causation" rules in logic. But saying there is no relationship is also an overstatement.

I have not stated that there is no relationship. I stated that correlation alone provides no information regarding cause-and-effect.

Robert Gentel wrote:

There is a relationship between correlation and causation and correlation is often used to help narrow down the items to test against causation.

Yes. If there is no correlation, then there is no cause-and-effect.

Robert Gentel wrote:

All of that being said, the graph Bill posted doesn't provide much insight. We already know that it's theoretically possible for capital punishment to decrease crime, so showing a correlation doesn't give us any new information.

It doesn't tell us much because the direction of the relationship is not established. For example, as murder rates rise perhaps they trigger harsher law enforcement and punishment (including capital punishment) and perhaps it's the harsher law enforcement that is reducing the murder rate. Perhaps the increase in the murder rate is what triggers society to re-institute the capital punishment and the other elements of their response to the crime wave (increased incarceration, more police etc) might be responsible for the subsequent reduction in the murder rate.

Yes. I will also point out that executions and murders are definitely linked, in that if we had zero murders we would then have zero executions.

Robert Gentel wrote:

This is not something I'm arguing with you, but being dogmatic about correlation not having anything to say about causation is as wrongheaded as thinking that it always does.

I never stated that "correlation never has anything to say about causation."

I stated (and maintain) that a proving correlation provides no information about whether a causal relationship actually exists.

Robert Gentel wrote:

Correlation does say something about causation, it doesn't say that it is there but it does say something about the likelihood of it being there.

Only in terms of being able to rule out causation.

Robert Gentel wrote:

After all, with no correlation there's often no possibility of causation so at the very least it's something you can use in the process of elimination.

That's what I'm saying.

2 Replies

@DrewDad,

Let's say that in a portion of text, the letter "A" is always followed by the letter "B".

If I tell you that I see a letter "B", can you tell me the letter that proceeds it? No. The letter "B" provides no information about the preceding letter.

Causation is always accompanied by correlation.

If you find a correlation, what can you assume about causation? Nothing. Correlation provides no information about the causal relationship.

2 Replies

@DrewDad,

DrewDad wrote:

Nope. Correlation by itself gives you no information about cause and effect.
[...]
Agreed. Causation requires correlation.

Then correlation gives you some information that reflects on causation. You can dispute how much that information is worth, but saying that it doesn't give any information about that is an overstatement.

Quote:

I have not stated that there is no relationship. I stated that correlation alone provides no information regarding cause-and-effect.

Well you really don't every get a scenario where there is no information other than that. So let's try an example scenario with little other than correlation as the input data:

You are told that if you guess between two possible causes of someone's allergic reaction, you will win a million dollars. The choices presented to you are peanuts and aspirin. Both are common allergens and you are given this additional information:

90% of the times the individual takes aspirin, he gets an allergic reaction. 0% of the times the individual ingests peanuts he gets an allergic reaction.

Which would you pick? And if you did not arrive at this pick through the use of the correlation data what are you using?

Quote:

I stated (and maintain) that a proving correlation provides no information about whether a causal relationship actually exists.

Ok, so if you are playing poker and you notice that almost every time someone bets all in they are bluffing and you have a decent hand and they go all in would you say you have no information to work with or would the correlation give you some indication about the probability that the player is bluffing?

It does give information DrewDad, because correlation shares a relationship with causation. The information it gives you is often not strong enough to draw any conclusions about causation, but it can change the probabilities you will find a causal link.

In my peanuts or aspirin scenario, you have one shot at a million dollars. You are clearly going to be guessing, and solid evidence of cause and effect are not going to be available to you.

The choice is a no-brainer based on the correlation information you have been given, because as you know there is a relationship.

To say that it gives no information is an overstatement. If there is a relationship it's going to give you some information about the likelihood of finding a causal link.

Quote:

Only in terms of being able to rule out causation.

Being able to rule out causation is information about causation. If correlation can give you this data then it stands to reason that correlation can sometimes give you information about causation.

1 Reply

@DrewDad,

Clarification: "Correlation provides no information about the causal relationship" is meant in regard to when a correlation is established.

Absence of correlation implies absence of cause-and-effect.

1 Reply

@DrewDad,

DrewDad wrote:

Causation is always accompanied by correlation.

If you find a correlation, what can you assume about causation?

That absent any other information I can't rule out causation.

Quote:

Nothing.

Well DrewDad, that is something but this doesn't seem to be going anywhere so we'll have to agree to disagree then.

0 Replies

@DrewDad,

DrewDad wrote:

Clarification: "Correlation provides no information about the causal relationship" is meant in regard to when a correlation is established.

Absence of correlation implies absence of cause-and-effect.

Even with this clarification, there's still data to be had as related to probability.

If I told you that I ate a peanut and an aspirin and had an allergic reaction and that once out of a 100 times I ate peanuts in my life I had an allergic reaction but 99 times out of 100 times I took aspirin I had an allergic reaction which would you pick for the million? Would you find it just as likely that the peanut caused the reaction?

It's certainly possible, and the correlation alone doesn't prove anything one way or the other but there is a relationship between correlation and causation and there is some level of information being given by correlation about causation.

0 Replies

@Robert Gentel,

Robert Gentel wrote:

DrewDad wrote:
Nope. Correlation by itself gives you no information about cause and effect.
[...]
Agreed. Causation requires correlation.

Then correlation gives you some information that reflects on causation. You can dispute how much that information is worth, but saying that it doesn't give any information about that is an overstatement.

No, absence of correlation gives you information that reflects on causation.

Robert Gentel wrote:

Quote:
I have not stated that there is no relationship. I stated that correlation alone provides no information regarding cause-and-effect.

Well you really don't every get a scenario where there is no information other than that. So let's try an example scenario with little other than correlation as the input data:

You are told that if you guess between two possible causes of someone's allergic reaction, you will win a million dollars. The choices presented to you are peanuts and aspirin. Both are common allergens and you are given this additional information:

90% of the times the individual takes aspirin, he gets an allergic reaction. 0% of the times the individual ingests peanuts he gets an allergic reaction.

That is not just a correlation, so you are carrying the discussion beyond what one can determine from correlation.

Robert Gentel wrote:

Drewdad wrote:

I stated (and maintain) that a proving correlation provides no information about whether a causal relationship actually exists.

Ok, so if you are playing poker and you notice that almost every time someone bets all in they are bluffing and you have a decent hand and they go all in would you say you have no information to work with or would the correlation give you some indication about the probability that the player is bluffing?

You haven't given any information on how often the player goes all-in without bluffing, so correlating bluffing with all-in provides no information on whether all-in is caused by bluffing.

Perhaps they go all-in on every hand. (Which would make for a short game, but still.)

Robert Gentel wrote:

To say that it gives no information is an overstatement. If there is a relationship it's going to give you some information about the likelihood of finding a causal link.

No, Robert, it doesn't.

Ice cream sales and shark attacks are correlated. Shark attacks are not caused by ice cream sales, and ice cream sales are not caused by shark attacks. They are both caused by warm weather and people going to the beach.

The number of cavities in elementary school children and vocabulary size have a strong positive correlation. Kids don't get cavities because they talk; cavities and vocabulary both increase as kids get older.

Global temperatures are correlated with the decline in the pirate population. So what?

A correlation, by itself, provides no information about causality.

Robert Gentel wrote:

Drewdad wrote:
Only in terms of being able to rule out causation.

Being able to rule out causation is information about causation. If correlation can give you this data then it stands to reason that correlation can sometimes give you information about causation.

Every time I've said "correlation provides no information about causation" I have been referring to correlation being present, not to correlation being disproved. I have been consistent on this point, even before you joined the conversation.

Drewdad wrote:

There can be no conclusion made regarding the existence or the direction of a cause and effect relationship only from the fact that A and B are correlated.

2 Replies

@DrewDad,

DrewDad wrote:

Every time I've said "correlation provides no information about causation" I have been referring to correlation being present, not to correlation being disproved. I have been consistent on this point, even before you joined the conversation.

I modified my aspirin/peanut question to you to include correlation on both items. So which would you guess and why?

1 Reply

@DrewDad,

Robert Gentel wrote:

Ok, so if you are playing poker and you notice that almost every time someone bets all in they are bluffing and you have a decent hand and they go all in would you say you have no information to work with or would the correlation give you some indication about the probability that the player is bluffing?

Whoops! I mis-read.

Again, this is more than a correlation. You have the correlation (how you achieved it, I don't know. To prove bluffing, they would have to lose the hand, which means they'd be out of the game.) You then form a hypothesis, which you are intending to test.

1 Reply

@Robert Gentel,

Robert Gentel wrote:

DrewDad wrote:
Every time I've said "correlation provides no information about causation" I have been referring to correlation being present, not to correlation being disproved. I have been consistent on this point, even before you joined the conversation.

I modified my aspirin/peanut question to you to include correlation on both items. So which would you guess and why?

It's still more than correlation. You have a well-established medical theory that certain foods cause allergic reactions. You've ruled out everything except peanuts. Now you plan to test your theory by eliminating peanuts from your diet.

1 Reply

@DrewDad,

Quick clarification on poker, there are a number of ways you can know they are bluffing without the person being out:

1) He can show the bluffs
2) He can win even though he bluffed
3) He can lose to smaller stacks
4) He can be playing a cash game and rebuying

And there's not more information to go off of (that I gave) other than the correlation, but the peanut/aspirin example is a better one to answer anyway.

1 Reply

@Robert Gentel,

I've figured out what's been bothering me with these scenarios.

We're not talking about cause-and-effect.

I never claimed that correlation cannot be informative, or that one cannot make decisions based on correlations.

With poker, the cause of going all-in is some process in the player's mind. Low cards do not decide to bluff; the player does. All-in might correlate to low cards, but saying that the player going all-in is a result of low cards would be false.

If the barometer changes, I expect the weather to change. The barometer doesn't cause the weather, but I can still know that something is going to happen.

Correlations tell you that things are related, and you don't have to know why they are related in order to make decisions based on them.

But correlations, by themselves, do not inform you as to the cause of the relationship.

0 Replies

(All the same, I can't go wiggle the needle on my barometer and change the weather. This is why Bill needs to establish cause-and-effect on his graph.)

1 Reply

@DrewDad,

DrewDad wrote:

It's still more than correlation. You have a well-established medical theory that certain foods cause allergic reactions. You've ruled out everything except peanuts. Now you plan to test your theory by eliminating peanuts from your diet.

I modified it so that both peanuts and aspirin can't be ruled out. One has a 1/100 history of being correlated to the allergic reaction, the other has a 99/100. The real cause could be neither but you are gambling for a million on a guess between them.

If you had to guess, why wouldn't you think that one is more likely to be the culprit than the other based on correlation alone? Do you think they have equal chances of being the cause even though the correlations are vastly different?

1 Reply

@Robert Gentel,

Eating peanuts isn't the cause, though. The cause is a mistake in the immune system, which makes it produce the wrong chemicals when you eat peanuts.

Eating peanuts might be correlated with the allergic reaction, and one can avoid the allergic reaction by avoiding peanuts, but that's just a palliative.

1 Reply

@DrewDad,

DrewDad wrote:

(All the same, I can't go wiggle the needle on my barometer and change the weather. This is why Bill needs to establish cause-and-effect on his graph.)

That's a poor analogy because Bill's theory isn't that outlandish and the relationship he's commenting on is fairly well established in general, even if the particular effects he was touting are, in my opinion, exaggerated. The severity of the punishment does have some relationship with the willingness people will have to commit the crime so it's not like wiggling your barometer, we know that the severities of the punishments can make a difference and the question is how much, not whether it's possible.

If the punishment for murder were 5 minutes in jail do you think murder would be more common? I don't personally think there's a huge increase in deterrence between life sentences and death sentences but it's not outlandish to assert that there is a cause and effect.

1 Reply

@DrewDad,

That's a bit of semantics really, the peanuts are "causing" the body to do what is causing the reaction. But eliminate all that and the scenario stands.

You have potential allergen a, with a history of causing a reaction 1/100 times and you have potential allergen b, with a history of 99/100. Wouldn't you pick allergen b if you had to guess?

1 Reply

@Robert Gentel,

Again, I didn't say that correlations cannot inform one's decisions.

1 Reply

@DrewDad,

Yeah, but you said they gave no information about causation. So in the scenario you are trying to guess which one causes the reaction. Does the great different in correlation not give you any information about which is the more likely cause?

1 Reply

@Robert Gentel,

As I've stated before, my personal belief is that life without parole should replace the death penalty. My main reasons are cost (execution costs more than life imprisonment) and error correction (if someone is wrongfully convicted, new forensic techniques might allow us to detect it, rectify the error, and seek the real criminal).

The problem with Bill's graph, is that even if a correlation is established, it tells us nothing about cause and effect. Bill claims a strong one-way causality, that more executions result in fewer murders; this claim cannot be supported just on the basis of a correlation.

0 Replies

Killing people is the best solution.

Related Topics

Quick Links

My Account

able2know