Return to Website

Number Watch Web Forum

This forum is about wrong numbers in science, politics and the media. It respects good science and good English.

Number Watch Web Forum
Start a New Topic 
Scientific Statistics

Statistical Nonsense

is bubbling to the surface and someone is saying something about it.

Can we hope that One Armed Bandit Science will see its demise?

Re: Scientific Statistics

The Rutherford quote by Anthony Watts is quite well-known (“If your experiment needs statistics, you ought to have done a better experiment” ), but much less well-known is the identity of the experiment that Rutherford was strongly rumoured to have been attacking. It is Arthur Eddington's 1919 solar eclipse experiment to validate general relativity (GR).

The details of the experiment are:

Theoretical predictions for angle of bending of light by sun's gravity: 0.87" (Newtonian gravity) and 1.74" (Einstein's GR theory)

Experimental results at the two eclipse measuring locations: 1.98"±0.12" (Sobral, North Brazil) and 1.61"±0.30" (Isle of Principe, West Africa).

A two standard deviations criterion was then used by Eddington to claim confirmation of GR.

In the rebuttal of the article by Luboš Motl, he claims that physics works to 5 sigma in contrast to the soft sciences which generally use 2 sigma, and the soft sciences need to raise their standards. General Relativity is regarded by many people (but not particularly by me) as being the most important advance in physics in the 20th Century, but its validation, at least in Einstein's lifetime, was based on a '2 sigma signal'.

Re: Scientific Statistics

"Theoretical predictions for angle of bending of light by sun's gravity: 0.87" (Newtonian gravity) and 1.74" (Einstein's GR theory)"

If there was any reason for the star's apparent position to be slightly out of place under the Newtonian assumption, it cannot be down to the action of gravity. It would have to be due to some optical effect on account of the stellar atmosphere.

Re: Scientific Statistics

I'm quoting the figures directly from a copy of Eddington's book, "Space, Time & Gravitation", first published in 1920. The webpage on the link below gives a derivation for the 'Newtonian' angle prediction. It's one of these things where it depends on whether you regard light as being made up of particles or waves. The book argues that the refractive index effect of the Sun's corona is negligible.


Re: Scientific Statistics

Thanks Dave for posting that link. Fascinating stuff. I remember the Rutherford quote but never knew that the man was referring to that famous experiment.
Slighly off topic but did you see the piece in Physics World (Feb 26)?
"Both answers correct in century-old optics dilemma"
Is there anybody on this Forum that can understand this?"

Re: Scientific Statistics

Interesting. I was labouring under the impression that space curvature was the source of the apparent deflection of starlight, and not that Newtonian gravity could lead to a similar phenomenon.

Re: Scientific Statistics

It means that all our understanding of the real world is in the form of metaphors. Two metaphors for light are a stream of particles and a series of waves. Even a mathematical desciption is amethaphor, because it captures only one aspect of a phenomenon. Science doesn't reveal some kind of absolute truth about reality but gives us a handle on things. (Philosophy just permits us to play a better game of Philosophy)

Re: Scientific Statistics

I've also got a comment on Luboš Motl's claim that all science has to do is work to a 5 sigma confidence level, and everything is OK. I vaguely remembered that cold fusion was supposedly demonstrated to this confidence level, and I managed to find a link written by one of the 1980s pioneers of cold fusion that confirms this:

What killed cold fusion off was its lack of reproducibility rather than any shortcomings in being statistically significant.