An alert reader sent me link to a stupid
article published by Reuters about the Olympics and Astrology.
It's a classic kind of crackpot silliness, which I've described
in numerous articles before. It's yet another example of pareidolia - that is, seeing patterns where there aren't any.
When we look at large quantities of data, there are bound
to be things that look like patterns. In fact, it would be
surprising if there weren't apparent parents for us to find. That's
just the nature of large quantities of data.
In this case, it's an astrologer claiming to have found
astrological correlations in who wins olympic competitions:
Something fishy is happening at the Olympic Games in Beijing. Put it all down to the stars.
Forget training, dedication and determination. An athlete's star sign could be the secret to Olympic gold.
After comparing the birthdates of every Olympic winner since the modern Games began in 1896, British statistician Kenneth Mitchell discovered gold medals really are written in the stars.
He found athletes born in certain months were more likely to thrive in particular events.
Mitchell dubbed the phenomenon "The Pisces Effect" (pisces is Latin for fish) after finding that athletes born under the sign received around 30 percent more medals than any other star sign in events like swimming and water polo.
Swimming and water-polo - but not diving or synchronized swimming. Why? Well, because if you add their data in, the 30% figure goes down, and doesn't look so impressive anymore.
In the history of the Games, the big winners in the overall medals haul were born under the signs of Capricorn, Aquarius and Aries. They boasted a significantly higher number of golds.
In other words, if you look at an arbitrarily chosen 1/4 of the year, athletes with birthdays in that 1/4th of the year have tended to be more likely to win in the olympics. And note the "significantly higher",
without any numbers to support it. It's a fake correlation: With 12 astrological signs, you'd expect to be able to find some way of breaking it onto fourths that produced one fourth that included an uneven distribution.
Checking out the birthdates among the Beijing winners produces some intriguing results.
For fencers looking to deliver a sting in the tail and make it to the podium, Scorpio is the right sign. Two of the three Beijing medallists in the men's individual sabre event were Scorpio, he said.
For pole vaulters charging down the track, it is better to be born under Taurus, the sign of the bull.
Look at the scorpio thing there. Wow, 2 out of 3 medals went to
people who's astrological sign is associated with their sport! Amazing, right? Not really. There are three different individual fencing
events at the olympics - epee, foil, and sabre; and each of them have
separate mens' and womens' events. So there are 18 medals in
fencing. Work it out: 12 signs, 6 groups of 3 medalists. Gosh, what are the odds that one of those 6 groups will include two people with the same
The fact that you can find a fencing competition with two medalists with the same astrological sign really isn't a surprise. But it sounds nice because Scorpio's symbol has a stinger, and it's got a strike sort of like a fencing attack. But that's post-facto rationalization; for just about any other astrological sign, you could find a way of justifying it. If it was Taurus or Aries, you could talk about the charge in to strike. Sagittarius, the archer, should be obvious. Leo, the lion, like a great hunter. Aquarius, the fluid strike and defense. And so on: "patterns" like matching astrological signs are inevitable, and the human mind can always find a reason for associating the sign with the event.
Same thing with the vaulters. Why would pole-vault be associated with a charging bull? Well, because it just happened that there were winners
in the pole-vault with that sign - and you could describing the run-up
to the vault as "charging". Again, you could find all sorts of justifications if it were a different sign.
Of course, since this all sounds silly, the article needed to
throw in some reason to pretend that the crackpot behind this
rubbish was credible:
Even Mitchell was surprised by his own findings which he said were conclusive "and I really mean conclusive".
"I am talking of odds against chance of hundreds of thousands to one", he said, explaining the research he undertook after being made redundant from his IT job.
"And just for the record, I know a thing or two about statistics. I have a PhD from Glasgow University on statistical ecology and a further 33 years working on statistical data analysis," he explained on his website.
I think I know why this guy was "made redundant" from his job. It's because he's an idiot.
The most classic mathematical error of probability is what I call
perspective errors. These are errors where you take an event after
its occurred, and then work out a probability for it as if you were predicting it before it occurred. That's something that anyone with
any clue about probability should be well acquainted with; it's an
absolute textbook abuse. In fact, pretty much every probability textbook
includes a discussion of one particular example of perspective errors: the
infamous card-shuffling probability problem. Take a deck of cards, and shuffle it. Look at the order of the cards. What was the probability of getting that particular order? Roughly, 1 in 1068.
But some card ordering had to be the result. The probability of getting
a result from shuffling the cards is exactly 100%; and the odds of any particular outcome are the same. You can't
take the result of the shuffle after the fact, and then say that it's
miraculous, because the odds of this happening were so small!
But that's what the crackpot astrologer is saying. He's taking a bunch
of facts that he's gathered after they've occurred, finding an apparent pattern to them, and then calculating the a priori odds of the pattern
he discovered. But patterns are inevitable.
There is a way to check things like this: there are some very nice
tools from the world of Bayesian probability that allow you to work
out the probability that the pattern you're seeing is really unlikely. But bozo's like this guy never bother to do that. For one thing, it's hard. Working out all the factors in the correct way is
a laborious process, where it's easy to make mistakes, miss elements, etc. But more importantly, the proper analysis has a tendency to make
an apparently impressive thing look unimpressive. You do a lot of work, and
and the end, that work tells you that you're wrong. No crackpot is going
to risk that!
And now, for the real prize of this piece. They saved the stupidest bit for last.
Explaining his eureka moment with all the zeal of a statistical crusader, he concluded: "Did you know that the distribution of Olympic swimming medallists against the tropical astrological zodiac signs can be almost exactly mapped by a polynomial function of the third degree?
"That's one to shut people up at a pub." (Editing by Nick Macfie.).
One of my least favorite topics back when I was studying CS in college was numerical analysis. I hated it. In NA, we spent a lot of
time on curve fitting. If you never went through the torture of an NA class,
curve fitting is a technique of taking a set of data points, and trying
to find a polynomial that fits those points.
Given any set of two points, you can find a line that goes through them. Given a collection of points that aren't a precise line, you can use linear regression to find the line that's the closest match (for various definitions of closest.)
Curve fitting is basically a generalization of that. Given any three points, you can calculate a quadratic equation that will fit them perfectly. Given a collection of data points, you can find the quadratic that's closest.
And so on. One interesting property of polynomials is that as the degree increases, it gets easier to create a polynomial that's very close to matching those points. Given almost any twelve points, I can find a cubic
equation that comes pretty close to matching them.
But the distribution of astrological signs isn't just any set of twelve points. In fact, it's a set of twelve points that we would expect to follow something close to a normal distribution. And for any set of twelve points with a normal distribution, I can guarantee that there's a cubic curve that's a really good match. There's just no way that that's not
an expected result.
Pretty much the only thing that that kind of reasoning is good for is impressing the folks down at the pub - with a few pints under
your belt, it might sound pretty good.