Buy The Book
Referencing this article on VORP, some BTF readers make some curious statements, while JC makes some puzzling ones.
I made this post at both blogs:
In the bottom of the 10th inning in today’s BOS game, CIN was down by 2 runs with 2 outs, with Jolbert Cabrera, a veteran journeyman player, on second base. With Griffey on base and the shift on, Jolbert easily took third on the next pitch, since no one was playing near third base.
Now, obviously being on third base is no better than being on second in that situation. In fact, you can make an argument that a foul ball occasionally hits a runner on third in fair territory, so that it is better to be on second, although I don’t think I’ve ever seen that, and it is exceedingly rare (if it has ever happened at all). Plus, being on second base, you can possibly steal the signs or the location and relay it to the batter, although runners rarely if ever do that anymore (I am not sure why). On the other hand (that is the last hand, I think), being on second might be a distraction to the batter, although certainly you are not going to be “jockeying” on the basepaths and you certainly don’t need to have any kind of a lead since it doesn’t matter if you score on a hit.
Now, even if he can walk to third on the pitch, which was apparently the case, there is a non-zero chance that you could trip and fall and get thrown out. I found it curious that a guy who has been playing baseball for almost 30 years would do that (take third).
RE chart for the entirety of Retrosheet years.
To read the first record: with the bases empty and 0 outs, there have been almost 2 million PA with almost 1 million runs scoring from that point to the end of the inning. The average (R/PA) is 0.494 runs, which is labeled as REOI (runs to end of inning). REOI_0 is the percentage of times that there were no runs scored at all in that inning (72.7% in this case).
Excluded are all partial innings, and home halves of 9th and later innings. I am only looking at the base/out state at the start of the PA.
Here’s the SQL
I don’t know when THT rolled out Fangraphs WPA stats, but I like it. They give you the team-level totals. The Angels for example are a total of +6.5 wins (meaning their actual wins minus losses is 13 games… as you can see, the player performances are a perfect match to their teams’ record). That breaks down as -2.0 wins for batting, +4.7 for starters, and +3.7 for relievers.
Fangraphs is likely using a run environment that is too high, meaning it gives too much credit to pitchers. The total for the 30 MLB teams should have batting as exactly 0.0 and starters+relievers as exactly 0.0. Checking now… the total of the 30 MLB teams is -51 wins on offense and obviously +51 on pitching. So, until David A. updated the run environment charts he uses for the win expectancies, you need to mentally add +51 wins per 70,000 or so PA to your hitters (i.e., about +0.1 wins per 200 PA), and remove 51 wins per 16,000 or so IP to your pitchers (about -0.2 wins per 70IP). As you can see, no big deal at the player level so far.
Also note that starters are way behind relievers, as they always are. Relievers get the advantage in that their run environment is being compared against a fixed point (say 4.50 runs per game), when in fact relievers, because they are relievers, should actually be compared to a lower run environment. But, the charts I provided David doesn’t allow for that to be handled. I can do it, but it’s a pain in the butt.
In any case, since you likely want to compare to replacement level, not average, you’ll have to make an adjustment anyway, so you might as well do it after-the-fact, and not in real-time.
Anyway, the reason I happen to discover THT’s WPA charts, is that Rally was talking about the Mariners fielding. It seems that Safeco is always at the center of issues with fielding stats. They’ve got close to the league-low in both infield and outfield fielding. It’s hard to believe that a team with Beltre and Betancourt and Ichiro can be that dismal. Either the other players are so dreadful as to bring them all down, or one or all of these guys aren’t as good as their reps.
At some point, I’ll be looking at Safeco’s PBP data to see if there’s something strange with their data.
Fangraphs is showing the average LI and average win expectancy for each game for a particular day in its sidebar (sorted for your ease). So, you can see which game is the most “exciting” based on these two measures (high LI means alot of swing potential, and a win expectancy that is close to .500 probably means that the game was close for a good part of the game).
Guy has it right when he says:
Fangraphs is starting to add the years, and 1974 was the first one. John Lowenstein and Tommy Hutton are your clutch hitters that year, while Yaz was the choker of the year. The leader in Leverage Index was Tom Murphy (who?) at 2.05. He also happened to be the best reliever that year. So, the Brewers used him perfectly.
Smart people make this mistake all the time, but friend-of-The-Book Joe Sheehan doesn’t:
...with a runner on second and no one out, you have a 62.5% chance of scoring one run. With a runner on third and no one out, you have an 82.7% chance of scoring one run. With one out and no one on, you have a 16.5% chance of scoring one run
He references BBTN, and we have it in The Book as well, but, I’ve also got it online right here. While he did say “one run”, he really meant “at least one run” and not “exactly one run”. And his numbers bear that out. And he was right to compute it that way. With one out and no one on, the chance of scoring exactly one run is 10%. But, the chance of not scoring at all is 83%, meaning the chance of scoring at least one run is 17%. And that is the number we care about. Invariably, we almost always want to know chance of scoring not at all, or its complement. (His 82.7% for runner on 3B looks mighty low. I have it at 86.4 for 1999-2002. Did he mean to say 87.2%?)
Good job to Joe in not falling into the trap. Admiral Ackbar has snagged many a fine folk.
Studes has a little program as to how to calculate WPA Above Baseline. He actually taught himself PHP so that we have a simple calculator. Sweet. I think the algorithm needs modification though. He says:
We’ve talked about something like this in the past, but we’ve never done anything about it. Eric wants to give it a go. I will also reproduce all my emails to him, for those interested in what I had to say:
Well, I’m not doing it team-by-team. But USS Mariner did it for their team. Plus, he also did it by looking at Ichiro Properties and Felix Properties, etc. Well done. I’m looking forward to all the other bloggers doing a similarly exhaustive article.
Note: The positional adjustment may look off by 0.5 wins, but if you look at position+replacement, it’s fine.
I get this question alot, so I might as well cement the response here:
How you wish to handle it is based on who is involved at the time the ball is in flight.
Let’s say that the win expectancy (WE) is .600 when the batter is at the plate. The pitcher makes a great throw causing the batter to pop high in the infield. 99% of the time, it’s a sure out, causing the WE to go down to .550. 1% of the time, it falls for a hit, for a WE of .700. The weighted average is .5515.
So, at the point of contact, you give the pitcher .600 minus .5515, or +.0485. Now, after the point of contact, the pitcher is out of the equation and the shortstop is now in the equation. If he catches the ball, the WE goes from .5515 to .5500. If he doesn’t it goes from .5515 to .7000.
It’s all a matter of tracking what the win expectancy is, at the point that the player is involved.
WAR is wins above replacement. Replacement is defined very specifically for my purposes: it’s the talent level for which you would pay the minimum salary on the open market, or for which you can obtain at minimal cost in a trade.
For nonpitchers, that level is set at -2.25 wins per 162 games, below the average for that league. Since the same stats of an average AL player is better than the same stats of an average NL player (i.e., the AL is the better league), we have different replacement levels. Those levels are -2.5 wins per 162 games in the AL, and -2.0 wins in the NL.
The positional adjustments are:
+1.0 wins C
+0.5 SS/CF
+0.0 2B/3B
-0.5 LF/RF/PH
-1.0 1B
-1.5 DH
These are needed so that an average fielding 1B is not valued the same as an average fielding SS. The DH and PH should be at -2.0 wins (that is, a poor fielding 1B and a DH have the same fielding+positional value). However, as per The Book, it’s harder to DH, so we give them a 0.5 win boost. The PH gets an extra 1 win boost over the DH, because PHing is much harder.
For pitchers, that level is set at .380 win% for starters and .470 for relievers. The exact same pitcher will perform much better in a relief role than in a starter role. So, you need different baselines. You can read the relevant chapter in The Book for that. As with nonpitchers, you need different baselines in each league: .370, .460 in the AL, and .390, .480 in the NL.
For closers, there is a closer replacement level of .570 win%. Any wins above this level gets multiplied by his Leverage Index (LI). That is, while he is not responsible for the LI he finds himself in, he is responsible for his talent that allows him to take advantage of the extra leverage (sort of like Ozzie Smith gets to play SS and reaps the benefit of all the extra opps). We are only giving credit to the closer for the leverage above the .570 win% level. Credit GuyM for this insight.
I’ve been meaning to do this for the longest time, and someone finally beat me to it. It looks like he did a Markov on it. (Note: he calls it WPA, but it’s really WE.)
I was hoping somebody would do this. The sum of each event’s WPA/LI tells you exactly the impact of an event, without the effect of the leverage. If someone can move someone over, WPA/LI will capture it. And thankfully, Pizza was the one who took the plunge. Next time someone talks about “productive outs”, Pizza has the list as to who can do it. Hopefully, he can share a more extended list, with numbers.
http://www.tangotiger.net/wins.html
(LWTS, RE data are in the comments below)
More about WPA/LI (linear weights by game state, with the leverage aspect deflated). Not for the faint of heart:
I’ve been late in continuing my answer to an earlier post regarding the real-time and retrospective look at the play-by-play of a game. Basically, the answer is:
I have this laying around and a few people always ask me for “how often does each base/out frequency” occur. So here you go:
Recent comments
Older comments
Page 1 of 57 pages 1 2 3 > Last »Complete Archive – By Category
Complete Archive – By Date