THE BOOK--Playing The Percentages In Baseball

Jayson Stark has a column today on ESPN.com debating who has been the best player in the AL. His choices are Josh Hamilton and Carlos Quentin. Take a wild guess as to what dumb statistic they lead the league in and take another guess as to how good their defense is.

http://proxy.espn.go.com/chat/chatESPN?event_id=21923&campaign=rss&source=MLBHeadlines

I would prefer that articles like Stark did be written without numbers. 

Old sportswriters complain that all the numbers that sabermetricians present takes the fun out of watching the game.

I think that all the numbers that sportswriters present takes the fun out of reading about the game.

What sportswriters do is exactly what the writers of Star Trek: The Next Generation did: take all of the great quality of the original version, and replace it with techspeak.  That’s all the kept doing on TNG, techspeak after techspeak.  You can always hear the writer’s conferences where they write out the script, then leave out a block of space for (insert intelligent techspeak here) as they continue with the script.

In the old Star Trek, the only one with techspeak was Spock, and you knew it sounded perfectly fine from him.

This is what Stark and Chass and the other sportswriters do.  They use numbers for the sake of using numbers, and not illuminating in any way.

Which sportswriters come closest to the good old Star Trek?  Or have they all “evolved” into the TNG version?

I have said this before and I’ll probably say it again.  Here is what I wrote in the comments section about lwts (and similar stats) and MVP awards:

I’ve said this many times before, but context-neutral linear weights (which is what lwts is - context-neutral, depending upon whether you park or even opponent adjust, of course) is a terrible gauge for an MVP award.  Of course it will generally closely track how a player does in high leverage situations, it will closely track RBI and runs scored, and will closely track how a player is perceived in terms of his “value” to a team.  But in no, way, shape or form should it be the gold standard.

In fact, I find it ludicrous to even use it as a starting point.

A player can have great lwts and do virtually nothing to help his team win.  And vice versa.  Within certain limitations of course.  Not to mention the fact that there is no clear definition of what “helping to win” or “valuable” even means.  For example, some people think that if a player keeps providing win advancement (as when he has a high WPA) then he was valuable, regardless of whether his team actually won any of those games or not.  Another camp, which I happen to be in, but I don’t really care much one way or another, is that if a team doesn’t actually win those games in which a player contributes in terms of win advancement, then that player has provided no value whatsoever.

Pretty much the same thing with positional adjustments, and certainly the voters do not much consider positional value.

So the only thing this list does for me is tell me who has the best context-neutral stats, offensively and defensively.  That is it.  If you want to discuss MVP, you had better discuss and invoke a lot more than that.  Not that a discussion of MVP interests me.  It doesn’t.  One man’s MVP is another man’s non-MVP and that is the sum total of it as far as I am concerned.

We have this discussion annually.  And MGL summarized the issue well in terms of there being 3 camps:
1. context-neutral
2. context-specific, regardless of final outcome
3. context-specific, knowing who won

I agree with MGL that it’s silly to consider #1 for MVP.  You can have great hitting stats when your team is down by 10 or up by 7, and who cares.  You really did nothing to help your team win.

#2 is about “helping in trying to win” and #3 is about “helping in a win”.  I believe that #2 is what we care about, and MGL says #3.

So, I’ll keep bringing back my example:
http://www.fangraphs.com/wins.aspx?date=2004-10-08&team=Angels&dh=0&season=2004

That was Game 3 of the 2004 division series between Sox, Angels, that the Sox won to finish the series.  If you look at the chart, you will see that Guerrero and Ertad were the stars of the Angels that day, along with K-Rod.  On the Sox side, Ortiz, Foulke and Arroyo were their stars. 

When you go to the Play Log, you will see the most dramatic play (prior to extra innings): with the score 6-2 for the Sox in the top of the 7th and 2 outs, bases loaded, Vlad was at bat.  The chance of the Sox winning was 90%.  Vlad hit the slam.  Tie game.  That added +.325 wins.

The Angels lost in extra innings (Ortiz 2-run HR).

Now, I don’t see how you can treat what Vlad did the same as if he struck out in that situation.  The Sox went from a 90% chance of winning to 57.5%.  That play was enormously valuable to the Angels.  That they did not ultimately win does not mean that Vlad provided no value.  And that is because, in baseball, the object is for players to help in their team’s ATTEMPT at winning, and not in their contributions in wins.

The proposal for #3 is that Vlad and KRod were less valuable than Mike Timlin, who faced 5 batters, got 2 outs, and gave up that HR to Vlad!  How can someone argue that Timlin was more valuable than Vlad?  But, if only performances in wins counts, then that’s exactly what is being said.

Would we consider the worst player on a 16-0 Patriots team to be “more valuable” than the best player on a 0-16 team, simply because that team never won a game?

Sorry, but #3, if you go to its logical conclusion, cannot make sense.  We agree that #1 doesn’t really make sense either.  That leaves us with #2 as the one that makes the most sense: contributions towards helping a team trying to win.

And that means using either WPA or WPA/LI.

I think there can be a happy medium.  I would not argue that on a team that goes 0-162, that all players are equally invaluable (zero value).  But I would argue that the most valuable player on a 0-162 team is not nearly as valuable as an equivalent player on a winning team. (And I think that 99% of people would think the same.)

I think you can argue for my perspective but you can’t really argue for Tango’s.  In other words, I can get behind #3 or some combination of #2 and #3, but I cannot get behind #2 by itself with no regard for whether a team wins any games or not. 

And the voters are certainly in the “combo” camp with a lean towards #3.  They are much more likely to give the award to a player on a team that either makes the post-season or comes close.  In order for a losing team or one not close to a post-season berth to have an MVP player, that player has to be far above the other players in whatever stats the voters are looking at.

The reason I can get behind #3 alone is that Tango’s definition of value - a player who contributes towards helping his team win, whether that teams wins or not - is HIS definition.  Certainly not THE definition of value in this context.  An equally (perhaps better) plausible definition is, “that player who contributed towards his teams actual wins.” After all (using that line of thinking), in reality, someone cannot be of value if what they do ultimately leads to no value whatsoever.

Again, it is really a matter of opinion or perspective, but I do agree with Tango who agrees with me on one thing - that #1 is silly and I tire of seeing criticisms of MVP winners because so-and-so (else) had better VORP or EQA (or lwts or whatever).  You see that all the time from the analyst/bloggers.

On the one more hand, if Tango is going to argue for #2 because, after all, whether a team wins or not, the only thing that counts is that a player helped his team “attempt” to win the game, you can certainly use a similar argument to support #1, using the same logic.  That logic is, “I can’t control when I get my hits and outs - everyone knows that - I can only try and get my hits every time I come to the plate.  And by doing so, I will eventually help my team win games.”

Isn’t that similar logic to #2, which is, “I can only do my best to advance my team’s WE.  I cannot fully control whether they win or not?”

#2 and #1 require similar arguments to support them.  So I don’t think you can fairly be dead set against #1 but fully support #2 at the exclusion of #3.

But that is just my opinion on something that I care little about.  Voters can (and do) vote for whomever they want for these silly awards.

First, I’ll have to take back that WPA/LI is plausible.  It isn’t because you will get a great WPA/LI regardless of inning, score.  So, only WPA.

Secondly, the combo of #1 (attempt to win, regardless of game state) and #3 (final result if a win) is #2 (attempt to win, based on game state, regardless of final score, but because a team wins, each good thing is accentuated).

You see, if you look at a winning team’s WPA (always +.50 wins) and you look a losing team’s WPA (always -.50 wins), that gap (1.0 wins) won’t be reflected in their static linear weights.  If you look at the component numbers for each game, you will see that the winning team minus losing team probably shows a theoretical gap of 0.3 linear weights wins.

So, WPA already takes care of part of the issue about whether a team wins or not.  The winning team shows a static linear weights wins of say +0.15 wins, but WPA gives them, always, +0.50 wins.

I think if you are going to argue for a “combo of 2 and 3”, you’ll have to explain exactly what you mean, because #2 is a combo of #1 and #3 and I don’t see why you would necessarily want to weight the 3 portion of that combination more than the 1 portion (or vice versa), any differently than what WPA already does (#2).

What you can do that is sort of WPA-ish is to concede that each win is worth +.50 wins, and therefore, if the LWTS for the team that game comes out to say +.15 for the hitters and +.15 for the pitchers, then you give out an extra +.20 wins to the players in some fashion (probably by overweighting all the positive events).

This way, rather than worrying about the timing of the event by inning score, you are simply conceding that all contributions in wins are overcounted.

Come to think of it, I think this may be the best solution, and if this is what MGL was thinking in terms of “combo”, then I’d be happy with that.

Basically, it would be something like:
1. take hitter’s stats
2. double all his positive contributions in wins
3. double all his negative (outs basically) contributions in losses
4. add it all up

Here’s a link to BR where you can try it yourself:
http://www.baseball-reference.com/pi/bsplit.cgi?n1=pujolal01#outcb-outcb

So for a rough estimation would you take their OPS in wins, double it * games won.  Then subtract 2*(1-OBP from losses) * games lost.  The question arises of how to present this number in a fashion that is easily understood, like wins or runs.  It could be left as a unity metric, any value greater than 100 shows that this player was a positive contribution overall.  But it just seems a little cumbersome.  Ideally we would just use win advancement in wins and loss advancement in losses instead of OPS and “out rate”, but that is a lot of work to go back and do.

2003 for Puljols would look like…
(1.289*2)*83-((1-.356)*2)*74 = 118.662

I would present it exactly as I stated.  You double-count the positive events in wins and double-count the outs in losses.  If he had 25 HR in wins and 12 in losses, then he gets 62 HR. 

Or, if you need to align it properly, multiply his positive events in wins by 1.5 and his outs in wins by 0.75.  Multiply his outs in losses by 1.25 and his positive events in losses by 0.5.

In the above case, his 25 HR become 37.5 and hi 12 HR become 6, for a total of 43.5

I’m looking at the league split stats for 2007.

In wins, the batting team shows Linear Weights Runs of +1.5 runs.  And of course, the pitching team in wins also shows LW runs of +1.5 runs.  So, as a team, the winners are +3.0 runs.  We expect them to be +5.0 runs in order to be +.50 wins.

To get +5.0 Linear Weights runs for the winning team, for batters you need to add 6.7% to their hitting events and subtract 6.7% from their outs; for pitchers you subtract 7.7% from their hits and walks allowed and add 7.7% to their innings.

So, if you want to weight the stats so that the performance in wins gets more weight, that’s what you do.

ARod for example is having a bad clutch year this year (according to WPA).  Worst in the league.  He has a total of 16 HR in wins and 10 in losses (26 total).  If we weight his HR in wins at +6.7% and remove 7.7% from his HR in losses, he gets a total of 26.3 HR.  If you go through all his stats, you will find that he’s doing pretty much what the league does in wins and losses.

His tOPS+ in wins is 125 and in losses it’s 68.  That is, when you compare his OPS in wins to his overall OPS, it’s 25% higher. 

The league average in 2007 was 128 and 70.

125/68 = 1.84
128/70 = 1.83

So, from this measure, ARod’s performance in wins and losses mirrors the league.

And yet, a show of hands please, who here believes that ARod’s performance should not warrant any discounting?

Ortiz in 2005:
http://www.baseball-reference.com/pi/bsplit.cgi?n1=ortizda01&year=2005#outcb-outcb

125/64 as his relative OPS in wins and losses, pretty much mirrors the league.

2005 was his clutch year to end all clutch years.  So, I can’t really get behind a stat where performance in wins is overweighted, especially since:
a) it doesn’t seem to matter as the splits seem to mirror the league
b) it doesn’t really reflect the perceptions of the fan in any way

Going back to the earlier discussion - I’m in the #2 camp with Tango. Assuming for argument’s sake that’s the “correct” method for choosing an MVP, isn’t the next step to move from the implied assumption that every regular season game is of equal importance to a method that somehow weights WPA based on the importance of each game?

Voters sort of do this already - this is why they usually vote for players whose teams made the playoffs or at least came really close. And they will sometimes look at September performance because those games are more important (obviously, that’s only the case if the team is in a tight race). But, it seems like there should be a more systematic way of doing it.

This is probably a little too narrow minded, but if we define the goal of every team as making the playoffs, then we could look at how much each win or loss increased or decreased the team’s chances of making the playoffs based on the standings on that day. So, on opening day, every team’s games are of the same importance, but on September 20th, a team that is tied for the division lead would be playing a very important game and a team that is mathematically eliminated or that has already clinched the division would be playing a game of no importance. And if you’re playing the team that you’re battling for the division, then your game is twice as important.

This system is probably a little too rigid. I don’t know if it’s fair to say that certain games don’t count at all (the Orioles probably do care whether they win 60 games or 80 games), so maybe there should be an arbitrary minimum level of importance for each game. It also potentially puts too much importance on the final days of a tight race (i.e. Red Sox and Yankees tie for the Wild Card, have a one game playoff, and Julio Lugo hits a walk-off grand slam when the team is down by 3 runs with 2 outs - is he the AL MVP?). And there’s some value in getting home field advantage in the playoffs, so that would need to be factored into the equation somehow.

Also, I’m not sure what the ideal way is to calculate odds of making the playoffs. I lean towards just assuming that every team has a 50% chance of winning every game (or with a home/road adjustment) because it doesn’t seem fair for us to punish/reward players based on our expectations of what’s going to happen. One flaw with doing it that way is that it doesn’t allow us to account for the fact from the Yankees’ perspective, an April game against the Red Sox is likely going to wind up being more important than a game against the Orioles even if it’s not clear from the standings at the time.

"to a method that somehow weights WPA based on the importance of each game? “

As you discuss, it puts a tremendous weight on the final game of a tight race.  Heck, game 163 would put one team from .500 of making the playoffs to 1.000.  One game, therefore can add +.5000 wins.

The other 162 games bumped them from +.250 to +.500, or an average of +.0015 per game.  No one’s going to listen if you weight one game 300 times more than the other.

A player’s objective is to help the team win that game.  It is not to help his team get into the playoffs.  Otherwise a team that is mathematically out of it won’t even bother to show up.  But, players perform exactly as I’ve stated: the goal is to help the team win that game.

"A player’s objective… is not to help his team get into the playoffs”?

I think that’s taking it a little too far. I agree that the kind of calculation I’m proposing (let’s call it PPA (Playoff Probability Added)) would produce some screwy-looking results. It probably wouldn’t be wise to use it as the sole basis in determining an MVP choice, but I think it could provide some useful information to be used in addition to WPA (and defense, baserunning, etc.).

Let’s say the Red Sox and Yankees play each other the final weekend of the season and come into the series tied for the wild card. The Red Sox win the first two games and clinch the wild card. Alex Rodriguez goes 0-5 in each game, grounding into double plays in two huge spots. Sunday night, A-Rod goes 4-4 with a walk-off homer. His WPA might be positive overall for the series, but his PPA would be extremely negative. It seems to me that most Yankee fans would agree that PPA is correct in this case.

In order to figure out which method to use, you really have to define what you believe has value.  Does a run have any value if it doesn’t affect a win? The consensus here seems to be no.  But why does a win have value? It’s closer to a run in the big picture than it is to the final goal, which is to make the playoffs (at least in terms of MVP voting, where October is ignored). 

For instance, the Angels have already locked up a playoff spot, for all intents and purposes.  Their wins the rest of the way don’t matter much more than the Padres do.  No individual game really means much in terms of those teams respective playoff odds. 

If we’re going to say that, for purposes of MVP voting, the tangible thing that has value is making the playoffs, then it doesn’t make any sense to include any performance that doesn’t result in a playoff spot.  This is, of course, the BBWAA’s position as a whole. 

So, there’s really two possible ways to look at this - the most valuable player on the teams who made the playoffs, or something like WPA/LI - the guy who did the most with the opportunities he was given. 

I don’t see any reason to believe that we should give a player value credit for performance that he had nothing to do with, so I go for option B.

"I think that’s taking it a little too far. “

I don’t think so.  Getting into the playoffs is a byproduct.  It’s not the objective.  If it is the objective, then what do players on non-contending teams do in September?

It is the objective of fans maybe, and of management maybe.  But not of players.  There’s nothing a player can possibly do different, if his team is 3 games back or 30 games back.  There is something a player can possibly do different if there’s a runner on 1B and you are down by 1 or down by 10.  Or a runner on 3B with 2 outs, or with less than 2 outs.

The only thing a player can do is the batter/pitcher matchup and the runner/defense matchup.  It’s there that he can help his team win the game.  Everything else is a byproduct.

...the guy who did the most with the opportunities he was given.

Oooooh, I like that.  That is the perfect definition of WPA/LI.  You are given a series of game states, 600 or so in a season of varying flavors of complexity, but each one equally weighted… how well do you do in each one?  That’s WPA/LI.

If we look at “Value” in this context, of simply given equally weighted opportunities, regardless of whether your HR will help your team get much closer to the goal of winning the game, then #1 becomes a proper definition of MVP, and is no longer silly.

That is, you do your work every day on the job.  Sometimes, your performance can land a million dollar deal.  Other times, a thousand dollar deal.  Your performance however isn’t worried about the magnitude of the deal, but rather simply leveraging your skills to the tasks at hand.

Is it “valuable” if you land a series of thousand dollar deals, but, through no fault of your own, simply the luck of the draw, the million-dollar deal falls through?

I guess it all depends at this point.

I think what becomes apparent is that you must subscribe to one point-of-view.  You can’t be wishy-washy about it and say that you consider “everything”, which really means you are considering nothing.

At this point, you can make a fairly strong case for any of the three options.

This is a very interesting discussion.  I particularly liked Dave Cameron’s post 15.  He makes a point that I have hazily considered before, but articulating it much better than I ever did: that there is no “stable” position between going for WPA/LI or something like that, and giving it to the best player on a playoff team (or better: the best player on a team that barely made the playoffs).  Tango seems to like an intermediate position, but it is hard for me to think of a principled reason for doing so.

One question: don’t lwts and eqa and all the situation neutral offensive stats track WPA/LI pretty darn closely?  And if so, isn’t it not that bad to use them in MVP discussions if WPA/LI is acceptable?

Yes, WPA/LI will track LWTS very closely.  WPA/LI is more nuanced (things like K with runners on 3B and less than 2 outs is captured perfectly by WPA/LI and ignored by LWTS). To the extent you can argue for #1, LWTS or EqA is a suitable replacement.

That is, you do your work every day on the job.  Sometimes, your performance can land a million dollar deal.  Other times, a thousand dollar deal.  Your performance however isn’t worried about the magnitude of the deal, but rather simply leveraging your skills to the tasks at hand.

Is it “valuable” if you land a series of thousand dollar deals, but, through no fault of your own, simply the luck of the draw, the million-dollar deal falls through?

Yes, yes, yes, we know that!  We’ve talked about all this before.  And my usual response is:

We are talking about giving out an award here - no one gives out an award to the guy in the company who did a really good job but the million dollar deal falls through with no fault to him.  He might get a private pep talk from an enlightened boss or even a raise since the boss knows that he did a good job and that ultimately the deal fell through but that had nothing to do with him.

But, he does not get a trophy that says, “Great job even though we lost all of our deals and our company went bankrupt through no fault of your own!”

That ain’t the way it works in real life.

I know, and you know, and most of the people who visit this site know that these players create “value” even though their team does not necessarily win any games or make the post-season.  But, that is NOT what the award is for, at least in the eyes of the voters and most fans.

As I also often say, if you want to, just create another award called, “Most context-neutral valuable player, regardless of what your team actually does” award.

You can’t give out some criteria for an award after you change the PREVAILING definition of the award.

Doesn’t it seem strange to accept the BBWAA definition of the award and then use stat-geeky approach to pick the award?

Frankly, I would rather recognize and discuss the best players, not those who have created the best stories.  Those get enough play anyway.

The genius of BBWAA is that the lack of strong definition makes this a debate into perpetuity, and otherwise intelligent folks lose their heads over this.