Just how self-correcting is science?

In an extremely timely article in the new issue of American Scientist, Joseph Grcar looks at what fraction of publications in various disciplines are basically corrective (that is, comments, corrigenda,  corrections, retractions, or refutations).  He finds in the sciences in general the correction rate is about 1-1.5% of publications.  This is probably a bit of an underestimate, in my view, since there are new works published that are essentially soft refutations that may not be detected by the methods used here.  Likewise, some fraction of the body of published work (constituting the denominator of that fraction) has no impact (in the sense of never being cited).  Still, that's an interesting number to see.