I think the main reason is to discover if the authors provided a new contribution or just restate their previous work with minor modifications. This is obviously much easier to spot self-plagiarism when the reviewers know where to look.

On the other hand people feel that this gives an advantage to well-known authors because they get the benefit of doubt. Actually because of this, it is quite common to have double-blind reviews in computer science. There is a good discussion of pros and cons of this in the following paper:

http://doi.acm.org/10.1145/1517480.1517492

Having had worked at a research granting council for 10 years and been responsible for upholding the integrity of the council's peer-review process, I see the answer to this in a pragmatic way.

When a grant application is received, the persons processing it and seeking peers to review it need to know who submitted the application in order to identify any potential conflicts of interest. That said, once we did our best to do this, we'd remove any identifying information so the external reviewers (2 minimum), hopefully, were unaware of who was the applicant (but as many of your know the academic world is small and reputations precede people, meaning it can be quite simple to determine, in a small country like Canada, who in which discipline area is doing what).

As far as the adjudication committee goes, it was not blind reviews for similar reasons - we have to give the members the ability to identify potential or real conflicts of interests so they could remove themselves from the discussion as needed.

With regard to questions of preferential treatment for well-known researchers, that is always a risk but I found that a strong program officer would, like I did several times but not regularly, insist that the reviewers judge the application solely on its merits using the published evaluation criteria and not on information not contained therein.

I have seem highly though of researchers send in applications that should never have been submitted and in most cases they were denied funding because, while the committee had confidence that the applicant was capable of doing a good job, they were not comfortable in recommending tax payer money be used to support a half-baked research and work plan. This is not to say that the researchers in question would not take every opportunity to decry the outcome for 'the best application I ever wrote'.

I have some experience in medical writing, peer-review, editing and so on, and I have to conclude that blinded peer-review does not mean un-biasness.

If you think Canada is small, come to Croatia ;) Whole country (non-English speaking) of a size of Montreal oriented on two, maybe three Journals. And, you could only imagine how here “reputations precede people”...

I think there are two major reasons:

(1) cost consideration: non-double-blind review helps referee save time spending and thinking efforts when he/she knows who is the author. Actually, it allows behivour of laziness.

(2) risk consideration: non-double-blind review helps referee take lower risk of making a wrong or silly decision.

Personally, I hate non-double-blind review. However, many journal editors have their own considerations or limits of conditions. In short, it does exist and it should exist in some situation.

The double-blind review may be difficult to implement (especially in physics, mathematics and computer science) also because the results from a paper submitted to a journal are often already available in a preprint or a conference presentation, and then it is not too difficult for the reviewer to identify at least some of the authors.