Posted by jtrinsey on 3/19/2013 11:35:00 PM (view original):
I got the PPI thing, however, if you look at Woo again as an example, if you add up his total "chances" (A + PO + E + "+" + "-"), it is significantly less than 0.57 per inning, so even if he had a 1.000 FP and 0 "-" plays, he'd still rate as a negative fielder, which doesn't really make sense. I think it makes more sense to add up all of the chances that a player had (A + PO + E + "-"... I'd leave out + plays as they are essentially "bonus" plays) and take what an average fielder (using historical fielding data from the league) at that position would have done. For example, in the world data I tracked, it seemed an average shortstop was about 0.972 FP and about +8 in +/- per 1200 innings.
So for all the chances Woo got (655), I would have expected an average shortstop to make 655*.972 + 8*(1211/1200) = 644.7 outs, whereas Woo actually made 636 outs. This makes him about -8.7 outs worse than average, and, estimating an out as worth about 0.3 runs, he was about -2.6 in FRAA, by my calculation. Obviously this is significantly different than the -16 you have him credited with.
Of course, I deliberately picked the one where we differed the most that I could find in few minutes of scanning. Most of them are pretty close and often within 1 run of each other. I don't know for sure that my way is better, but I suspect that it is less prone to noise because you are essentially forming your baseline comparison off average performance x average chances whereas I am forming from average performance x actual chances. Something to think about.
The "grey area" thing is interesting. I've never thought about that before, but it certainly seems like it could be the case. The way I see it, you're overestimating + and - plays. For example, a + is listed as 4 plays, which would be 0.8 runs, however, an fly out is (in MLB) only worth about 0.3 runs, and not all + plays are outs, some just save the runner from advancing an extra base. However, if the "grey area" exists and every + play is a proxy for a few more good plays that don't show up, then maybe it's on point.
In any case, it's more of a theoretical discussion. Your spreadsheet is really good and I look forward to parsing it a little bit more. Thanks for sharing.
Alright, I promised this for jtrinsey earlier today when I got home.
I took a look at historical Hardball Dynasty – Fantasy Baseball Sim Games - Player Profile: Pat Woo
stats. First number his Yoo's PPI, the second his the season average PPI for SS.
Season 8 - .581, .612
Season 9 - .534, .600
Season 10 - .544, .557
Season 11 - .548, .549
Season 12 - .577, .551
Season 13 - .591. .562
Season 14 - .550, .558
Season 15 - .528, .557
Season 16 - .551, .565
Season 17 - .571, .552
Season 18 - .500, .558
Season 19 - .505, .571
Season 20 - .504, .563
Season 21 - .486, .561
I'm going to group these seasons into two distinct groups, seasons 8 through 17, and 18 through 21, because there is a stark contrast after 17. From 8-17, he averaged a PPI of .558, which is just under the league average of .566 for that time period, or about 2 runs below average over 1000 innings. From 18-21 he averaged .499, well below the average of .563 for the time period, or about 13 runs by my estimation over 1,000 innings.
When looking at his ratings it's easy to see why they dropped off. After season 17 his range dropped 4 points down to 81, and he lost a point in arm strength. But lets look at recommended ratings first before discussing Woo specifically: 80, 85, 85, 85 is what WIS gives us for Shortstops to be league average. From 8-17, Woo was 85,88,82,76, give or take a couple points while he was developing in the earlier seasons. So he's got above average range and glove, but below average arm strength, and way below average arm accuracy.
My theory is that his range was compensating for his below average arm during his career, making it appear he was league average, but when that started to drop off then it started to reveal his faults. By the time we get to season 19, his range is down to 80, exactly average, and his arm strength is down to 78, seven points below average.
This is where I'm going to get into the "grey area" I mentioned. It is my belief that there is a "grey area" in between + and - plays, where a player might make a play that might require above average ratings, but might not necessarily be worthy of being distinguished as great or poor by the engine.
Think of how Zone Rating breaks the field up into different zones for each position. If a player makes a play on a ball outside his zone, he is credited with it , but if he doesn't, it doesn't get count against him. I figure these are + plays in the engine. Now I think there might be another zone, within the larger one, that makes up plays a player should absolutely be making, and if he doesn't, then he is credited with a - play.
Now this is where the grey area comes in. If a player makes the stop on a ball, good for him, he gets credited with the PO or A, if not, no biggy. I thought over the course of a season this might make itself evident by getting more PO and A, but after running numbers for two years, I found this not to be the case in some instances.
I present to you these three individuals
Here are their season lines, along with their fielding ratings that season:
Player A: 297 PO, 573 A, 44 E, 0 +, 14 - 75, 80, 83, 82
Player B: 267 PO, 521 A, 29 E, 0 +, 6 - 79, 82, 87, 84
Player C: 272 PO, 512 A, 17 E, 31 +, 0 - 91, 86, 96, 91
I left innings out for a reason, but I will say A and C played about the same amount of innings, while player B played about 200 less than A and C.
Who is better? Who would you want playing shortstop given equal hitting ratings?
Obviously Player C, right? 31+ plays, which broke a record in the world he plays in, while only 17 errors over the course of a season? The guy is a perennial GG candidate. Not so fast says an un-weighted PPI, which says Player B is the best, at about 10 runs, while Player C is second in at around 4. Player B, with his 14 - plays and 44 errors comes in at 2 runs.
A weighted PPI, with + plays worth four, and - plays worth negative two, changes this. Player C becomes the clear front runner, with 19 runs, Player B is second at 6.3, while Player A is -4.
I will admit that the weights I chose are entirely arbitrary, and I did no statistical work to come up with them, but I figured they are better than running with nothing. I'm willing to work with people to try and come up with a better way of finding out how to weight + and - plays. Maybe it will require taking park effects, and pitching staff into account? It might be worth looking at league average GB/FB data, making adjustments for a the player's pitching staff and park factors. From what I understand of how the engine works, it first decides if a PA is a walk, hit, or an out, and then decides what kind of hit or out, with park effects first playing a role in outs/hits, and then on what type of hit. On this point, If anyone wants to have some theoretical discussions on how to go about this, please sitemail me.
What I don't like about going with just straight up historical field percentage is we know that errors are caused by either a player's glove or arm accuracy rating, while + or - are from range and arm strength. FP doesn't incorporate those two ratings into it because it's solely based on errors. Maybe PPI adjusted for pitching staff/park effects wouldn't be as bad as -16 runs, but in my mind he's definitely not -2 in a year where he's 7 points below rec arm strength and 9 points below rec accuracy.
Player A: Hardball Dynasty – Fantasy Baseball Sim Games - Player Profile: Chick Conroy
Player B: Hardball Dynasty – Fantasy Baseball Sim Games - Player Profile: Orber Pescado
Player C: Hardball Dynasty – Fantasy Baseball Sim Games - Player Profile: Ed Stockton - Yes, he is my player.