I should have mentioned that I have also seen some papers on this question purporting that regression is not the appropriate analysis, that Markov chains should be used. In a Markov chain, the various possible "transition states" are laid out and an analysis of them is carried out. (e.g., 0,1, 2, or 3 runners, 0 outs; 0, 1, 2, or 3 runners 1 out, etc. ..further refinements like 2 strikes vs. 1 or 0 could be defined as well) Anyway, when a Markov analysis is done, I seem to remember the sac bunt came out as a winning strategy even in many pro situations/teams esp. NL where there is no designated hitter. Haven't looked at these stats for a while, though.
It's also impt to see that since few pro teams use the strategy in general, there is little data on how well the strategy works, in general, and therefore sabermetrics gets inaccurate. At the college level this is not true, and as I said, I have no doubt Augie has commissioned a grad student or two to study this.