A comp title—short for "comparative title," sometimes just called a "comp"—is a title used as a predictive tool to help ballpark expected sales on a new title. Publishing houses invariably try to pick their own titles as comp titles, since they generally solicit POS (point-of-sale, i.e. through the register) data from their accounts that are more accurate than BookScan's. If the title is by an author who's previously published with the house, the author's previous title is often used; if not, a title similar in format, publishing season (fall, spring, or summer) and content is used. Houses also want to pick comps that are relatively recent, as it's unlikely POS data will be available for a book that was published in, say, 1989 (or even as recently as 2000).
As I've said before (and I'll say again): you're going to have astoundingly little, if any, say over which title(s) are picked as comps for your book. While you can certainly mention a comparative title or two in your query letter (hint: your book is not Harry Potter meets Twilight meets The Girl with the Dragon Tattoo), the decision is ultimately up to the house, and it's pretty unlikely that your suggestions (which may or may not be informed by appropriate sales figures) will survive from your query letter through your agent's pitch to your editor and your editor's pitch to the publisher.
You can, however, ask your agent to find out what your comp title(s) is/are, then drive yourself moderately insane looking up sales figures and Amazon rankings for that/those title(s). (Remember that BookScan is 1.) expensive and 2.) only reports about 75% of the marketplace, and actual POS information from the publisher won't be shared with you.)
Und jetzt, meine Autoren, an update regarding e-books!
I think comp titles will still be necessary in THE FUTURE (the future... the future...), but not as means by which initial buys are determined. (After all, it's not like Amazon needs to buy 3,000 electronic files of a book in order to sell 3,000 e-books.) I think they'll generally be used as guides to determine target markets and advertising based on historical results, e.g. "We did this with Book X and it worked really well, and since that title's a good comp for Book Y, let's do something similar."
Comp titles will, I think, also be somewhat useful in terms of extrapolating sales figures for budgets and the like, but I imagine more accurate and nuanced predictive tools/methods will be developed as the shift to e-books continues.
Tomorrow's topic: self-publishing, redux!