Thursday, February 26, 2015

Causation IS Correlation!

Let’s get to the core of the Swanson paper and start talking about correlations.

“Correlation is not causation”.  How often have we seen this mantra on pro-GMO websites!  It’s as if they only have to say the magic words and they win the argument.  The picture is reinforced by graphs like this:

You may recognize this meme, it’s been floating around on pro-GMO websites for quite a while.  It’s kind of fun because it’s so ridiculous—calculated to really piss off organic food fans.  And the intended take-home message is that correlations are pretty well valueless.  To counter that, I’ve deliberately entitled this post “Causation IS Correlation”.
How can that be?  If I say “Bob is not Bill” and “Bill is Bob”, haven’t I contradicted myself (unless I’ve cheated by bringing in a different Bill and/or a different Bob)?  Sure, but talking about individuals is very different from talking about sets.  To cover all possibilities, consider two circles, one containing the set of possible causes for a given phenomenon, the other the set of the things that correlate with that phenomenon:

                    Image result for two circles

A:                           Correlations                     Causes

                              Image result for two circles overlapping
B:                                      Correlations       Causes

                                    Image result for one circle inside another

C:                                      Correlations/

For any phenomenon X (say autism) there will be both correlating-causative  and correlating-noncausative factors.  In the A diagram (valid only for Bill-and-Bob cases) causes and correlations are entirely separate, have nothing whatsoever to do with one another.  This is the picture that I’m sure GMO supporters would like to leave you with.  In the B diagram, correlations and causes overlap, so that autism might fall into the overlapping area (where its cause(s) is/are) or outside it, where organic food-sales lie.  GMO supporters wouldn’t mind you leaving with this picture, either.

But the only diagram that fits reality where sets are involved is diagram C, where the circle containing sets of posible causes is entirely enclosed by the circle containing sets of correlations.  What this means is that, for any given phenomenon, there may be correlations that are not really causes, but whatever the cause is, there will be accompanying correlation(s).  So “correlation is not causation” is only a half-truth.  The other half is the title of this post.  Wherever there is a cause there will be a correlation, so correlations form an invaluable tool for shrinking the search space in which the true causes for any phenomenon will be found.  That's why pretty well all sciences use correlations at one time or another--especially when true causes are not readily apparent.  And it's not readily apparent why the U.S.,  a country that prides itself on having the world's most advanced medicine, should have so many diseases that are getting steadily worse

Nobody, least of all Swanson et al., is saying that correlation proves causation.
Here’s all that that paper claims: “The significance and strength of the correlations show that the effects of glyphosate and GE crops on human health should be further investigated.”  No more than that.  The only thing you might question is whether the correlations are strong enough and significant enough to warrant such investigation.

How do you test for strength and significance?  First the strength of a correlation; according to Wikipedia, “There are several correlation coefficients, often denoted ρ or r, measuring the degree of correlation. The most common of these is the Pearson correlation coefficient.”  This, then, is a natural choice for Swanson et al.  Once you arrive at a coefficient, its strength can be measured by its closeness to the ideal: +1.0.  This is to some extent context-dependent—as Wikipedia points out, “A correlation of 0.8 may be very low if one is verifying a physical law using high-quality instruments, but may be regarded as very high in the social sciences where there may be a greater contribution from complicating factors.”  The causes of things like autism or cancer obviously come somewhere between these, so the fact that at least twelve of the Swanson correlations between a negative health condition and GE crop/glyphosate use are higher than 0.95 should be enough to disturb anyone.

But there’s still the issue of significance.  After all, autism/organic-food-sales yielded a 0.99 correlation.  However, purveyors of the GMO jokegraph conveniently forgot about the Bradford Hill criteria.  There are eight of them, of which the most relevant in this case are:

1.     Plausibility: A plausible mechanism between cause and effect. (OK for Swanson: glyphosate has known toxicity. Not OK for organic food sales or organic food consumption--what do you suppose everyone was eating a couple of hundred years ago?)

2.     Coherence: Coherence between epidemiological and laboratory findings increases the likelihood of an effect (there are plenty of laboratory findings for glyphosate damage if you care to look for them—note that most if not all of the papers that give glyphosate a clean bill of health evince no awareness of the facts provided in “Unsafe at Any Dose?”. But there are no lab findings that organic food damages your health.)

Even if all eight criteria are satisfied, as Bradford Hill points out, that still doesn’t amount to proof until the mechanism causing the effect has been indisputably demonstrated.  But correlations as strong and as significant as those the Swanson paper points out demand further and deeper investigation.  Adding to their significance is the fact that these correlations involve not just one or two conditions but nearly two dozen.  One or two might be chance, but two dozen?  Come on!

There is at least one further consideration that must be taken into account if we are to properly evaluate the significance of this particular smoking gun.  It involves cancer.

The National Cancer Institute’s Cancer Trends Progress Report (2011-12) at (apparently the most recent available) shows that while overall cancer incidence rates are falling, the rate for eleven cancer types is rising.  Swanson et al. cover five of those types—pancreatic, thyroid, kidney, liver, and myeloid leukemia--while their sixth, bladder cancer, is not separately listed in the Report.  Why are these cancers bucking the trend?  Nobody knows.   “The causes of pancreatic cancer are mysterious. Although certain risk factors have been identified, the story is far from complete” (WebMD).  “We don't know what causes thyroid cancer” (Cancer Research U.K.)  “It’s not clear what causes most cases of liver cancer” (Mayo Clinic).  “Doctors don't know the causes of kidney cancer” (WebMD).  “We don't know exactly what causes bladder cancer” (Cancer Research U.K.)

Now for the big question: what do the pancreas, the thyroid gland, the liver, the kidneys, and the bladder have in common?  Answer: all but one are directly involved in metabolism, the process by which your body converts what you eat and drink into energy, while that one, the thyroid, regulates the energy thus produced.   We can therefore conclude that whatever is driving the increase in incidence can only be something in the environment that we ingest, that appeared relatively recently, and that is increasing in use.  Apart from pesticides and foods made with GMOs, how many other things can you think of that meet all these criteria?

Finally, please note the green trend lines in some of Swanson et al.'s figures (#s 7, 10 through 15, 24 and 26, for those of you who have the paper to hand).  Nobody is claiming glyphosate or GMOs as the sole cause of the selected diseases, nobody is even claiming that either of these is the sole cause for recent increases in their incidence.  For several of the conditions described, numbers were already increasing from 1980 or even earlier.  What the green trend lines show is the incidence that would have been predicted if no additional cause(s) had emerged after 1990.  But in all nine of these figures, the actual incidence is far higher than the predicted incidence.  These trend lines not only show that some new factor(s) must be present, but they pinpoint  the exact time that incidence started to rise faster than the trend--the same time for each of the nine conditions, which just happens to be the year in which large-scale spraying and consumption of GMO foods really took off. 

I need go no further.  If the Swanson et al. paper is right, the health of millions of Americans is at stake.  On the one hand is all the empirical evidence that they have collected; on the other there are only repeated assurances that GMOs and pesticides are safe, flying in the face of a fact that to the best of my knowledge no GMO advocate has even tried to dispute—the fact that, as I described in “Unsafe at Any Dose?”, we can no longer trust the “safe-dose levels” on which those assurances are based.  The contentions of this paper MUST be investigated and they must be investigated NOW!