David Hogg sends in this bizarre bit of news reporting by Robert Evans:
Until now, in the four decades since it was first posited, no one has convincingly claimed to have glimpsed the Higgs Boson, let alone proved that it actually exists.
At an eagerly awaited briefing on Tuesday at the CERN research centre near Geneva, two independent teams of “Higgs Hunters” – a term they themselves hate – were widely expected to suggest they were fairly confident they had spotted it.
But not confident enough, in the physics world of ultra-precision where certainty has to be measured at nothing less than 100 percent, to announce “a discovery.”
In the jargon, this level is described as 5 sigma . . .
So far, so good. But then comes this doozy:
As one scientist explained, that level of accuracy would equate to the 17th-century discoverer of gravity, Isaac Newton, sitting under his apple tree and a million apples one after another falling on his head without one missing.
Huh? A free Higgs boson to the first person who can come up with a plausible theory of what this “one scientist” could possibly have been thinking.
I believe this is referencing a repeated trial. One million trials without a miss would imply a very tight confidence interval.
I think that on the day that this happened to the unfortunate Newton, there were an infinite number of monkeys taking the day off from typing to remove the apples as they fell.
If 5 sigma is “nothing less than 100 percent”, then what is 6 sigma? Maybe it is a billion apples hitting Newton in the head.
I was also surprised by that.
As I pointed out below, high energy physicists routinely require a result to be at or beyond 5 sigma in order for them to consider it “real”.
That is seriously one of the funniest, most ridiculous pseudo-scientific pieces of bullshit I’ve ever heard. I can’t stop giggling though, so I’m really happy it was published.
At least the mainstream media seems to have finally stopped calling it the “god particle” which is a ridiculous mutation from Leon Lederman calling it “that goddamn particle”.
Would that you were right. I heard it used several times over the past few days on TV.
Might be fun to try to define and simulate a null and alternative hypothesis, build a virtual machine and simulate type one error and power for various critical value and numbers of virtual apples.
I’m confused, how can a hypothesis test ever be used to determine that a model is correct? Also, what’s the null hypothesis?
I believe he meant to say:
” … that level of correctness is equivalent to the 19th century discoverer of water, Jons Jakob Berzelius, shooting himself in the brain with a 50 caliber sniper rifle 2 million times in a row without missing”
Is there just a scratch of irony that the phrase “certainty has to be measured at nothing less than 100 percent” is in reference to a discipline with a foundational principle called the “uncertainty principle”?
High energy physicists generally use p-values, and 5 sigma (two-sided) tail area is just a bit under 10^{-6}. Of course we know that observed p-values are frequently misinterpreted as, for example, “the probability that the null hypothesis is true”, and similar mistakes. My guess is that his figure of a million apples is somehow vaguely connected with the two-sided p-value at about 5 sigma, along with being confused about what p-values mean.
Forgot to note: High energy physicists generally require 5 sigmas to consider the thing they are looking at to be “real”.
I thought you a p-value is for falsifying a model. How do they decide their specific model is correct based on a p-value?
“you a” = a
I’m not familiar with the details of this particular experiment, but apparently they are looking at features that differ from some baseline (the “no Higgs” baseline). They look for features that are some number of sigmas away from the baseline. At least that’s my understanding.
So they are falsifying a model that has no Higgs, apparently.
I’m surprised at the surprise, anyone who has spent much time talking to scientists will surely know that p-values are far more commonly misinterpreted in the standard manner, than correctly understood. Not that this really explains the quote about apples!
Sorry, but I am a scientist, and spend a lot of time talking to my scientist colleagues, who apparently understand p-values better than the scientists you talk to.
I am a scientist (astronomer), and I’ve seen a lot of misunderstanding of p-values in the scientific community. Partly this is because many scientists learn their statistics from other scientists who are not statisticians. Very few scientists learn their statistics from statisticians.
Tim Harford’s podcast, “More or Less”, this week has a piece on physicists’ misunderstanding of p-values, in connection with the Higgs announcement:
http://www.bbc.co.uk/podcasts/series/moreorless
http://downloads.bbc.co.uk/podcasts/radio4/moreorless/moreorless_20111216-1645a.mp3
It’s the first article of several.
The probability of an apple tree large enough to have a million apples existing is very small. Thus, if a million apples fell from a tree, and fell in a bell-shape curve, it would prove… yeah… I got nothin’.
According to Bill Jefferys, 5-sigma corresponds to a little less than one in a million. So if a million apples fall without missing, the number that miss is less than one in a million. Duh.
To get a p value of 95%, you’d have to have less than 5 miss out of 100.
It’s a common misunderstanding that gravity is concerned with simple attraction between bodies. Gravity is actually the theory of why things fall where they fall and what they hit when they do so. Hence, the number of Newton-directed apple plummets is directly proportional to the certainty of the theory of gravity.
I suppose Rosencrantz and Guildenstern Are Dead, but the real question here is how long are we going to be Waiting for the Godot Particle?
this is the weird thing to me – generally speaking you can’t have a null hypothesis that’s “not model X” because there are many possible null hypotheses that correspond to “not model X”. Falsifying one of them doesn’t prove model X. Take for example the null model of independence of variables – you can’t prove a specific correlation structure with a hypothesis test, you can only falsify a specific correlation structure (e.g. independence).
Maybe there’s something about the particle physics that makes this all makes sense, but it’s confusing to me as an outsider…