THE BOOK--Playing The Percentages In Baseball

What about Bill James games score method, or something along those lines?  Would something like that averaged (or whatever, I’m not a math guy) over a season be a good indicator?

Tango - Could you explain to me the advantages of WPA/LI over RE24?  From your previous explanations of WPA/LI it seems that it provides some context of the situation, but not as much as WPA, which is also what RE24 does.  I just haven’t figured out why it might be superior to RE24 which seems much less complex to compute than WPA/LI.

RE24 is a precursor to WPA/LI.

Both properly handle the situation with a runner on 3B and less than 2 outs (a K is a big minus to the offense), only WPA/LI takes that particular situation and distinguishes it whether the batting team is up by 2, down by 1, in the 3rd inning, or it’s the bottom of the 9th.

So, WPA/LI is pretty much RE24, but more.  Say WE24, or more closely (if you have 18 half innings, and 21 score differentials to consider, from -10 to +10), we can call it WE9072.

It’s not exactly that though.  RE/boLI, where boLI is the LI of that base/out state, would be a better comparison point.

Now I really am confused.  What you described in post #2 sounds exactly like WPA to me.  I thought the purpose of dividing by LI was to neutralize much of the game context (inning and score) because LI would vary in parallel to WPA for those factors. Have you done a rank order comparison of players’cumlative RE24, WPA/LI, and WPA?  Does WPA/LI fall somewhere between RE24 and WPA for all players?

I’m having trouble understanding the Dickson point. The Cubs were shut out that day, the best defensive result the Phillies could’ve hoped for, and I’m pretty sure Dickson was the primary reason for that. How can he not get a star of the game?

Greenback06 - because the fielders behind Dickson and luck had more to do with the favorable team outcome than Dickson did (5 BBs, several XBHs)

#5--here’s how I understand it: if a certain PA has an LI of 5, WPA will count it five times as much as a typical PA while RE24 will count it exactly the same as any other PA. WPA/LI (which is an awful name, incidentally) will also count it exactly the same as any other PA.

Just as RE24 is custom linear weights by base/out state, WPA/LI is custom linear weights by game state. Dividing WPA by LI is merely a shortcut for getting at the custom linear weights.

In other words, if you figured out the correct LW coefficients for runners on 1st& 3rd/1 out/bottom of the 7th/tie game...and then got the correct coefficients for every other base/out/inning/score situation...and then applied those custom linear weights to players’ performance in each game state...you’d come up with the exact same number you’d get from simply dividing WPA by LI.

I think so, anyway.

Anthony - Thanks for trying to help me, but I don’t see how you are reconciling the first two paragraphs of your response.  According to Fangraphs RE24 and WPA are calculated with the same methodology.  You assign a value for both the before and after states of a play from either the run expectancy tables (RE24) or the win expectancy tables (WPA) and then you subtract the before state from the after state, including any runs scored on the play in the after state for RE24.

So it seems to me that your first sentence of paragraph 2 should read that “RE24 is custom linear weights by base/out state, and WPA (not WPA/LI) is custom linear weights by game state.  Which still leaves me wondering what the effect of dividing WPA by LI will be.  I had believed similarly to what you stated in paragraph 1, that dividing WPA by LI basically cancelled out the leverage of the game state and left you with a number comparable to RE24 in meaning.  In a other words, a calculation of win value by base/out state with the effects of inning and score cancelled out by dividing WPA by LI.  Which would seem to be the same result that you would get by dividing RE24 by average runs per win.  That’s why I would like to get Tango’s version of what advantages WPA/LI offers over RE24 since he invented the concept.

I also still have trouble grokking WPA/LI. One thing I would love to see are some specific comparisons of all the methods for different situations. For example, what value does a player get in wOBA, RE24, WPA, RE24/boLI, WPA/LI, and WPA/LI * boLI for a game-winning home run? For a home run in a blowout? Etc.

Also, I’d be interested in any follow-up to some of the points raised by dcj in these threads:

http://www.insidethebook.com/ee/index.php/site/comments/more_updates_at_fangraphs/

http://www.insidethebook.com/ee/index.php/site/comments/the_monthly_wpa_primer_and_me_thread/

I’m interested in the pitching vs. fielding issue that cropped up in the Primer discussion.

Why couldn’t you do as follows:

In the case of any BIP, give the pitcher the credit/debit for the type of BIP—ground ball, line drive, infield fly, or outfield fly.  The difference between that and the ultimate outcome (hit/error vs. out) goes to the fielder.

You would have different WPA’s by base situation.  For example, a ground ball with a man on first and less than two out would be worth more to the defense, because of the chance of a DP.  Conversely, with a man on third and less than two out, an outfield fly would be worth less (due to possible SF), as would a ground ball (more hits if the infield is in).

I think this would truly capture, as WPA/LI intends, the “feel” of the game.  Say the Yankees have a 1-run lead with two out and runners on second and third in the 9th; Rivera is on the mound and Mauer is up. If Mauer pops up, Rivera gets virtually all the credit.  The fans say, “How about that Mariano?”

If, instead, Mauer hits a line drive that’s caught by Teixeira, the fans say, “Boy, Tex really saved the day.”

LI, if I’m understanding correctly, is more-or-less the equivalency of how many normal PAs a given PA is worth.  The outcomes of a PA could be worth 5 times more than normal in a given high leverage situation.  In both RE24 and WPA/LI, the PAs are always worth 1.  Both are custom linear weights designed to fit the game state, but because RE24 only uses partial game state as its inputs (because it only cares about run expectancy, so additional information that only relates to win expectancy is discarded) and WPA/LI considers additional information, they sometimes arrive at different weights for different events.  The overall value of those events will still be the same, but that value might be divided differently.

The classic example is bases loaded, 2 outs, bottom of the 9th of a tie game.  In terms of run expectancy, a home run is worth 4 runs minus whatever the run value of having the bases loaded is.  A walk is worth 1 run.  So in that case, the HR is worth much more than a walk.  In terms of win expectancy, a HR and walk are exactly the same; 1 - whatever the win expectancy was at the start of the play (divided by the leverage index, for WPA/LI).  So if you use RE24, a HR is worth more than it should be and a walk less.  WPA/LI brings the value of the HR down and raises the value of the walk until they are both worth the same amount. 

“In the case of any BIP, give the pitcher the credit/debit for the type of BIP—ground ball, line drive, infield fly, or outfield fly.”

That’s essentially what tRA is.  I’m not sure if it’s available anywhere on a single-game level, though, so that much might have to be calculated on your own.  I’m not sure.

Kinkaid - There were 187079 batter events last year.  There were 212 walkoffs.  I am well aware of the classic example that you gave above.  Walkoffs have always been problematic in the calculation of a player’s offensive value.  That is why partial innings are eliniinated from the calculation of RE tables. 

I am more interested in the overall impact of the differences between RE24 and WPA/LI on the ranking and evaluation of players, in other words, the 186867 other plays.  From your above description it sounds like you would predict that WPA/LI would always be between the RE24 and WPA of a play.  Is that correct?

No, I don’t see why that would be the case.  To use the walk-off scenario again (I would use something else, but I don’t have win probabilities or LI in any of the databases I use, so I have to stick with the obvious conceptual examples), a players WPA/LI would be smaller than his WPA because LI is greater than 1.  His WPA/LI could also be smaller than his RE24 depending on what he does, because some events are favoured by RE24, like a HR.  It may be that that tends to be the case (I have no idea), but I see no reason why is would always be the case, and there are plays where that is not true.

Hopefully Tango has more insight on what specifically alters the rankings, but as far as I can tell, they are just different because they use different inputs. RE24 is different from linear weights because linear weights assumes that you can measure a player’s impact by crediting him with the average value of each event across all situations, while RE24 assumes that the differences between different base-out states have an appreciable impact on what happens.  If you want to consider how a hitter changes his hitting approach with a runner on third and less than 2 outs, for example, then you might prefer RE24, because linear weights would just say that a sacrifice fly is a fly ball that happened to come in an advantageous situation, whereas RE24 would say something more like he hit a fly ball because it was advantageous in that situation.

If you want to not only consider the unique values of different base-out states, but also of the inning, half of the inning, and the score, then you would use WPA/LI instead.  Linear weights assumes a fly ball with the bases empty or with two outs should be measured the same as one in a sac fly situation (unless you include sac flies as in your linear weights equation, which you could).  RE24 assumes that a fly ball with a runner on third and less than two outs has to be measured differently than a fly ball with none on, but that that fly ball is measured the same at any point in the game, be it the top of the third or bottom of the ninth or whenever, and whether you are tied or down 8 runs or up 3 runs.  WPA/LI assumes that a sac fly in the bottom of the 9th has to be measured differently than one in the top of the third, and that one in a tie game has to be measured differently from one in a 3-run game.  They all just make different assumptions about how much context to consider, from no context (linear weights), to base-out state (RE24), to base-out state and score and inning (WPA/LI).  They give you different results because they use different inputs.

I guess another specific scenario where WPA/LI and RE24 would differ is a sac fly.  In a tie game, or down 1 run, the run is very valuable in terms of win probability, and the out is not so costly compared to that value.  Down 5 runs, the out is more costly and the run less valuable.  So WPA/LI would value a sac fly in a tight game more than it would value a sac fly trailing late in a game.  RE24 would consider them the same.  I don’t know if there is a systematic way they affect how players rate, but a strong proponent of WPA/LI would probably argue something like the hitters who adjust their approach based on the inning and score as well as the base-out state will tend to rate better in WPA/LI, or possibly more versatile hitters who are more capable of adjusting their approach.  I might predict Albert Pujols will have a higher WPA/LI than RE24 going forward if I think he has the ability to adjust his approach to optimize more scenarios than RE24 can differentiate between, and Ryan Howard will have a higher RE24 than WPA/LI going forward because if I don’t think he has the ability to adapt his approach to those situations beyond just producing the most runs possible given the base-out state.  I don’t know if that is true of those players or not, but that would be a difference between how those two rate players. 

As for WPA, I can’t have any idea of how that will compare to WPA/LI unless I think I can tell someone is clutch or is a choker, so I can’t tell whether it will be on the same side of WPA/LI as RE24 or on the opposite side.

I just realized my last point about not knowing whether a WPA/LI will be between WPA and RE24 for a player over all plays is not the same as what you were saying, where that would be true for each individual play.  For that to be the case, though, it would have to be impossible for the RE24 of one event to be above WPA/LI in a given PA but to be below it for a different event, which I don’t think is true (i.e. the walk-off GS vs. the walk-off walk), though I don’t have the actual data to verify that.

I looked back through the archives and I realized that I am basically asking the same questions that studes and David Gassko were asking in the “Unleveraging Win Probability” thread back in September 2007.  I’ll try and restate my specific questions.  If WPA/LI is unleveraging the WPA doesn’t that take out much/all? of the inning/score factor that makes the game state different from the baseout state?  It is clear that WPA has a large component of leverage in its value.  It is clear that dividing by LI reduces that component, possibly even eliminates it.  What is not clear to me is if the method for computing LI produces a factor that exactly analogous to the amount of game state leverage in WPA.  If it is not, then what is left over from dividing WPA by LI might not be a good measure of a batters game state ability but instead just be an artifact of a mathematical manipulation.  If it is, then I don’t see what the difference is between WPA/LI and RE24, which is what studes was asking in the earlier thread.

We had another thread on this a few months ago where I tried to figure out WPA/LI.  Tango did his best to explain it, but I never quite got there.  The problem is that most fans (I believe) think of leverage as having two components:  1) baserunners/outs and 2) score/inning.  If a metric includes 0, 1 or 2 of these, we can get our head around it.  But WPA/LI cannot be explained in these terms (whereas RE24 can). 

WPA/LI may be the “perfect” stat that Tango claims.  But I will immodestly say that if I (and apparently Peter) can’t understand it, after making a real effort, the probability of it ever gaining widespread acceptance is approximately zero.

I think I also had a thread called Situational Wins, and an accompanying article:
http://www.tangotiger.net/wpali.html

The entire methodology is there.

The best way to think about it is the bases-loaded, tie game, bottom of the 9th scenario.  You have to continually keep this situation in mind.

The other thing to always keep in mind is that each PA has an equal (overal) impact, regardless of the game or baseout state.

If you can do that, then you are almost all the way there.

It took it being explained to me quite a few different ways to get it.  I think something along these lines worked though.

WPA gives you credit for doing the right things in the right situation - a bunt might be a good move in certain situations with certain scores.  It weights any contribution that you have by how much it will help you win the game.  What it also does, is weight important plate appearances more than unimportant ones.

What WPA/LI does, is take away the second of the two.  It puts all pa’s on the same scale, but still gives you more credit for things that will help you win the game, based on the score/inning/ base-out state.

So it takes into account “situational hitting”, but gets rid of “clutch”.

How is this different than RE24?  The end of game is a good example.  Bunting could be another one.  A bunt in the first inning is most likely going to have a worse WPA/LI than a bunt in a late inning tie game with the same base-out state.  With RE24, they’ll be the same.

Remember WPA - WPA/LI = Clutch. 

Hope that helps.

What I don’t get, is how WPA/LI helps us in the Clemens situation.  It doesn’t take luck or fielding out of the equation.  Wouldn’t Dickson still come out on top?

"A bunt in the first inning is most likely going to have a worse WPA/LI than a bunt in a late inning tie game with the same base-out state.  With RE24, they’ll be the same. “

Ooooohhhh… I like that.  I LIKE that.

Same applies with the bases loaded walk by the way.