Is the Universe Screwy?


In April 1997 Borge Nodland and John P. Ralston created much excitement in the media by their claim for evidence of "cosmic birefringence", i.e. that the plane of polarization of radiation gradually rotates as it travels through the universe.

The result was first announced at the PANIC '96 conference: see their presentation "Cosmologically Screwy Light". A longer account was published in Physical Review Letters. 78, 3043, on 21 April 1997, entitled Indication of Anisotropy in Electromagnetic Propagation over Cosmological Distances.

There is a popular account (with fancy graphics) on Borge Nodland's web site, and much the same material is available as a Physics News Graphic. The "discovery" was also covered by the New York Times, the London Daily Telegraph and other newspapers. A more sceptical note was sounded by a news item in Science.

Unfortunately, a fatal flaw in the argument was found withing three days by Eisenstein & Bunn. This was spelt out in embarassing detail a few days later by Carroll & Field. Far from being an effect significant at the level of 0.6%, the anisotropy found by Nodland and Ralston is less than that expected by chance in about 50% of cases. There is a good summary, which also has links to even more on-line press coverage of the original announcement, on Sean Carroll's website,

I got interested in this because it was obvious immediately that the claimed rotation was inconsistent with the known polarization structure of distant radio galaxies and quasars, which is one of my main research interests. Using data on half a dozen objects from a couple of recent projects I was able to set an upper limit on "cosmic birefringence" about thirty times lower than Nodland and Ralston originally claimed: a short comment describing this has been submitted to Phys. Rev. Lett., and has been placed on the Los Alamos preprint archive as astro-ph/9704285. The same point is made, with even more stringent limits, in a paper by Wardle, Perley & Cohen: see also their NRAO press release.

Nodland and Ralston are not the type to give up easily. They have posted a very critical response to Eisenstein and Bunn. However Eisenstein & Bunn have appended a robust defense to their on-line preprint, finding it unnecessary to change their original text in any way.

An even more critical response to my comment has been posted by Nodland and Ralston. Embarassingly to me, they did find a mistake in my calculations based on their theory, but this does not affect my main argument or results. Given that my results, taken at face value, are obviously inconsistent with their theory, they imply that I have deliberately set out to mislead by selecting grossly unrepresentative data. This is not a move I have previously encountered in scientific debates, and it is difficult to know how to deal with it, except to deny the allegation. Following the example of Eisenstein & Bunn, I have appended a couple of pages to my on-line preprint countering Nodland & Ralston's objections and explaining how the data was really selected; I have also revised the comment itself to fix the mistake and respond to a couple of their minor criticisms, and to shorten it, as it turned out to be 10% over the allowed length.

Update: September 1997

Yet another take on the original statistical analysis has been submitted: Laredo, Flanagan & Wasserman use Bayesian statistics, in which there is a lot less room for debate about the "right" way to approach the problem. Again, they find there is no significant effect. Several theorists have now published explanations of this (non)-effect: links are on Sean Carroll's page.

Ralston & Nodland presented an update on their work at a recent conference: they seem to be retreating from their original explanation (as it is ruled out by Wardle, Perley & Cohen and my own work, although they won't admit this in so many words) and concentrating on the claim that their statistical analysis has turned up something interesting. They illustrate this with a graph that looks a lot more convincing than the ones they published, but in fact it is the same data, manipulated according to one of the oldest tricks in the statistical book: if X is not correlated with Y, you can multiply both axes by a third variable (in their case the factor cos[gamma]) and lo and behold, a wonderful correlation! (Cos[gamma] is correlated with itself!)

The 8th September 1997 issue of Phys. Rev. Lett. (Vol 79, Number 10) publishes the paper by Wardle, Perley & Cohen (page 1801), the comment by Eisenstein & Bunn (p. 1957) and a shortened version of Nodland & Ralston's reply (p. 1958), in which, according to Physics News Update they "refute" Eisenstein & Bunn. I hope Update means "deny"; in my opinion, this reply just emphasises N&R's poor grasp of statistics.

Meanwhile my paper was rejected by Phys. Rev. Lett. on the grounds that it does not meet their technical definition of a "Comment". I plan to publish a slightly longer version as a free-standing paper.

Back to Jodrell Bank home page
Last modified: 1997 September 16
J. P. Leahy