Killing people is the best solution.

@JPB,

JPB wrote:

Ok. But the trouble begins when you start drawing conclusions from these non-statistical suggestions.

Depends on what conclusion you draw. If you are drawing the "it's causality" conclusion, agreed but I can think of others where it's not always the case.

Quote:

The highlighted phrase is what brought me back into this thread. Again, colloquial interpretations is one thing. Conclusions drawn on inferences that don't have statistical meaning is another. Just don't start using the word "significant" and I'll go back to my corner.

Not a "colloquial" interpretation of data, you've conflated the colloquiality of the logomachy over "imply" (that Drew disavows but that is a polemic part of that maxim among some of the scientists I quoted) to the interpretation and are portraying the interpretation as being non-scientific.

Even in science, the correlation has statistical meaning as it relates to causation. I don't know what exactly you mean by "significant" (though I think I get your drift) but it's something. I wonder if anyone's has calculated the correlation between correlation and causation....

1 Reply

@DrewDad,

DrewDad wrote:

I see immediate problems with both scenarios. I don't think this is something that lends itself to empirical testing.

Of course it is! You are disputing claims about probability, and this kind of simple probability can easily be tested by brute force with a very low margin of error.

Since I can do the work for the a/b scenario fairly easily, what problems do you see in my test method?

3 Replies

@Robert Gentel,

Because you are assuming that causation is randomly distributed among all correlations. Or you are making some assumption about the distribution.

0 Replies

@Robert Gentel,

Let's go back to this:

Drewdad wrote:

What does it mean if no correlation is found?
If no correlation is found, then one can rule out any kind of cause-and-effect relationship.

What does it mean if correlation is found?
If a correlation is found then one of several scenarios is in play:

1. B is dependent on A. (A change in variable A causes a change in variable B.)
2. A is dependent on B. (A change in variable B causes a change in variable A.)
3. A and B interact with each other. (A change in variable A causes a change in variable B, which causes a change in variable A, etc.) (Self-reinforcing)
4. A and B are mutually dependent on something else. (Unknown third factor)
5. The relationship between A and B is so complex that while they appear related, it is impossible to determine what the actual relationship is. (Coincidental. "Co" meaning "together" and "incident" meaning "occurance". Not coincidental meaning "accidental".)

The problem with making any assumption based on just a correlation is that you are then making the assumption about the distribution of these five options. As if you can assign a probability to each of them. But from correlation alone, you have no information about the distribution of the options.

1 Reply

@Robert Gentel,

Robert Gentel wrote:

Of course it is! You are disputing claims about probability, and this kind of simple probability can easily be tested by brute force with a very low margin of error.

I see an immediate selection bias. How do you chose among the vast number of correlative studies, and how do you know which of these studies can be shown to have causation?

For that matter, there can be correlative studies where causation is present, but where it could not be proved.

Robert Gentel wrote:

Since I can do the work for the a/b scenario fairly easily, what problems do you see in my test method?

Because "A" is always followed by "B" makes no claims as to the randomness of the text. "ABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB" is just as valid as "ABCDEFGHIJKLMNOPQRXTUVWXYZ." That's the point.

1 Reply

@Robert Gentel,

Quote:

Depends on what conclusion you draw. If you are drawing the "it's causality" conclusion, agreed but I can think of others where it's not always the case.

Of course, but then there are generally other confounding inputs where the causality has additional evidence beyond the simple relationship between the two variables of interest (you've stated as much previously).

My reference to significance is again a statistical term that is oftentimes brought into discussions where no formal significance testing has been performed. If you want to show that a correlation of 0.750 is not zero then that's easily done. If you want to compare a correlation of 0.750 to a different correlation coefficient and state that it's significantly different than a correlation of 0.999 or 0.500 then that can also be done, but if you want to make the leap that a correlation of 0.999 makes a stronger case for causality than a correlation of 0.500 then you're on thin ice without additional knowledge of the underlying factors in the data.

1 Reply

@DrewDad,

DrewDad wrote:

Because "A" is always followed by "B" makes no claims as to the randomness of the text. "ABBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB" is just as valid as "ABCDEFGHIJKLMNOPQRXTUVWXYZ." That's the point.

I understand that DrewDad, and probability as expressed in a partial percentage or ratio is never right for one past scenario. It's either 100 or 0 for this kind of thing when it's predetermined.

Not all assumptions are created equal and that's kinda the whole point of probability. So as long as there are diverse scenarios, the assumption can be tested by a statistically relevant sampling of them.

So if I were to make the assumption that higher correlation equals greater probability of causation than not, it won't mean much for one pre-determined scenario. But as long as the sample is diverse at all (contains both scenarios where the B is caused by an A or not) there will be a difference in the overall outcome in a big enough sample if I pick A versus any other letter based on that rule and the greater the sample the more reliably it will show a larger probability for A.

0 Replies

@DrewDad,

DrewDad wrote:

But it still gives you no information about whether causality actually exists, or if it doesn't exist.

It says that it is more probable that it exists than if there were no correlation. Probability doesn't say whether it does or does not exist, it attempts to measure the likelihood of that still unknown scenario.

Probability is information, even if it's unsatisfactorily significant to you.

0 Replies

@JPB,

JPB wrote:

Of course, but then there are generally other confounding inputs where the causality has additional evidence beyond the simple relationship between the two variables of interest (you've stated as much previously).

Sure, like more correlation (an example to make the "only using correlation" hold true) for one.

If you establish that there is a correlation between correlation itself and causality probability in your dataset then a single correlation within your dataset equals higher probability within your dataset.

0 Replies

@DrewDad,

DrewDad wrote:

Let's go back to this:

I read that the last X times you copied it from Wikipedia. If it didn't work then (ascribe it to stubborness on my part if that makes it easier) I don't see how it would work now.

Further down on the same Wikipedia page you are copying are the arguments I've been making but I'm not going to just copy and paste it X times as my argument. It's still there on the page and can be read at anyone's leisure.

Quote:

The problem with making any assumption based on just a correlation is that you are then making the assumption about the distribution of these five options. As if you can assign a probability to each of them. But from correlation alone, you have no information about the distribution of the options.

Really? What if you have correlation information between the individual correlation coefficient for each item in a dataset and the causality?

Do you argue that this scenario is impossible? Or that you can't derive any information about the probability of causality this way?

Just establishing this correlation alone can tell you about distribution. For example it can tell me whether you've intentionally created a dataset where this does not work.

If you read back, I was very careful to say that correlation "can" suggest causation, not that it "does" (as it was couched between you and Bill). Can you envision no such scenario (I'll dust off my imagination if so)?

3 Replies

@Robert Gentel,

I go back to it being an information theory issue, not a statistical issue.

If X then Y (Causality means there is correlation).

Code:



X          Y

1          1

0          1

0          0

X tells you about Y
Not Y tells you Not X.

Knowing only that Y=True provides no information about the state of X.

I'm done with this for tonight. I have to take out the trash, and get the girls ready for bed.

0 Replies

OK, my head officially hurts now.

0 Replies

@Robert Gentel,

Robert Gentel wrote:

If you read back, I was very careful to say that correlation "can" suggest causation, not that it "does" (as it was couched between you and Bill). Can you envision no such scenario (I'll dust off my imagination if so)?

Damn it Robert, that is just not true. I was no less careful to make it very clear that I do not think correlation proves causation, only that it can imply it. The only place you've really deviated from my argument is that you knew about the 1996 discovery and initially ducked the 3 decades that everyone with a speck of common sense recognized that the correlation between smoking and cancer at least implied causation.

I can't tell you how amusing it is to see you have now had practically the exact same dispute with DrewDad I had... And that you've conceded practically every point I made... not conceded so much as presented even better arguments for same. This is the intellectual honesty I was looking for (because I know you disagree with my purpose, but habitually will nonetheless generally admit the obvious side-points that are valid.) This is what I found lacking after several pages and posters chimed in, basically to agree that "Bill must be wrong" (a couple delivering their message by falsely attacking me personally) while all the while ignoring the obvious side-points DrewDad was dogmatically denying (and still is)... which in turn allowed Drew-Dad to assume his cut and paste nonsense was valid as some kind of an absolute law of science. I see he even hit you with the same "Good Lord" as if the scientific community would lose their collective mind because someone pointed out an obvious exception to laboratory rules in real life situations.

Thank you for finally stepping up to the plate. (You too, JBP)

1 Reply

@OCCOM BILL,

OCCOM BILL wrote:

Damn it Robert, that is just not true. I was no less careful to make it very clear that I do not think correlation proves causation, only that it can imply it.

I'll take your word for it Bill, I didn't read the entirety of the exchange and while I remember "does" being used it may have been DrewDad (and as a side note it'd be pretty funny if you guys were also on different does/can pages at times as well).

Quote:

I can't tell you how amusing it is to see you have now had practically the exact same dispute with DrewDad I had...

I know. Once I got into it I knew that wouldn't slip your notice. But my initial comments weren't intended to get me into this debate.

Quote:

This is the intellectual honesty I was looking for (because I know you disagree with my purpose, but habitually will nonetheless generally admit the obvious side-points that are valid.)

I saw the graph, I saw a post by DrewDad with state data or graphs and largely ignored the rest. I was frustrated with the strength of conviction thing ignored almost everything after DrewDad's "taste the shoe moron" (or something like that) post.

When I saw a simple example of fact being rejected with extra strong conviction I popped in. I really didn't want to get into it initially (and my opinion about the edification of doing so hasn't changed) but this weekend there were a couple of times where I had nothing else to do in a hotel in Guatemala the boredom got the better of me.

I don't see this as a matter of intellectual honesty, I see this as me still not having all the self-control I want. I still think I'd be better off doing something else but saw potential common ground every now and then. That those glimpses of progress in the discussion always returned to square one is why I didn't want to jump in in the first place, not because I wanted to deny any part of anyone's argument.

1 Reply

@Robert Gentel,

Robert Gentel wrote:

OCCOM BILL wrote:
Damn it Robert, that is just not true. I was no less careful to make it very clear that I do not think correlation proves causation, only that it can imply it.

I'll take your word for it Bill, I didn't read the entirety of the exchange and while I remember "does" being used it may have been DrewDad (and as a side note it'd be pretty funny if you guys were also on different does/can pages at times as well).

That would be pretty funny, but not possible: I've re-iterated it exactly like that too many times.

Robert Gentel wrote:

Quote:
I can't tell you how amusing it is to see you have now had practically the exact same dispute with DrewDad I had...

I know. Once I got into it I knew that wouldn't slip your notice. But my initial comments weren't intended to get me into this debate.

I can't blame you for that. Your superior communication skills didn't get you any further.

Robert Gentel wrote:

Quote:
This is the intellectual honesty I was looking for (because I know you disagree with my purpose, but habitually will nonetheless generally admit the obvious side-points that are valid.)

I saw the graph, I saw a post by DrewDad with state data or graphs and largely ignored the rest. I was frustrated with the strength of conviction thing ignored almost everything after DrewDad's "taste the shoe moron" (or something like that) post.

When I saw a simple example of fact being rejected with extra strong conviction I popped in. I really didn't want to get into it initially (and my opinion about the edification of doing so hasn't changed) but this weekend there were a couple of times where I had nothing else to do in a hotel in Guatemala the boredom got the better of me.

Time to get your passport stamped again already? I hadn't meant to suggest that you were particularly guilty of intellectual dishonesty, but collectively yeah. Two guys, both behaving like blowhards: One guy gets a pass for denying obvious points from several different posters while they deliver personal shots to the other guy... even though the other guy is actually making more sense. It shouldn’t matter who delivered your doorbell example, for instance, it should be recognized as undeniably obvious. I went with the cancer example because the life-and-death nature of the dilemma lends itself nicely to an argument for the expediency of breaking with traditional (wait-till-we-have-more substantial-proof) theory. I wouldn’t want my cancer specialist to subscribe to DrewDad’s Dogma, while contemplating possible treatments. I’d rather have one who wasn’t afraid to mix in his best judgment.

Robert Gentel wrote:

I don't see this as a matter of intellectual honesty, I see this as me still not having all the self-control I want. I still think I'd be better off doing something else but saw potential common ground every now and then. That those glimpses of progress in the discussion always returned to square one is why I didn't want to jump in in the first place, not because I wanted to deny any part of anyone's argument.

I'm just glad you had the extra time. The mutual admiration club thing had little to do with you, and I strongly suspect no one else was planning on offering up as unbiased of an opinion. Thanks again.

0 Replies

Ok, to be honest I've only read up to the part about me programming a script to test probability, and then I got excited, because it's simple. I haven't read past, but here is a simple php script:

Code:



<?

$causationBs = 0;

$nonCausationBs = 0;

for($i=0;$i<10000;$i++)

{

	$letter = rand(1,26);

	if($letter == 1)

		$causationBs++;	

	if($letter == 2)

		$nonCausationBs++;

}

echo "There were $causationBs Bs following an A, and $nonCausationBs Bs that were followed by another letter";

?>

Robert, you can put that up somewhere if you want. I ran it locally, and was finding that there were as many causationBs as there were nonCausationBs, which means that there is a MUCH greater chance that a B is following an A. (It follows A half the time, and the other 25 letters the rest of the time). I could break it down to show how often it followed each letter, but I really don't see a need.

1 Reply

@Robert Gentel,

Robert Gentel wrote:

I read that the last X times you copied it from Wikipedia. If it didn't work then (ascribe it to stubborness on my part if that makes it easier) I don't see how it would work now.

I was hoping someone might actually comment on it and point out where the logic fails.

0 Replies

@Nick Ashley,

You're making assumptions about the character stream other than the single rule that "A" always has a "B" after it. (i.e., only 26 characters, randomly distributed, etc.)

The rule "A" is always followed by "B" is followed in both of these character streams:

BBBBBBBBBBBBB

and

ABABABABABABAB

The rule itself, absent all other information does not provide you with information about what precedes any particular "B".

1 Reply

So then, one of the obvious assumed benefits of capital punishment, that it acts as an effective deterent, can safely be removed from the "yes" column and placed in the "don't know" column?

2 Replies

@Eorl,

I would say two things about that:

1. Bill's graph provides no new information for the debate about deterrence.
2. The preponderance of evidence indicates that there is no scientific support for the idea that the death penalty is an effective deterrent.

1 Reply

Killing people is the best solution.

Related Topics

Quick Links

My Account

able2know