Killing people is the best solution.

Define "cause." You dismissed my earlier explanation as "semantics," but now you want to argue the same point.

I don't really think the definition of "cause" is in dispute in the scenario I posed, but if you'd rather not answer I'm not going to push it.

I've already stated that correlations can inform one's decisions. If allergy attacks are associated with peanut intake, then by all means avoid peanuts.

But allergies require a "sensitizing dose" as well as a "shocking dose." Is the shocking dose the cause or is the sensitizing dose the cause? You call this "semantics" but for doctors seeking to understand the actual mechanisms, it is important.

Cause-and-effect is very hard to pin down.

Even simple things are difficult to pin down, if you really look at them.

I flip a switch, the light comes on.

But if the electricity is out, the light doesn't come on. The switch is not the cause of the light. A complete electrical circuit, with voltage source and a device capable of emitting photons, is the cause.

All of those details can be discarded. You can take this and genericize it all the way and the scenario still stands.

You have to guess between two possible causes. One has a high correlation with the effect, the other has a low correlation. With no other information and while required to guess to win the million you would pick the higher one because it does give you some information about probability between causes. It's not infallible information, hell it's not even very reliable. But it is some information.

Anywho, this is going nowhere fast so I'm going to try to extricate myself from this debate.

And the higher correlation means that the two events are more closely related. That's what correlation measures.

This debate seems to be ending, which is fine, because I really don't need to add anything to it. But I am bored, and will chime in on a couple things

The following is what I believe to be fact:

1. In most of the examples I read nobody was proving correlation. Saying "when I do X, Y happens" doesn't prove correlation. For example, ringing a doorbell and hearing the chime doesn't prove they are correlated, let alone causation. To prove correlation, you would have to prove that "the more X happens, the more Y happens". So you would have to prove that as the amount of times the doorbell rings increases, so does the amount of times the button is pushed. However this really wasn't what the debate was about, and doesn't change anything about the points being made, so it doesn't matter. I'm Just saying...

2. In the ice cream / shark attack example, it was said that while they were correlated, they were actually caused by warm weather and people going to the beach. This is false. They are simply more closely correlated to warm weather then to each other.

3. If 2 things are proven to NOT be correlated, then it is IMPOSSIBLE for there to be causation. It's not a matter of "in most scenarios". If there is Causation, there is correlation. If there is not correlation, there is not causation. Guaranteed.

4. Correlation isn't a binary thing. Its not a matter of either they are correlated, or they aren't. There are levels. This is important, because the higher the correlation between 2 things, the greater the probability that there is causation. This is evident in Roberts peanut example. He was implying that aspirin and allergies were more closely correlated then peanuts and allergies, and this fact is useful.

So yeah, those are the thoughts that popped into my head while reading. I'm not chiming in on the actual debate itself, as I find it too silly.

Too silly? Hmmm.

I think there's a good reason this is being discussed in such depth, and it's not all about ego.

It's because statistics are so easily and readily abused, and used as propaganda. When a graph is published that appears to suggest an underlying truth (assumed causation), people tend to accept it, at least in part, but it's possible to juxtapose two truths in such a way that the result is a lie.

It's not silly when people are being executed as a result of fuzzy statistical analysis, it's tragic.

It was just such a correlation between "declining moral standards" and teaching evolution in schools that led to creation of the Flying Spaghetti Monster and the Pirate/Global Temperature graph. It took that graph to demonstrate to the public and the Kansas School Board that correlation does not equal causation.

It is in precisely these situations , (death penalty v murder rate), that we need to apply the most rigorous analysis. Lazy thinking, assuming some probable causation without further evidence, is perpetrating the common assumption that Capital Punishment is an effective deterrent, and jeopardizing it's potential elimination.

Hoist on my own petard!

It wasn't the fact that it's a debate about correlation vs. causation that made it seem silly to me. It was the fact that I honestly feel that DrewDad and Robert are on the same side, pretty much. They both would agree that proving correlation does NOT prove causation. Instead they were arguing about what information, if any, correlation does give you. However, when given concrete examples, I feel they STILL show up on the same side. So the argument is about weather the information is coming from the correlation, or other known data. At this point we are pretty far from the statement "Correlation does not prove causation" and because I think that is really the heart of the matter, the argument is silly.

Who rains on whose parade?

What he said.

If that bothers you, then you won't be a fan of most of my posts. I tend to type quickly, trying to make sure my fingers can keep up with my thoughts. After that, I go back through and correct everything that firefox underlines as a spelling mistake. This means that my posts may have grammatical errors. I have always been bad with grammar, and I don't consider posting on forums important enough to bother spending alot of time proofreading and making sure everything is perfectly correct. If I did, it would feel too much like homework, and I wouldn't bother posting at all. To those that this bothers, sorry.

It doesn't bothers me the least, Nick.

I just found it funny and hope my failed attempt to be jocous doesn't bother you either..

Sorry, I just misinterpreted your post. Poking fun of the strange sentences I create is no problem at all (I can take a roasting as well, just ask Robert). I've ran into some people that are truly upset that my posts contain errors, and mistakenly thought you were one of 'em.

No Worries!

Are you, like, Australian or something?

He might have picked that one up from me.

Very true, my aversion to absolutism in language got the better of me there and I softened the absolutism erroneously.



This, to me is a big part of where the argument lies.

- We all know correlation doesn't equal causation.

- We all know that correlation doesn't imply causation, as long as you are using a statistician's meaning of "imply".

- The colloquial meaning of "imply" is to suggest, and yes correlation can suggest causation. This is what is in dispute.

I'll let Edward Tufte's quotes summarize my position, he considered the "correlation does not imply causation" maxim to be misleading, and proposed these two alternatives:



That's where we are stuck, the very different meanings of "imply" are confusing the issue. Correlation isn't causation, we all know that. But it does give us information about causation. It's not a binary thing, and the differences between correlation coefficients can give us information about causation (even if it's not infallible).

I think of it like a "warmer, warmer, getting hot..." for causation with the caveat that is often misleading.

That's what the peanut/aspirin example was trying to illustrate. Both had a correlation one much stronger than the other and the difference is suggesting something about the relative likelihood of causation.

Of course, that's why DrewDad never answered the aspirin/peanut butter scenario. But that is what we were arguing about.

And to be fair, Bill probably thinks that's the same reason I didn't answer his pre-1996 smoking question, and implied that my memory was selective. But I did not have access to such reading material till 1994 and really don't know what correlation data was available before causation was established.

With the hindsight that the established causation provides, I'm not sure how valuable my answer is but if I had to answer Bill's question I would say that yes, the correlation was enough for a reasonable person to believe that it implied (colloquial) causation. (Your turn Drew for the cool million)

Thing is, I've seen the medical community get this wrong so many times (not their fault, they have to study based on correlation for ethical reasons) with things like high power lines, silicone implants and all that I'm not going to vouch for their correlation data without having reviewed it. I've reviewed their causation claims to my satisfaction and with that felt like the correlation study is moot.

I also haven't seen convincing data from Bill in regard to the murder/capital punishment correlation and don't agree with that either. This puts me in the odd spot of agreeing with Drew about the graph not being very meaningful, but agreeing with Bill to the effect that correlation can reasonably suggest causation.

Drew, pony up to this already so that we can get to Bill's graph. There's easier ways to argue against his conclusion than the whole "imply" logomachy.

I agree one hundred percent with this.

I've also stated that correlations (especially strong correlations) can inform one's behavior without causality having to be established. (Barometric pressure and weather. Robert's peanut/aspirin example.)

I do not agree that proving correlation suggests, implies, or hints at a causal relationship of any kind.

I refer again to this analogy:

If the letter "A" is always followed by the letter "B", and you see a letter "B", what can you assume about the letter preceding "B"?