From the same guy that did the SBNation write-up a while ago. His blog, Football Study Hall, is a pretty good read if you're into stats. This came out of his "Fun Stat Nerd Tidbit" for Louiville - number of returning starters for any position group have weak correlation with an increase in offensive production.

Louisville must replace four starting offensive linemen in 2011. This obviously isn't a good thing, but it might not be quite as bad a thing as we tend to think.
Here are some correlations between returning starters in a given unit and one-year change in Off. F/+:

Correlations To Change In Off. F/+
Returning QB: 0.264
Returning RB: 0.041
Returning WR/TE: 0.283
Returning OL: 0.167
None of these correlations are truly strong, but the correlations connected to the passing game are stronger than those most directly connected to the run game. Meanwhile the OL correlations are quite a bit weaker than those of most projection factors we observe in our Football Outsiders Almanac projections.

Correlations To Change in Rushing S&P+
Returning QB: 0.106
Returning RB: 0.086
Returning OL: 0.207
Data has been hinting more and more at the interchangeability of running backs, and this certainly backs that up a bit. There is but the slightest, weakest correlation between losing your starting running back and seeing your rushing numbers regress the next season. Here we see that there is at least a bit of a correlation between an experienced line and an improved run game, which doesn't hint at greatness for Louisville's Victor Anderson running behind a green line, but still ... these are still minimal correlations.

Correlations To Change in Passing S&P+
Returning QB: 0.273
Returning RB: 0.037
Returning WR/TE: 0.223
Returning OL: 0.067
An experienced quarterback can evidently do wonders for a passing game (who knew?), but these correlations are still surprisingly low. There is apparently an argument to be made that pure talent -- regardless of experience -- is what matters on offense.
Link: http://www.footballstudyhall.com/