Tom Tango responds to some questions about Win Probability Added (WPA) at The Book Blog. I like that Tom makes clear the intent of this statistic:
That is what WPA captures…. the quantification of your feelings as the game unfolds, assigned to the players involved.
WPA is not a way to evaluate the talent of a player. WPA is exactly the same as counting a PA 11 times when the bases loaded down by 1 with 2 outs in the bottom of the 9th and counting a PA as almost zero in a blowout game. It is basically ridiculous to think that one PA can inform you on the talent level of a player 1000 times more in one situation than another.
WPA/LI however might be a way to evaluate the talent of a player, since now each PA is exactly worth 1 PA. The only thing we are doing is realizing that baseball might not be a random game, and that a player might tailor his approach based on the base/out inning/score. We don’t know how much he does tailor his approach. That needs to be studied.
When I was reading this, however, something struck me about the FanGraphs game plots. They all make the assumption that the game starts with a 50/50 chance of either team winning. We know that’s not true, however, unless the teams are evenly matched. For example, when the .680 Yankees play the .269 Orioles tonight, the Orioles chance of winning should be around .15 by the log5 method. Even if you think the Yankees are really a 95 win team (.586 WPct) and the Orioles are a 75 win team (.463 WPct), the Orioles start with a chance of winning the game at .379. Shouldn’t the probabilities start and move from there?
Last week I was watching games while watching the live graph move and thought of a similar thing. I noticed that the line would move to a new percentage in a very generic way. It didn’t seem to take account for facts like… when the game is tied in the 9th, the home team actually has a better win probablity than the visitor….it just stuck the probablility at 50%. I don’t know much about how their graphs work, but I don’t think they’re accurate enough for me (yet) after watching them for a week.
David: if you account for the fact that Roy Halladay is pitching at home, you might start the Phillies at .700 to start the game. You “preallocate” +.200 wins to Halladay. Halladay’s in-game WPA is going to come out close to zero over the season, because you do that. There’s nothing wrong with either approach. The big difference is that it’s a huge effort to use identity of players to establish the win expectancies, all to get you to the same end point.
***
Devon: those are my win charts, and yes, you can do like you say and preallocate the win probs that includes the HFA. Once you do that, you can also put in the identity of the players. There’s nothing wrong with any approach. It’s a matter of how much preallocation you want, and how much in-game movement you want.
For example, say it’s the NBA and the home advantage is .620. You have a road and home player that both ended up scoring 30 points, 6 steals, 8 assists, 4 rebounds. They have the same performance. Would you be comfortable seeing one guy with a WPA of +.200 and the other with a WPA of +.120? Because, that’s what happens when you transfer part of the WPA to the home advantage.
@Tangotiger: Thanks, Tom.
Thats the same issue with a number of stats that assume league average teams with league average players at every position in the lineup and batting order.
Players who perform better than other players against good teams and on the road are not given enough credit.
Some players feast against bad teams and at home and are otherwise pretty average against good teams on the road, yet this is not taken into account when comparing players season ending stats.
Jason Bay for example had a 945 OPS against teams over 500 last year (893 aginst sub-500 teams),while Dustin Pedroia had a 705 OPS against the same teams (946 against sub-500 teams). In JD Drews career, he has a 835 OPS against 500+ teams, while it jumps to 951 against sub-500 teams. Kevin Youkilis OTOH has a more even career split with a 890 OPS against sub-500 teams, and 871 against 500+ teams. David Ortiz had a similar small spread in his career.
The league average spread in the AL in 2009 was 780 vs 751, so all other things being equal, players like Bay-2009, Youkilis and Ortiz (pre-2008) should be given more credit for doing well against good teams (as well as hitting well on the road).
“Thats the same issue with a number of stats”
Shouldn’t that be “virtually every stat”? I think I can count on one hand (one finger?) all the stats that are “strength of schedule” adjusted.
You can’t use the logic to bring down WPA for not doing something, and then not apply the same logic to everything else.