Another shooting in the USA. Remind me about the reason for having guns again. - by SubJeff
LarryG on 9/9/2012 at 16:56
Quote Posted by zombe
No.
Lack of correlation means there was no correlation found in the specific dataset where correlation was tested - nothing more. Correlation does not really tell anything about causation (besides a dubious probability for refining future research - which can help but does not necessarily have to).
causation -> correlation, NOT the other way around.
// end of bickering about logic
Yes. Causation implies correlation. SO if there is no correlation there can be no causation. Assume you have causation without correlation and you arrive at an immediate contradiction. The very definition of causation is as a particular type of correlation. All relationships between things are correlations. They are somehow associated together. We may not know the specifics for the relationship, but we do know that they are related in some way. The particular relationship between things where the first is the cause of the 2nd is what we call a causative correlation, also known as a causation relationship.
Since you seem to know a little about logic, perhaps you recall what a (
http://en.wikipedia.org/wiki/Proof_by_contrapositive) contrapositive proof is? Remember that a statement and its contrapositive are logically equivalent. Apply that to your statement:
[INDENT]causation => correlation[/INDENT]
and you get
[INDENT]lack of correlation => lack of causation[/INDENT].
Those two statements are logically equivalent. And that's why showing lack of correlation is sufficient to show lack of causation. Correlation is necessary (but not sufficient) for causation. Without it, no matter what else you have, you cannot have causation. It's really that simple.
zombe on 9/9/2012 at 19:01
Quote Posted by LarryG
if there is no correlation there can be no causation.
If one would express it as "if it is impossible to find a correlation then there can be no causation." - i would agree (with reservations). However, proving a negative is, shall i say, rather difficult (outside mathematics) - making the sentence useless.
What i am getting at is that the data set might be influenced by other things (bad sample or other negative influences that nullify the correlation). Failing to find a correlation does not mean there is no causation - just that we have failed to find a correlation (with or without an underlying real causation).
Quote Posted by LarryG
causation => correlation
and you get
lack of correlation => lack of causation.
Agreed. However, "lack of correlation" (as a total negative) is impossible to establish (all you can say with certainty is that you have not found a correlation) - which is why i do not accept the second one.
Perhaps i did not express clearly enough what i protested against ...
Quote Posted by zombe
Lack of correlation means there was no correlation found in the specific dataset where correlation was tested - nothing more.
... but i do think you actually agree with what i was trying to say. No?
jay pettitt on 9/9/2012 at 22:35
Quote Posted by LarryG
Yes. Causation implies correlation. SO if there is no correlation there can be no causation. Assume you have causation without correlation and you arrive at an immediate contradiction. The very definition of causation is as a particular type of correlation. All relationships between things are correlations. They are somehow associated together. We may not know the specifics for the relationship, but we do know that they are related in some way. The particular relationship between things where the first is the cause of the 2nd is what we call a causative correlation, also known as a causation relationship.
Since you seem to know a little about logic, perhaps you recall what a (
http://en.wikipedia.org/wiki/Proof_by_contrapositive) contrapositive proof is? Remember that a statement and its contrapositive are logically equivalent. Apply that to your statement:
[INDENT]causation => correlation[/INDENT]
and you get
[INDENT]lack of correlation => lack of causation[/INDENT].
Those two statements are logically equivalent. And that's why showing lack of correlation is sufficient to show lack of causation. Correlation is necessary (but not sufficient) for causation. Without it, no matter what else you have, you cannot have causation. It's really that simple.
I think it's awesome that you went and got the numbers and actually had a look at them. Seriously awesome.
But...
Your dataset is not gun ownership vs gun killings. It's gun ownership vs gun killings along with literally everything else thrown into the pot for good measure: butterflies flapping their wings, socio-economic factors, survey biases, revolutions, war and peace. Literally war and peace.
If you were in a controlled experiment where you could tailor your tests and narrow the variables down to only the ones you want to investigate and you knew you were expecting a simple linear correlation - then you could perhaps be confident that a wholesale lack of correlation was telling you something interesting.
But that's not what is happening here. Here you've got dozens of variables interacting in complex, non-linear, not understood ways.
You need a different mindset when dealing with the woolly kinds of data you find in the wild - the sort you get in systems sciences like sociology and epidemiology. You do a test, note something interesting then test some more. When you've got lots results in the bag, you can maybe start to say that you're confident in your hypothesis - but it's the combined mass of the results from many tests looking at the thing from different angles that gives you the confidence, not a single result - not even if it's a very strong result.
For this stuff you need to start thinking in Bayesian terms.
LarryG on 9/9/2012 at 23:06
I don't know who you are arguing with.
Quote Posted by LarryG
SO: we can be confident that the UNODC data gathered through its annual crime survey does not support the ideas that more guns = more gun homicides or that less guns = more gun homicides or that more guns = less gun homicides or less guns = less gun homicides. Instead it supports the idea that pure gun availability does not drive gun homicides and that other factors should be looked at to determine reasonable candidates for causative factors in gun homicides.
The only conclusion that the data supports is that the number of gun homicides cannot be predicted based on the average number of guns per person
alone. That's all. We don't have anything else useful in this dataset to try and correlate gun homicides to, so we cannot make any other conclusions about the causes of gun homicides from it. But that single conclusion eliminates a huge number of simplistic arguments, such as were made earlier in this thread, showing them to be simply untrue. More guns does not make for less homicides on its own. Less guns does not make for less homicides on its own. If you want less homicides, you need to look elsewhere. That's what I'm trying to explain.
Stop focusing on gun ownership counts, and start looking at who is owning the guns, why they have them, what they plan to do with them, what kind of people they are, what their economic status is, etc. etc., etc.. And if you can find some common threads between who owns the guns and the numbers of gun homicides, you might be onto something. But please just quit with the simplistic arguments. The data does not support those conclusions.
CCCToad on 10/9/2012 at 04:10
Quote Posted by LarryG
Stop focusing on gun ownership counts, and start looking at who is owning the guns, why they have them, what they plan to do with them, what kind of people they are, what their economic status is, etc. etc., etc.. And if you can find some common threads between who owns the guns and the numbers of gun homicides, you might be onto something. But please just quit with the simplistic arguments. The data does not support those conclusions.
Careful here. Keep making sense instead of tossing around talking points and somebody might start accusing you of hating poor people.
Vasquez on 10/9/2012 at 04:31
CCCToad, one of the factors linked to violent crime in studies is economic inequality. That's a fact, not a biased opinion.
CCCToad on 10/9/2012 at 05:28
Isn't it bleeding obvious that my post was being sarcastic?
Vasquez on 10/9/2012 at 05:32
Sorry, my head is full of snot, maybe I'd better stay away from these threads until I feel better ;)
jay pettitt on 10/9/2012 at 08:03
Quote Posted by LarryG
I don't know who you are arguing with.
You. I'm pretty sure I left clues to that effect in the post.
Quote:
The only conclusion that the data supports is that the number of gun homicides cannot be predicted based on the average number of guns per person
alone.
Actually it doesn't. You've way to much confidence that you can apply stats from global data to other scales. The conclusion you're drawing is far stronger and further reaching than is supported by your analysis. At very best were talking shades of grey here.
Quote:
That's all. We don't have anything else useful in this dataset to try and correlate gun homicides to, so we cannot make any other conclusions about the causes of gun homicides from it.
Poppycock. There's likely a tonne of useful information in that data set. Chimpy's reanalysis that grouped countries with similar socio-economic characteristics was a genuinely interesting progression. And the criticisms you made of it were utterly unfounded.
Quote:
But that single conclusion eliminates a huge number of simplistic arguments
We could have a discussion along those lines if your analysis produced 'that single conclusion', but it didn't.
Quote:
such as were made earlier in this thread, showing them to be simply untrue
Mostly your analysis doesn't say much of anything about arguments in this thread. It doesn't support them, it might even tinge them with a slight shade of grey (though actually I doubt it) but it isn't enough on its own to exclude any of the arguments that have been made. Nobody has argued that global homicides should track global gun ownership.
Quote:
More guns does not make for less homicides on its own. Less guns does not make for less homicides on its own. If you want less homicides, you need to look elsewhere. That's what I'm trying to explain.
I probably agree with you, but the data/analysis we've got so far doesn't support that. Other analysis/data might, but we've not looked at it in this thread.
Quote:
The data does not support those conclusions.
We can't know that until we've looked.
Quote:
Stop focusing on gun ownership counts, and start looking at who is owning the guns, why they have them, what they plan to do with them, what kind of people they are, what their economic status is, etc. etc., etc.. And if you can find some common threads between who owns the guns and the numbers of gun homicides, you might be onto something. But please just quit with the simplistic arguments.
Simplistic arguments are not the problem right now. The problem is simplistic analysis. Simple logic of the sort you're trying to apply to the problem won't work here - there are too many unknowns. When you've got a problem like this you need to think about confidence and probability like a Bayesian.
Besides, it's all moot.
Look, somebody asserted that gun control arguments were bunk because the opposite should be true; he believed that more guns should lead to less violence. You got some global data and challenged him, but he was almost certainly talking on a regional and state level. The best you can really say about your analysis was that it was inconclusive - it didn't really say anything about local/state scales. A simple look at global data certainly didn't support the idea, but should you expect it to?
Regardless, the burden of evidence is still with Mr GunControlsArePoop.
SubJeff on 10/9/2012 at 09:17
And herein jay illustrates why I'm out.
:thumb:
