I’ve always thought personally that the 8th inning is far more important for building back momentum (for a team that is currently losing) or for putting the game away (for a team that is currently winning). Of course, you occasionally get the 9th-inning rally but it just seems that more things happen in the 8th.

Tango, already did. That 20.5% figure is only among those who appeared in relief for the game. I double checked it and it’s accurate.

I agree, and I’m a huge Padre and Trevor fan. But you really can’t argue with it anymore. The selection committee knew the controversy would start when they finally decided to induct Bruce Sutter, which opened the door for Rollie Fingers, and gave hope to Goose Gossage, Lee Smith, Dave Righetti, and Mariano Rivera, among others. Don Sutton, a good starting pitcher, is in based on wins, which itself was based in large part on his longevity. The selection committee is of course anxious about setting precedents and thus opening the excusive door too wide too quickly. But even though the Bruce Sutter induction pretty much ensures that Trevor Hoffman and the others mentioned above will get in, I still think it’s a stretch.

Guy, it surprised the heck out of me too. To answer your question, I only looked at one team at a time. So, each team in a game had a reliever (assuming that they used one) who got a “most WPA added” award (or at least a least WPA subtracted!) So, the “savior” is only competing against players on his own team.
Also, looking through the top 50 in WPA (I assume we’re looking at the same fangraphs chart), shows a bunch of guys who have 0 or 1 game started, but who are also not their team’s primary closer. (Some are guys who have been setup guys but became closers, e.g. Alan Embree). In the top 15 or so, I see R. Soriano, Betancourt, and Okajima.
A few thoughts occur to me on why this might be happening: Closers are generally more consistent than their middle relieving brethren (hence, I believe why they got the ninth inning gig). “Saviors” added an average of .077 wins during their time on the mound with a SD of .068. All other relivers added an average of -.008 wins (so they basically broke even), but with an SD of .151 wins. Since starters now usually go 6 innings, the closer generally gets a chance at a save when the guys in front of him (pitching in some high leverage 7th and 8th inning situations) do their job (i.e. preserve the lead, hopefully involving some positive WPA). So, it’s not surprising that the guys who had those assignments rack up some good WPA points in front of the closer.
Closers generally do positive things consistently, so they build up the WPA, while the middle relievers are less reliable, so while one game, they build up the WPA, the next game, they give it all back. It’s a theory.

Ah, a small mistake in my methodology from my previous comment. -.008 is the average for all relievers for both teams, except the guy who recorded the save. That would be why they all break even. I restricted it just to relievers from the winning team (i.e., the team that got the save). The closer numbers are the same, but the average reliever on the winning team in a game where a save was recorded, but he did not record that save added .047 wins on average, but with a standard deviation of .132. So, we’d need to know how many scores in that distrbution fall above .077. In that distribution, .077 has a z-score of +.22. That means of all the relievers WPA scores, 42% of scores will fall above .077 on a pitcher-by-pitcher basis. If a team uses two relievers to get to the closer, odds are roughly 66% that one of them will outscore the closer. If they use more than two, those odds go up.

Actually, you need to compare the two distributions — all closers are not .077 in every game. Also, the more relievers in a given game the fewer IP for each, and so lower WPA on average. I’m not sure how this all nets out, but .40 looks a lot more realistic than .20.
Still a surprising result, I think. The main contributors, in addition to your observation that 8th inning LI is often quite high, look to be:
1) 3-run saves are really very low LI;
2) 1- and 2-out saves don’t generate much WPA, even if LI reasonably high;
3) another reliever often goes 2 IP, a big edge even if his LI is a bit lower.

didnt feel like reading all the comments, so someone might have said this already….
the win expectancy numbers have to be skewed so that there are fewer “swings of fate” in the 9th inning, because the best pitcher pitches that inning, preserving the win. if the best pitcher pitched the 8th inning for the last 30 years, then the win expectancy numbers would be different for everyone. someone who is smarter than I am can extrapolate on that thought, im not in the mood to think right now

I did do some thinking on the subject. Tango has advocated the use of context neutral wins (WPA/LI) added as a way to calculate clutch. When I get back to my laptop, I’ll take a look to see which relievers are adding the most context neutral wins.