Adjusted WPA

by user Airjakub2

One of my new favorite statistics is Win Probability Added (WPA). The concept of WPA is very simple- it is the effect a player has, in a given circumstance, on his team's chances of winning a game.

For example, if a hitter comes up with runners on first and second and two out in the bottom of the seventh inning of a tie game, his team has a 63% chance of winning. If that batter hits a home run, that percentage goes up to 95%. In this at-bat, this hitter had produced a WPA of .32.

Recently there has been more focus on WPA. The percentages above were calculated using this site. A couple years ago The Hardball Times' Dave Studeman wrote this fantastic overview of WPA. And there is even an entire site devoted to WPA- FanGraphs.

These sites piqued my interest in this statistic, but I found something missing. Stats like VORP and MLVr are adjusted for positions, but I had never seen this done with WPA.

I took the 2006 WPA statistics from FanGraphs, and sorted them by position. For each position, I then calculated the average WPA that came from an at-bat at that position. The results are as follows.

1B: .25
DH: .19
LF: .13
RF: .13
3B: .07
CF: -.01
2B: -.04
SS: -.08
C: -.18

Check out my site, Vegas Watch, for more details and VWI updates.

DNLLegend

925 days ago

Score 2

One of my problems with WPA is that it assumes that a non-event is possible. Using your example to illustrate:

Player comes into game, and his team has a 63% chance of winning. The problem is that, when his plate appearance is over, that percentage is going to change. The player SHOULD be measured against what the typical batter would DO in that situation -- that is, against the result, not against the situation itself.

What WPA does, by measuring from the situation and not from the result, is overvalue on-base percentage. Why? Because typically, the Win% will go down if the player makes an out (without driving in a run), and up otherwise. But most often, the former is going to be true. Therefore, the correct baseline-stat is actually lower than the one being used.

While everyone is being over-penalized, the bad effect is felt most by low OBP players who are losing a higher percentage of WPA points.

Permalink | Reply

DNLLegend

925 days ago

Score 0

I'm amazed that DH is, compared to RF and 1B, so low. Can you do AL-only numbers?

Permalink | Reply

Airjakub2Soccer Kid

925 days ago

Score 0

I agree that the DH component is weird. But I got these numbers by using data from 2003-06 (which is a ton of data, trust me, I had to go through all of it). When I actually calculate Adjusted WPA on my site I actually use the .25 number for DHs, since I think that is obviously more realistic. I strongly disagree with your complain about WPA. By definition, the average WPA is 0. Think of it this way- ignoring home field advantage, each team starts each game with a WPA of .5. So for each game, one team will have a cumulative .5 WPA, and the other a -.5 WPA, which is obviously then distributed among that teams players. The average WPA in each situation is also 0- that is the point of the stat. It is true that players make outs more often than not, but it is also true that the average non-out plate appearance has a larger impact on the game than the average plate appearance in which the player makes an out. So thse two factors balance each other out.

Permalink | Reply

DNLLegend

925 days ago

Score 0

That isn't really responsive to my gripe. We're trying to measure how good a player is compared to the average player. WPA doesn't do that because

it assumes that a player can leave a situation with the same value in which he entered. He can't.

When, e.g., Jose Reyes comes off to lead off a game, there are only five outcomes: no runners on and out, a runner on first and no outs, a runner on second and no outs, a runner on third and no outs, or a run scored and no outs. The situation he was met with -- no runners on and no outs -- is not a possible outcome. Therefore, he should be measured against the typical outcome, and not the given situation.

In this case, we can easily measure the typical outcome. It's going to be about:

• .65 * none on, one out
• .25 * runner on first, no outs
• .07 * runner on second, no outs
• .01 * runner on third, no outs
• .01 * run in, no outs (solo leadoff homer)

If we aggregate that, and compare it to Reyes' actual result, we'll get a true idea of how he fared in the situation versus the average player. If we do it versus a 50% winning percentage, we're over-valuing OBP.

Permalink

Airjakub2Soccer Kid

925 days ago

Score 1

All of this is true. But if you were to aggregate it, the aggregate would be the same as the original situation every time. This makes sense- in fact, this is the way it has to be. This is because Win Probability, by definition, is what percentage of the time the average team, with a collection of average players, would win the game. So Win Probability assums that an average player is up, and thus is actually an aggregation of that average player's performance. Basically, the entire point of WPA is doing exactly what you say in your post. The Win Probability to start off a game is .5 because, on average, each team's Win Probability after that first hitter is .5. So whatever Reyes does to raise his team's WP above .5 in that situation is that much better than an average hitter would have done.

Permalink | Reply

MetsJetsDevilsDraft Pick

924 days ago

Score 0

And if I combine my players DUI with his AARP, I should be able to create an NRA above 1. However, I often find that the ADL gets in the way of my pitcher's KKK, which was on the scoreboard until John Maine struck out his 4th batter in the 7th inning. Thank you for once again trying to reduce the game of baseball to a bunch of formulas and algorythms. I don't need Excel or evern a abacus to tell me that with he game on the line, historically speaking, I would rather Mike Piazza at bat than Alex Rodriguez. I don't need to read the Hardball Times or study Billy Beane's bathroom habits to tell me that I would rather Mariano Rivera have the ball in the 9th inning with a 1 run lead and basis loaded than Armando Benitez. How, you might ask, can I do this without the aid of a computer. You see, I have these things called eyes, and they go very well with this other thing called baseball sense.

Permalink | Reply

Davis21wylieMVP

924 days ago

Score 0

Sounds like someone woke up on the wrong side of the rock...

Permalink

MetsJetsDevilsDraft Pick

924 days ago

Score 0

No. Its just kinda my thing to rip into people who invest too much time trying to reduce baseball into a series of algorythms. Just ask bball3345.

Permalink

Davis21wylieMVP

924 days ago

Score 0

And it's just kinda my thing to subtly mock people who rip into people who invest too much time trying to reduce baseball into a series of algorithms. Just ask Bball3345.

Permalink

Airjakub2Soccer Kid

924 days ago

Score 0

I thought I had a lot of time on my hands at work but you- you, my friend, have an extraordinary amount of time on your hands.

Permalink

Tyrone BriggsHall of Famer

924 days ago

Score -1

A statistical study needs to be conducted pronto to determine the validity of statistical studies in pro sports.

Permalink | Reply

Anonymous Fanatic #1

924 days ago

Score -1

All you've done is say that 1Bs are good hitters. Would anyone disagree with this? You've also pointed out that Cs are bad hitters. WPA, IMO, really shouldn't be used in aggregate. They're game-situational and should be used game by game. The only way to use them in aggregate, maybe, is using WPA+ and WPA-. If you just mush up all the #s together I think you lose any relevance. And really, you can't just adjust for position, since position doesn't really matter. What you should be adjusting for is lineup slot. That's what matters.

Permalink | Reply

Airjakub2Soccer Kid

924 days ago

Score 0

Adjust for lineup slot? Seriously?

Permalink