Orgtheory has an excellent post in which several editors of leading journals in sociology explain what is a good review for them. Worth reading in full; everything applies directly to political science.
A small excerpt:
In my experience as editor, a common complaint among reviewers is that authors have not clearly articulated their work’s core contribution to knowledge. Many readers of articles in scholarly journals may not be reading with an interest in the particular empirical case, but are reading instead for the general ideas or the work’s theoretical argument.
This is a point that I make all the time when I give feedback on PhD projects (not so much when doing reviews for journals), but somehow I have the impression that it does not come across the right way.
In related news, the American Economic Association is abandoning “double blind” peer review. Reviewers will know the identity of the authors (but not vice versa). It seems that Political Analysis has similar plans.