Answering "Who Writes Wikipedia?" in terms of number of surviving words is, no doubt, better than using edit counts. But it is also may not be the best approach for the future, if we are really switching to a "quality over quantity" mentality. At the least, there should be careful choices about what kinds of articles to analyze, before we put too much weight on results like these. My intuition is that Featured Articles and Good Articles have a significantly larger portion of established editors as the main contributors, even by the word count metric.
-Ragesoss
FA's and to some extent Good Articles are not the reason that we are #17 on the most visited sites. If those were the only articles we had I doubt we would have 1% of the readership that we currently enjoy. I think articles past a specific size or maybe size and age woudl be a better metric, though a random sampling of 10k non-stub articles woudl probbly give us what we want.
SKL
I think (as I said in the reply that seems to have gotten lost in the flood) that level of citation (citations per unit length, or similar) would be a good parameter for choosing the sample. With regard to what makes us the #17 most-visited site, you are right (of course). It's defintely not the FA's and GA's. But high-quality articles (by whatever idiosyncratic measure each reader uses) are going to be responsible for future growth in readership and attracting the the most capable and knowledgable editors, (probably) more so than continued growth in the number and size of the unverified articles and content that good anons typically add.
As anecdotal evidence, I was talking to a new professor at my school yesterday (a historian of molecular biology) who was gushing over the Rosalind Franklin article (props to User:Wobble). He's now thinking about coordinating some of his course assignments with WikiProject History of Science.
To repeat what I say on the "expert rebellion" pages, I think we are handling the transition from a quantity-emphasis to a quality-emphasis the right way (i.e., slowly). The leaving experts either have a different set of priorities (i.e., they think that the real-life authority of editors should matter), or are leaving because of problems that we are working on (i.e., edit creep and the inability to preserve a "good" version). But metrics like AaronSW's are important for determining how fast we should change policy/software to favor ease of participation vs. verifiability and/or stability.
I guess ideally, we could have a measure based on each type of sample: random articles, FA's, GA's, long articles, old articles, high-citation articles, most-visited articles. -Ragesoss