“Correlation does not prove causality.” You have heard the one-liner uttered by clueless social skeptics probably one thousand times or more. But real science rarely if ever starts with ‘proof.’ More often than not, neither does a process of science end in proof. Correlation was never crafted as an analytical means to proof. However this one-liner statement is most often employed as a means of implying proof of an antithetical idea. To refuse to conduct the scientific research behind such fingerprint signal conditions, especially when involving a risk exposure linkage, can demonstrate just plain ole malicious ignorance. It is not even stupid.
When a social skeptic makes the statement “Correlation does not prove causality,” they are making a correct statement. It is much akin to pointing out that a pretty girl smiling at you does not mean she wants to spend the week in Paris with you. It is a truism, most often employed to squelch an idea which is threatening to the statement maker. As if the statement maker were the boyfriend of the girl who smiled at you. Of course a person smiling at you does not mean they want to spend a week in Paris with you. Of course correlation does not prove causality. Nearly every single person bearing any semblance of rational mind understands this. But what the one who has uttered this statement does not grasp, while feeling all smart and skeptickey in its mention, is that they have in essence revealed a key insight into their own lack of scientific literacy. Specifically, when a person makes this statement, three particular forms of error most often arise. In particular, they do not comprehend, across an entire life of employing such a statement, that
1. Proof Gaming/Non Rectum Agitur Fallacy: Correlation is used as one element in a petition for ‘plurality’ and research inside the scientific method, and is NOT tantamount to a claim to proof by anyone – contrary to the false version of method foisted by scientific pretenders.
To attempt to shoot down an observation, by citing that it by itself does not rise tantamount to proof, is a form of Proof Gaming. It is a trick of trying to force the possible last step of the scientific method, and through strawman fallacy regarding a disliked observer, pretend that it is the first step in the scientific method. It is a logical fallacy, and a method of pseudoscience. Science establishes plurality first, seeks to develop a testable hypothesis, and then hopes, …only hopes, to get close to proof at a later time.
Your citing examples of correlation which fail the Risk Exposure Test, does not mean that my contention is proved weak.
… and yes, science does use correlation comparatives in order to establish plurality of argument, and consilience which can lead to consensus (in absence of abject proof). The correlation-causality statement, while mathematically true, is philosophically and scientifically illiterate.¹²³
2. Ignoratio Elenchi Fallacy (ingens vanitatum): What is being strawman framed as simply a claim to ‘correlation’ by scientific pretenders, is often a whole consilience (or fingerprint) of mutually reinforcing statistical inference well beyond the defined context of simple correlation.
Often when data shows a correlation, it also demonstrates other factors which may be elicited to demonstrate a relationship between two previously unrelated contributing variables or data measures. There are a number of other factors which science employs through the disciplines of modeling theory, probability and statistics which can be drawn from a data relationship. In addition these inferences can be used to mutually support one another, and exponentially increase the confidence of contentions around the data set in question.²³
3. Methodical Cynicism: Correlation is used as a tool to examine an allowance for and magnitude of variable dependency. In many cases where a fingerprint signal is being examined, the dependency risk has ALREADY BEEN ESTABLISHED or is ALLOWED-FOR by diligent reductive science. To step in the way of method and game protocols and persuasion in order to block study, is malevolent pseudoscience.
If the two variables pass the risk-exposure test, then we are already past correlation and into measuring that level of dependency, not evaluating its existence. If scientific studies have already shown that a chemical has impacts on the human or animal kidney/livers/pancreas, to call an examination of maladies relating to those organs as they relate to trends in use of that chemical a ‘correlation’ is an indication of scientific illiteracy on the part of the accuser. Once a risk relationship is established, as in the case of colon disorders as a risk of glyphosate intake, accusations of ‘correlation does not prove causality’ constitute a non-sequitur Wittgenstein Error inside the scientific method. Plurality has been established and a solid case for research has been laid down. To block such research is obdurate scientific fraud.²³
4. Correlation does not prove causality… however, even weaker in strength of inference is an implicit refutation by claim of coincidence.
Most often, when one poses the ‘correlation does not prove causality’ apothegm, they are attempting to enforce an implicit counter-claim to coincidence in the observed data. While this is the null, it is also most often not an actual hypothesis – nor can such a claim be made without evidence. Most often the evidence in support of a correlation being merely coincidence, is in fact weaker than the evidence in support of causality. A position of epoche is warranted – not denial, in such circumstances.
Calling or downgrading the sum total of these inferences through the equivocal use of the term ‘correlation,’ not only is demonstrative of one’s mathematical and scientific illiteracy, but also demonstrates a penchant for the squelching of data through definition in a fraudulent manner. It is an effort on the part of a dishonest agent to prevent the plurality step of the scientific method.
None of this has anything whatsoever to do with ‘proof.’
A Fingerprint Signal is Not a ‘Correlation’
An example of this type of scientific illiteracy can be found here (Note: a former article entitled Correlation Is Not Causation in Earth’s Dipole Contribution to Climate by Steven Novella, which was dropped by Discover Magazine). There is a well established covariance, coincidence, periodicity and tail sympathy; a long tight history of dynamic with respect to how climate relates to the strength of Earth’s magnetic dipole moment. This is a fingerprint signal. Steven Novella incorrectly calls this ‘correlation.’ A whole host of Earth’s climate phenomena move in concert with the strength of our magnetic field. This does not disprove anthropogenic contribution to current global warming. But to whip out a one liner and shoot at a well established facet of geoscience, all so as to protect standing ideas from facing the peer review of further research is not skepticism, it is pseudoscience. The matter merits investigation. This hyperepistemology one-liner does not even rise to the level of being stupid.
Measuring of An Established Risk Relationship is Not a ‘Correlation’
An example of this type of scientific illiteracy can be found inside pharmaceutical company pitches about how the increase in opioid addiction and abuse was not connected with their promotional and lobbying efforts. Correlation did not prove causality. Much of today’s opiate epidemic stems from two decades of promotional activity undertaken by pharmaceutical companies. According to New Yorker Magazine, companies such as Endo Pharmaceuticals, Purdue Pharma and Johnson & Johnson centered their marketing campaigns on opioids as general use pain treatment medications. Highly regarded medical journals featured promotions directed towards physicians involved in pain management. Educational courses on the benefits of opioid-based treatments were offered. Pharmaceutical companies made widespread use of lobbyist groups in their efforts to disassociate opiate industry practices from recent alarming statistics (sound familiar? See an example where Scientific American is used for such propaganda here). One such group received $2.5 million from pharmaceutical companies to promote opioid justification and discourage legislators from passing regulations against unconstrained opioid employment in medical practices. (See New Yorker Magazine: Who is Responsible for the Pain Pill Epidemic?) The key here is, that once a risk relationship is established, such as between glyphosate and cancer, one cannot make the claim that correlation does not prove causality in the face of two validated sympathetic risk-dependency signals. It is too late, plurality has been established and the science needs to be done. To block such science is criminal fraud.
Perhaps We Need a New Name Besides Correlation for Such Robust Data Fit
Granger Causality – a refutation of a post hoc ergo propter hoc claim in that a variable X that evolves over time Granger-causes another evolving variable Y if predictions of the value of Y based on its own past values and on the past values of X are better than predictions of Y based only on Y’s own past values (or especially another assumed causality variable X2). Granger causality may not indicate direct causation; however suggests a common mechanism of some type via a fingerprint signal means which is much stronger than mere correlation.
Both of these examples above elicit instances where fake skeptic scientific illiteracy served to mis-inform, mis-lead or cause harm to the American Public. Correlation, in contrast, is simply a measure of the ‘fit’ of a linear trend inside the relationship between a two factor data set. It asks two questions (the third is simply a mathematical variation of the second):
- Can a linear inference be derived from cross indexing both data sets?, and
- How ‘close to linearity’ do these cross references of data come?
- How ‘close to curvinlinearity’ do these cross references of data come?
The answer to question number 2 is called an r-factor or correlation coefficient. Commonly, question number 3 is answered by means of a coefficient of determination and is expressed as an r² factor (r squared).³ Both are a measure of a paired-data set fit to linearity. That is all. In many instances pundits will use correlation to exhibit a preestablished relationship, such as the well known relationship between hours spent studying and academic grades. They are not establishing proof with a graph, rather simply showing a relationship which has already been well documented through several other previous means. However, in no way shape or form does that mean that persons who apply correlation as a basis of a theoretical construct are therefore then contending a case for proof. This is a relational form of the post hoc ergo propter hoc fallacy. This is a logical flaw, served up by the dilettante mind which confuses the former case, an exhibit, and conflates it with the later use, the instance of a petition for research.
Correlation Dismissal Error (Fingerprint Ignorance)
/philosophy : logic : evidence : fallacy/ : when employing the ‘correlation does not prove causality’ quip to terminally dismiss an observed correlation, when the observation is being used to underpin a construct or argument possessing consilience, is seeking plurality, constitutes direct fingerprint evidence and/or is not being touted as final conclusive proof in and of itself.
THIS is Correlation (Pearson’s PPMCC) It does not prove causality (duh…)¹²
This is a Fingerprint Signal and is Not Simply a Correlation³∋
There are a number of other methods of determining the potential relationship between two sets of data, many of which appear to the trained eye in the above graph. Each of the below relational features individually, and increasingly as they confirm one another, establish a case for plurality of explanation. The above graph is not “proving” that glyphosate aggravates diabetes rates. However, when this graph is taken against the exact same shape and relationship graphs for multiple myloma, non-Hodgkin’s Lymphoma, bladder cancer, thyroid disease, pancreatic cancer, irritable bowel syndrome, inflammatory bowel syndrome, lupus, fibromyalgia, renal function diminishment, Alzheimer’s, Crohn’s Disease, wheat/corn/canola/soy sensitivity, SIBO, dysbyosis, esophageal cancer, stomach cancer, rosacea, gall bladder cancer, ulcerative colitis, rheumatoid arthritis, liver impairment and stress/fatty liver disease, … and for the first time in our history a RISE in the death rates of of middle aged Americans…
… and the fact that in the last 20 years our top ten disease prescription bases have changed 100%… ALL relating to the above conditions and ALL auto-immune and gut microbiome in origin. All this despite a decline in lethargy, smoking and alcohol consumption on average. All of this in populations younger than an aging trend can account for.
Then plurality has been argued. Fingerprint signal data has been well established. This is an example of consilience inside an established risk exposure relationship. To argue against plurality through the clueless statement “Correlation does not prove causality” is borderline criminal. It is scientifically illiterate, a shallow pretense which is substantiated by false rationality (social conformance) and a key shortfall in real intelligence.
Contextual Wittgenstein Error Example – Incorrect Rhetoric Depiction of Correlation
The cartoon to the left is a hypoepistemology which misses the entire substance of what constitutes fingerprint correlation. A fingerprint signal is derived when the bullet-pointed conditions exist – None of which exist in the cartoon invalid comparison to the left – this is a tampering with definition, enacted by a person who has no idea what correlation in this context, even means. A Wittgenstein Error. In other words: scientifically illiterate propaganda. Conditions which exist in a proper correlation, or more, condition:
- A constrained pre-domain and relevant range which differ in stark significance
- An ability to fit both data sets to curvinlinear or linear fit, with projection through golden section, regression or a series of other models
- A preexisting contributor risk exposure between one set of unconstrained variables and a dependent variable
- A consistent time displacement between independent and dependent variables
- A covariance in the dynamic nature of data set fluctuations
- A coincident period of commencement and timeframe of covariance
- A jointly shared arrival distribution profile
- Sympathetic long term convex or concave trends
- A risk exposure (see below) – the cartoon to the left fails the risk exposure test.
Rhetoric: An answer, looking for a question, targeting a victim
Fingerprint Elements: When One or More of These Risk Factor Conditions is Observed, A Compelling Case Should be Researched¹²³
Corresponding Data – not only can one series be fitted with a high linear coefficient, another independent series can also be fitted with a similar and higher coefficient which increases in coherence throughout a time series both before and during its domain of measure, and bears similar slope, period and magnitude. In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a strong case for plurality especially if a question of risk is raised. To ignore this condition, is a circumstance where ignorance ranges into fraud.
Covariant Data – not only can one series be fitted with a high coefficient, another independent series can also be observed with a similar fit which increases in coherence as a time series both before and during its domain of measure, and bears similar period and magnitude. Adding additional confidence to this measure is the dx/dy covariance, Browning Covariance, or distance covariance, etc. measure which can be established between the two data series; that is, the change in x(1)…x(n) versus y(1)…y(n). In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a very strong case for plurality especially if a question of risk is raised. To ignore this condition, is a circumstance where socially pushed skepticism ranges into fraud.
Co-incidence Data – two discrete measures coincide as a time series both before and during its domain of measure, and bear similar period and magnitude. Adding additional confidence to this measure magnitude consistency which can be established between the two data series; that is, the discrete change in x(1)…x(n) versus y(1)…y(n). In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a moderately strong case for plurality especially if a question of risk is raised. To ignore this condition, is a circumstance where arrogant skepticism ranges into fraud.
Jointly Distributed Data – two independent data sets exhibit the same or common arrival distribution functions. Adding additional confidence to this measure magnitude consistency which can be established between the two data series; that is, the discrete change in x(1)…x(n) versus y(1)…y(n). In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a moderately strong case for plurality especially if a question of risk is raised. To ignore this condition, is a circumstance where arrogant skepticism ranges into fraud.
Probability Function Match – two independent data sets exhibit a resulting probability density function of similar name/type/shape. Adding additional confidence to this measure magnitude consistency which can be established between the two data series; that is, the discrete change in x(1)…x(n) versus y(1)…y(n). In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a moderately strong case for plurality especially if a question of risk is raised. To ignore this condition is not wise.
Marginal or Tail Condition Match – the tail or extreme regions of the data exhibit coincidence and covariance. Adding additional confidence to this measure magnitude consistency which can be established between the two data series when applied in the extreme or outlier condition; that is, the discrete change of these remote data in x(1)…x(n) versus y(1)…y(n). In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a moderately strong case for plurality especially if a question of risk is raised. To ignore this condition, is a circumstance where even moderate skepticism ranges into fraud activity.
Sympathetic Long Term Shared Concave or Convex – long term trends match each other, but more importantly each is a departure from the previous history and occurred simultaneously, offset by a time displacement, are both convex or concave and co-vary across the risk period. Adding additional confidence to this measure magnitude consistency which can be established between the two data series; that is, the discrete change in x(1)…x(n) versus y(1)…y(n). In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a compellingly strong case for plurality especially if a question of risk is raised. To ignore this condition, is a circumstance where even moderate skepticism ranges into fraud activity.
Discrete Measures Covariance – the mode, median or mean of discrete measures is shared in common and/or in coincidence, and also vary sympathetically over time. Adding additional confidence to this measure magnitude consistency which can be established between the two data series; that is, the discrete change in mode and mean over time. In this instance as well, a preexisting risk exposure has been established. This does not prove causality, however is a moderate case for plurality especially if a question of risk is raised. To ignore this condition is not wise.
Risk Exposure Chain/Test – two variables, if technical case were established that one indeed influenced the other, would indeed be able to influence one another. (In other words, if your kid WAS eating rat poison every Tuesday, he WOULD be sick on every Wednesday – but your kid eating rat poison would not make the city mayor sick on Wednesday). If this condition exists, along with one or more of the above conditions, a case for plurality has been achieved. To ignore this condition, is a circumstance where even moderate skepticism ranges into fraud activity.
These elements, when taken in concert by honest researchers, are called fingerprint data. When fake skeptics see an accelerating curve which matches another accelerating curve – completely (and purposely) missing the circumstance wherein any or ALL of these factors are more likely in play – to say “correlation” is what is being seen, demonstrates their scientific illiteracy. It is up to the ethical skeptic to raise their hand and say “Hold on, I am not ready to dismiss that data relationship so easily. Perhaps we should conduct studies which investigate this risk linkage and its surrounding statistics.”
To refuse to conduct the scientific research behind such conditions, especially if it involves something we are exposed to three times a day for life, constitutes just plain active ignorance and maliciousness. It is not even stupid.
epoché vanguards gnosis
¹ Madsen, Richard W., ” Statistical Concepts with Applications to Business and Economics,” Prentice-Hall, 1980; pp 604 – 610.
² Gorini, Catherine A., “Master Math Probability,” Course Technology, 2012; pp. 175-196, 252-274.
³ Levine, David M.; Stephan, David F., “Statistics and Analytics,” Pearson Education, 2015; pp. 137-275.
∋ Graphic employed for example purposes only. Courtesy of work of Dr. Stephanie Seneff, sulfates, glyphosates and gmo food; MIT, september 19, 2013.
What is your opinion of the “extraordinary claims requires extraordinary evidence”?
Yes, excellent question. That Sagan quip is THE single most abused one-liner in the entire arsenal of fake skepticism. By promoting an observation to status as a ‘claim’ – one can then apply the fake ‘Occam’s Razor’ – declare the claim to not be simple, and dismiss it as if it were a THEORY. Poof – follow that by the ‘science has found no data for (praedicate evidentia)’ and you are a scientist-skeptic master black belt of denial. This is exactly how it works.
A magic trick. A deception.
I blogged on that here:
http://theethicalskeptic.com/2012/03/18/deskeption-observation-vs-claim-blurring/
and here:
http://theethicalskeptic.com/2015/05/17/why-sagan-is-wrong/
TES :)