From the “correlation is not causation” department.
Guest Posting by Ira Glickstein
Since the Death Penalty was restored in the US in 1975, the Number of Executions per Year correlates with UAH Global Temperature Anomaly better than CO2 levels! So, if we want to reduce warming, cut the rate of capital crimes! (See graphic below.)
In the above graphic, the black line indicates US Executions per Year from 1975 to 2013 (Source: http://www.statisticbrain.com/death-penalty-statistics/ NAACP LDF “Death Row, U.S.A., Gallup Poll, Bureau of Justice Statistics”), the blue jagged line indicates monthly UAH Satellite-Based Temperature of the Global Lower Atmosphere from 1979 through 2013 and the red line shows the running centered 13-month average (Source: http://www.drroyspencer.com/2014/02/uah-global-temperature-update-for-january-2014-0-29-deg-c/).
In an earlier WUWT posting I showed that Total US Debt (public and private) as a percentage of US Gross Domestic Product (GDP) correlates with NASA GISS US Annual Mean Temperature Anomaly better than CO2 levels! So, if we want to reduce warming, cut the debt! (See http://wattsupwiththat.com/2011/02/16/forget-co2-us-debt-causes-warming/)
Therefore, based on the (faulty) idea that Correlation Implies Causation, we can solve Global Warming by reducing US debt and US capital crimes :^)
And the Good Lord looketh down upon the world, and He decreeath that whensoever mankind takes lives, even of killers, that the world will become hotter, as in hell. And when the rate of such executions shall diminish, so shall the rate of such warming. This can easily be correlated because there is not one report of any executions taking place during the last ice age. Keep the Friday Funnies coming!!
kadaka (KD Knoebel) says:
May 10, 2014 at 1:13 am
============
Thanks for the reply (and the others).
Good to have you back.
Causation, maybe not. A link? What’s so absurd about that.
In general, I think politicians who tend to favor capital punishment also favor lax climate regulation and debt spending, when you look at their actual records and not rhetoric.
Whereas despite the odd boondoggle, politicians who oppose capital punishment also tend to favor a functional EPA and sound spending practices.
Naturally, in countries where all parties agree to something like ending capital punishment, the correlation stops there.
But this doesn’t strike me as a funny fluke but rather a result of comorbid management practices.
Best WUWT thread ever!
Thank you all ladies and gentlemen for you contributions. I’m too inebriated to participate, but fortunately not too inebriated to roll around on the floor giggling hysterically. I can think of no other place that offers such incredibly dry wit and repartee when the subject matter is so uninspiring. I can only summise that the causation of such comments has a direct correlation to the intellect of the commenters. I’ll thank you once again before my stupor takes full control.
Ira Glickstein wrote:
[AGF: I have two digital clocks, A and B, and they are correlated 100% to an infinitesimal fraction of a second. Does that mean that A CAUSES B (or B CAUSES A)? Nope! Both are tuned to the National Bureau of Standards time signal, C. So C CAUSES A, and C CAUSES B. But A and B have no causal relationship with each other. I could destroy A and it would have no effect on B, and vice-versa. Ira]
Ira, how spooky, I’d just logged back in with the intention of writing a final attempt at showing how 100% correlation can never be assumed to infer causation with an example of two digital clock readouts in a sealed black box that happen to be 5 seconds apart; so B appears to lag A and is perfectly correlated with it but on internal inspection the clocks are entirely independent; but you’d beaten me too it…
Cheers and thanks for the feedback.
Mark
Actually that ‘infer’ should have been an ‘imply’, doh!
Ta
Mark
A lagging correlation between executions and temperature might exist [ similar, for different reasons, to the relationship between temperature and CO2 in ice core data].
Let us assume [and this may not be true] that there is more crime when it is warmer.
More people out on the streets, more alcoholic drinks consumed.
And then if it is very warm and the air conditioner is broken tempers fray etc.
So possibly more crimes that are deemed capital in various jurisdictions.
Assume also that most of the miscreants are arrested, tried and found guilty.
This process takes time.
So there should be a lag between high temperature and executions by about the average period between committing a capital offence and execution [ the average period on death row].
In China this time would be very short.
In the USA somewhat longer.
[GregK: Thanks, great point. In the USA Executions generally lag the capital crime by a decade or more due to the time it takes to catch the murderer, assemble the evidence, conduct the trial, and deal with the interminable appeals and judicial reviews. But, my graphic shows that Executions increase a bit BEFORE Global Warming, and continue high a bit AFTER. Considering the decade or more delay between the capital crime and the Execution, it is clear that (if there is a causal relationship) it is capital crimes that “cause” Global Warming. Ira]
Lewis P Buckingham says:
May 10, 2014 at 12:27 am
“Why stop there, since everyone breathes before they die, breathing must be a cause of death.”
==================================
Nope. It is the cessation of breathing that causes death. There’s 100% correlation there. Everyone who stops breathing, dies, so keep breathing as long as you can ;o)
Ira G. inserts at unknown time, listed at:
May 9, 2014 at 10:31 pm
[AGF: I have two digital clocks, A and B, and they are correlated 100% to an infinitesimal fraction of a second. Does that mean that A CAUSES B (or B CAUSES A)? Nope! Both are tuned to the National Bureau of Standards time signal, C. So C CAUSES A, and C CAUSES B. But A and B have no causal relationship with each other. I could destroy A and it would have no effect on B, and vice-versa. Ira]
===================================================================
Correct of course, but it remains absurd to claim no causal relationship involved with the synchronization of your two clocks. Like claiming no causation between train cars, one in front of the engine and one behind. A good climate example is the excellent Pleistocene T/CO2 correlation: they are parallel functions of ice sheet extension. Your objection is trivial. –AGF
[AGF: Thanks for your reply. OK, say scientists find a planet orbiting some far-away star and that planet happens to rotate on its axis once every 24 Earth-hours, in exact synchronization and 100% correlation with our Earth. The Earth or that planet could be destroyed with no detectable effect on the other. So, where is the causal connection? Caution; if you say the causal connection is that both Earth and the synchronized planet share a common origin of the Universe in the Big Bang, then everything in the Universe is causally connected to everything else, which is true, but which makes your claim that correlation implies some sort of causal connection absolutely trivial. Ira]
Climate Change to blame for Nigerian Girls Kidnapping: http://www.breitbart.com/Breitbart-London/2014/05/11/UK-Guardian-climate-change-to-blame-for-Nigerian-girls-kidnapping
Don’t read the Guardian, it’s the newspaper of idiots.
Re. Ira G. at May 11, 2014 at 9:04 am
There are no doubt thousands of planets in the universe which rotate with an average of within a few ms of 23 hours 56 minutes and 4.0916 seconds. A few of them might even share our mean solar day within limits. But within this huge universe I suspect there are no planets that could duplicate the earth’s rotational history to the nearest millisecond over the last 50 years or 50my. That would require cause and effect or improbable coincidence.
Regards, –AGF
====================================================================
…the animals that don’t eat die so the cause isn’t the eating, it’s the animals!
Gunga Din says:
“…the animals that don’t eat die so the cause isn’t the eating, it’s the animals!”.
Indeed it is the animals not the eating 🙂
The point being made was that it’s trivial to find examples of variables A and B that when, say, fed into a stats package (which has _no_ a priori knowledge of what A and B actually are) results in a correlation coefficient of 1; that is they are 100% correlated even if there is no link between them.
Too many times in research (in pretty much any field) a high correlation (regardless of value) has been taken to mean causation with the result that the received wisdom is then ‘if you change A then you can effect B).
Cheers
Mark
ajfosterjr says:
“Correct of course, but it remains absurd to claim no causal relationship involved with the synchronization of your two clocks. Like claiming no causation between train cars, one in front of the engine and one behind. ”
Once again I’m not sure if this is not down to a difference in definition. In the Ira clocks example the fact remains that A and B have no causal link (changing A does not effect B). They _do_ share a common underlying cause C, the time sync signal but the cause relationship is C A and B A.
If the definition by which you’re approaching this discussion means that a casual relationship can include both direct and indirect links then I’d say that for a system whereby you have A and B perfectly correlated then that would very, very strongly indicate some form of causal link, either direct or common third-party. Nevertheless there is always the possibility, regardless of how small, that the correlation may be purely coincidental (and down to chance, choice of sample size, measurement resolution, choice/knowledge of variables and so forth).
I’d also say that the conventional definition of a causal relationship (in the contexts being discussed on this thread) is a _direct_ one. However, that’s just my opinion :-).
Also note that from my perspective we’ve been discussing correlation when measuring two variables that we have no control over (ie we just sample them as they change / occur naturally). If however you take your measurements by specifically controlling one variable and measuring the resultant change on the other (eg changing voltage and measuring current) then that’s a different story all together…
Cheers
Mark
That should have read “but the cause relationship is C A and C B” obviously!
H.R.:
At May 11, 2014 at 6:01 am you say
In keeping with the so-called ‘logic’ which thinks correlation indicates causation, I write to expand on your observation.
If when you awake in the morning you discover you have stopped breathing then you have a problem.
So,
if you awake in the morning to discover that you have not stopped breathing then you do not have a problem.
And
when you do not have a problem then you have no reason to worry.
Be happy.
Richard
Mark Fawcett:
You spoil your good post at May 12, 2014 at 12:43 am by concluding
Sorry, but NO!
We have been discussing whether correlation indicates causation: it does not.
When an alteration to one variable is observed to repeatably induce a change in another variable then that is evidence of a causal relationship between the two variables.
In such a situation a correlation (which may or may not be linear) may be observed between the variables, but the correlation is not evidence of the causal relationship: the observation of the repeatedly induced change is evidence of the causal relationship.
The reason that no significant correlation may exist between the two variables is because there may be a confounding effect which reduces the confidence of the correlation to an insignificant level. In this case, either or both of the variables may be varied by the confounding effect while the effect of the causal relationship between the two variables would be much smaller than the confounding effect. In such a case the causal relationship between the two variables can be demonstrated by experiment but can usually be ignored (so is assumed to not exist) because it is overwhelmed by the confounding effect.
Richard
Richardscourtney says:
“Sorry, but NO!
We have been discussing whether correlation indicates causation: it does not.
When an alteration to one variable is observed to repeatably induce a change in another variable then that is evidence of a causal relationship between the two variables.
In such a situation a correlation (which may or may not be linear) may be observed between the variables, but the correlation is not evidence of the causal relationship: the observation of the repeatedly induced change is evidence of the causal relationship.”
Richard, couldn’t agree more with your statements which put into clear terms what I’d meant by ‘that’s a different story’ – I’d meant to imply that it was a totally different thing to correlation / causation not that it meant correlation means causation in that case. Apologies for the misleading turn of phrase 🙂
Cheers
Mark
richardscourtney says:
May 12, 2014 at 2:30 am
“In such a situation a correlation (which may or may not be linear) may be observed between the variables, but the correlation is not evidence of the causal relationship: the observation of the repeatedly induced change is evidence of the causal relationship.”
===================================================================
RC, let me repeat:
agfosterjr says:
May 9, 2014 at 9:41 pm
Tom Trevor says:
May 9, 2014 at 9:22 pm
“Even the type of correlation you are talking about doesn’t suggest causation.”
And:: “Even 100% correlation does not tell us what event is the cause and what event is the effect.”
=================================================================
What you mean to say with these contradictory statements is that while correlation doesn’t distinguish between cause and effect it certainly does indicate cause and effect. –AGF
&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&&
From the start I have claimed, and still claim, that correlation indicates causation. I didn’t say that correlation spells out any particular cause and effect. Or that it precludes a common cause for tandem effects. You write, “In such a situation a correlation (which may or may not be linear) may be observed between the variables, but the correlation is not evidence of the causal relationship: the observation of the repeatedly induced change is evidence of the causal relationship.”
But this has no bearing on the argument. If anything, it backs up my point: one determines cause and effect through correlation, when one thinks one knows the cause. Baby flips light switch; room lights up. Baby concludes cause and effect through correlation.
So to prevent further needless confusion, let me restate more specifically:
Correlation indicates causation (by a consensus of definition).
It does not distinguish between cause and effect.
It does not preclude a cause other than what is correlated.
If I back off from the claim of the absolute nature of correlation more problems are created than solved. E.g., what is the “correlation”? (say 95%) 95% of what? 100% correlation. Correlation unqualified is absolute. So the correct question would be, what is the measure of correlation?
If this claimed consensus of definition is not accepted, then I must revise: perfect correlation always indicates causation.
Does that help? –AGF
agfosterjr says:
May 12, 2014 at 12:51 pm
For sake of argument, let me agree that PERFECT correlation of A and B implies CAUSATION of the form A CAUSES B, or B CAUSES A, or some other common thing C CAUSES A and C CAUSES B.
OK, what if A and B are NOT PERFECTLY correlated? Say 95% or 90% ? say way less than 90% ?
In my graphic examples of Number of US Executions vs UAH satellite global temperature anomaly (1979 thru 2013), and my earlier example of US Public and Private Debt as a percentage of GDP vs NASA GISS temperature anomaly (1880’s to 2011) the correlation is visually appealing, but not quantified. My purpose was to show that Executions and Debt were BETTER CORRELATED with temperature anomaly than Atmospheric CO2 levels which NASA GISS, the IPCC, and the whole Climate “Hockey Team” claim is the MAJOR CAUSE of Global Warming.
We know, from the basic physics of the Atmospheric “Greenhouse” Effect, that Atmospheric CO2 IS A REAL CAUSE OF WARMING, but we (Skeptics) doubt the strength of that CAUSATION. My point is that the strength of CAUSATION (I.e, climate sensitivity to CO2) has been overestimated by the Climate Team by a factor of two or three or more.
I did not think that anyone would believe that Executions or Debt had a STRONG CAUSAL connection to warming. Their CAUSAL connection, if it exists at all, is quite WEAK. Therefore, if Executions and Debt have a STRONGER CAUSAL connection than CO2, which is what my graphic demonstrates, then the CAUSAL connection between human-caused CO2 must be very weak, indeed. QED
Ira
Hmmm…I wonder what a graph charting Mann’s lawsuits would show?
The hotter it gets under his collar, the richer his lawyers are?
Ira Glickstein, PhD says:
May 12, 2014 at 5:08 pm
You were probably not the author of the subheading: “From the “correlation is not causation” department,” but from there I took my cue: this is a common fallacy. Similar to such dicta as “nothing is absolute,” or the modified, “the only absolute is that nothing is absolute.” Such Hollywood philosophy as Wikipedia’s entry http://en.wikipedia.org/wiki/Correlation_does_not_imply_causation
is inexcusable coming from a science whose raison d’etre is the fact that correlation does imply causation. In fact typical definitions in effect equate correlation with causation. E.g., the first one Google brings up (without hitting ENTER):
cor·re·la·tion
ˌkôrəˈlāSHən/Submit
noun
a mutual relationship or connection between two or more things.
“research showed a clear correlation between recession and levels of property crime”
synonyms: connection, association, link, tie-in, tie-up, relation, relationship, interrelationship, interdependence, interaction, interconnection; More
STATISTICS
interdependence of variable quantities.
STATISTICS
a quantity measuring the extent of interdependence of variable quantities.
===========================================================
“interdependence of variable quantities”? = correlated values, functions of each other, effects of causes.
“a quantity measuring the extent of interdependence of variable quantities”? OK, definition #2, maybe, but this is better called “correlation coefficient.” Because as an abstraction, “correlation” is better taken as an absolute.
I hardly take issue with your method; I wasn’t responding to your post but to the subtitle noted above, and the fallacy it implies. I especially liked the statistical study of malaria posted last November http://wattsupwiththat.com/2013/11/07/new-study-shows-malaria-has-little-to-do-with-climate-but-more-with-household-size/
where by checking various correlation rates the authors apparently narrowed the cause to the number of family members sleeping in the same room. This is an extension of what the birds do when learning to distinguish between edible and inedible butterflies. We practice statistics because statistics have relevance in the real world. They relevance only because correlation does imply causation.
Thanks for your post and comments. –AGF
richardscourtney says:
May 12, 2014 at 2:09 am
http://wattsupwiththat.com/2014/05/09/friday-funny-forget-co2-us-executions-cause-global-warming/#comment-1634070
That expansion was all a good chuckle. Thank you. sir!
P.S. I got that one from my father. His wry observation was that coroners went to a lot of unnecessary fuss when the actual cause of death obviously was that the deceased stopped breathing. Sadly, he stopped breathing in 2008, but I still use his line when I can… always brings a smile to my face.
H.R.
richardscourtney says: May 12, 2014 at 2:09 am
H.R.:
At May 11, 2014 at 6:01 am you say
“It is the cessation of breathing that causes death. There’s 100% correlation there. Everyone who stops breathing, dies, so keep breathing as long as you can ;o)”
But if there is 100% correlation, and it is not evidently coincidence, then can you properly say whether A causes B or B causes A?
In H.R.’s example, would it be just as true to say that death causes cessation of breathing?
Consider – “death” can be considered to be total cessation of brain function. Following this, all inputs to the chest muscles cease, and breathing stops. So death causes cessation of breathing.
Further, consider a patient in an Iron Lung. Breathing has stopped – the Iron Lung does the necessary pumping of air in and out of the lungs, so the patinet survives (it is hoped). So here cessation of breathing has not caused death! QED,