THE BOOK--Playing The Percentages In Baseball

RE24/boLI and WPA/LI still don’t account for the value lost from Pujols batting vs. the next guy in the lineup.  The problem I have is if we did try to take this into account, we’d end up penalizing Pujols for being a better hitter than the guy behind him in the lineup which seems unfair.

Why doesn’t someone with the data look at the WPA/LI for the unintentional walks of various players and see what the avg was? I’d be curious to see what the typical UIBB was worth by Barry Bonds as opposed to another hitter who walked quite a bit, but wasn’t as feared, say R Henderson.

I’m not sure I understand the post by Tango, but I think I have a pretty good handle on the BB/IBB situation:

The IBB’s are easy.  If a player is IBB’d, you have no idea what he would have done had he been pitched to, regardless of who it is.  So you simply assign to that PA the average value of that player’s PA (for that season, probably).  There is simply no other way to do it.  The fact that the better player is IBB’d more often is already accounted for with that method.

The BB is a little tricky.  First of all, BB’s (NIBB) tend to be given in situations where a base runner is not that bad (for the offense).  That is why the lwts value of the BB is lower than that of the HP.  It is probably even a little lower, relative to the HP, if you look at win value rather than run value.  So, for most players, we can just use the .31 run value for all NIBB’s, and that is good enough.

But, for the better players, especially those that are deemed as “dangerous” by other players and managers, the average value of their BB is probably less than .31?  How much less?  I don’t know.  But we can certainly find out can’t we?  If it is .30 or even .29, then I wouldn’t worry about it. If it is .26, then we are talking about a 3-5 run difference for a player who gets 50 or 100 NIBB is a season, still not a huge deal.

So we can do one of two things:  Take the average actual value of all of a player’s NIBB and use that for his lwts value (for the NIBB), or we can assign different values for a NIBB, depending on the type of batter that we’re talking about or depending on the number of IBB’s that that batter gets.  For example, if a batter gets 20 IBB in a season, many of his NIBB’s might be semi-intentional, and overall each NIBB might be worth .25 or something like that. For a batter who gets 10 IBB, it might be .28, and for a batter who gets a couple of IBB or none, it might be .32.

No matter what you do, I just don’t see this as a big problem.  The main thing you want to be sure and do is to assign a value to the IBB correctly, which should be somewhere around a player’s average PA. It has to be close to that, since we have don’t know what the batter would have done had he been pitched to, and we have to assume that the IBB was close to a break even proposition.

Even if overall the IBB’s given too frequently, such that the average value of an IBB is higher than a generic PA for a player like Cabrera (who gets so many IBB’s - probably way too many), we don’t want to give him that much credit for the IBB, since generally we use lwts to estimate “context-neutral” value of a player and his true talent.  I am not sure we want to include “bad IBB’s” as part of a player’s talent…

I agree with the “known IBB” that we give then the average win value per PA of that batter.

For the rest of BB, we should be splitting by the 24 base out states.  We should look at the good hitters v bad hitters with 1b open and 2 outs and see if the good hitters are getting disproportionate number of walks in those situations (relative to the rest of their walks), thereby signalling some hidden IBB.

Lots of different ways to think about these walks.

"For the rest of BB, we should be splitting by the 24 base out states.  We should look at the good hitters v bad hitters with 1b open and 2 outs and see if the good hitters are getting disproportionate number of walks in those situations (relative to the rest of their walks), thereby signalling some hidden IBB.”

Right, we should basically compute the average lwts value of the walk for different kinds of hitters.  We don’t want to use the actual value of the BB for each player, based on the 24 base/out states, because then we are getting away from the idea of lwts, which is to remove context in order to smooth out the random fluctuation of when a player happens to get most of his singles, or walks, or what have you…

Right, and we can extend this to strikeout too.  Strikeouts with men on 3B and less than 2 outs are incredibly costly.  If there is a class of players that has a lower or higher K rate in those situations, then we should adjust accordingly their values.

It’s easy to see with walks, it’s about either the quality of player, or the “power” of the player, or the “high-contact” of the player, and the differences with 1 or 2 outs.

Not necessarily so straightforward with the 3B/K situation.  It might be more of the FB tendency and K tendency of the batter.

I don’t see how K’s are the same thing.  In one case, you have the pitchers throwing more balls to the better hitters when the walk is not that costly.  In the case of the K with a runner on third (and less than 2), pitchers are more or less hoping to K everyone.

Now, it is likely that some hitters are better than others at being situational in all of their at bats, including with runner on 3rd. Certainly some batters (and correctly so) try and make sure that they don’t K with a runner on 3rd and less than 2 (at the expense of something like power, probably), and other batters don’t seem to care (they still swing from the heels or take too many close pitches and strikes).

But, that is a different story altogether and regular lwts is not designed to capture that kind of talent, if that kind of talent even exists to more than a de minimus degree.  And we don’t know how much of that in any sample would be talent and how much would be random fluctuation without first determining the spread of talent in the population with respect to “situational hitting.”

Speaking of that, WPA minus regular lwts should give us a player’s exact “situational hitting” bonus right?  Has anyone ever looked at the year-to-year (or intra-class) correlation for that?  If there is some significant correlation (and I am not sure that there is), then rather than judging the future value of a player by their lwts projections, shouldn’t we be using a WPA projection regressed the appropriate amount?

"WPA minus regular lwts should give us a player’s exact “situational hitting” bonus right?”

Or, perhaps rather than WPA, maybe the lwts value of an event given the inning, score, base runners, and outs?  That is slightly different I think and I am sure how you would calculate it.  For example, using WPA if a player comes up with no one on base and his team is down by 5 runs in the 9th inning and 2 outs, the WPA value of a walk or a HR is pretty small and a player would get heavily “penalized” if he gets a single and we compute his “situational” bonus as WPA-regular lwts, since his WPA (in runs) might be .03 and regular lwts is .47, so he gets penalized by .44 runs.

However, he should get the same credit for a home run as for a walk since they are both worth almost the same thing in that situation, and the batter should know that. He should not be trying to hit a HR - he should just be trying to get on base....

Speaking of that, WPA minus regular lwts should give us a player’s exact “situational hitting” bonus right? 

Right, what you want is WPA/LI minus regular lwts (converted to wins).  You need to do the /LI so that you keep the impact as if it was LI=1, but you are allowing the game state to dictate the value of the HR or K etc *relative to each other*.

This would be called “situational hitting” value.  I think we looked at this once.

However, he should get the same credit for a home run as for a walk since they are both worth almost the same thing in that situation, and the batter should know that. He should not be trying to hit a HR - he should just be trying to get on base....

You are understanding WPA/LI perfectly.

In a bottom 9th, tied game, bases loaded, 2 out scenario, the walk and HR are worth the same, say +.34 wins.  And an out is worth say -.17 wins.

We scale those numbers to what we are used to, and that is that an out is worth on average -.027 wins.  That is, the LI was .17/.027 = 6.3.

So, the WPA/LI of any positive event (in that game state) is +.054 wins, and any out is -.027 wins.

Regular LWTS would say that a walk is +.030 wins and a HR is +.130 wins.  Now, in terms of “situational walking”, we see that the batter was smart, in that he got +.024 more wins than he normally gets.  And for “situational homering”, well, he didn’t so so hot, as his HR here was not worth as much as a regular HR.  (Consider it a cheap HR, since it may as well have been a walk.)

Situational Hitting wins.  Like it?

Yes, I like it. I want to know if it is a skill though (and to what extent) otherwise we are just measuring noise and implying that it is a skill, which I don’t like.

So the first thing I want to do is to compile WPA/LI minus lwts for every player for the last several years and then do a y-t-y (or intra-class) correlation. If we get something more than say, a .1 R, then I really like it.

We should do the same thing for pitchers.  If that is a skill for pitchers, it could certainly (partially) explain the difference between some pitcher’s ERA and ERC or even ERA and FIP.  Pitchers should definitely be pitching in such a way that they give up solo HR, hits with large leads, walks with runners on base and bases open, etc.  Using this metric would tell us (with noise of course) who the smart pitchers are and who are not.

No one has done this?

I think we have.  David Appelman may have done this.

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

Thanks Eli.  I just re-read that thread. So has anyone replicated the work and figured the correlations?  It is not that hard to do.  It has been on my long (mental) list of “things to do” for a long time…

I haven’t done it, but I’ll take a guess that it’ll be r=.10 to .20, at PA=500.

I took the top 200 players in PA from 1993-2009, and, maybe there’s a non-zero aspect to it that isn’t adjusted, but…

The average BtnWins on B-R.com (which I think is the Linear Weights as wins) is +9.2 wins, while the average WPA/LI is +12.4 wins.  So, it’s possible that the average player simply optimizes himself as a general rule.  Again, all depends how the stat itself has been zeroed out, etc.

I’m presuming that the WPA sums to zero at the league level, and so, WPA/LI should be zero all other things equal.

So, that’s one indicator there.

Aside from that, what I did was converted each LW wins and WPA/LI into wOBA, and figured how many SD they were apart.  For example, Tony Womack’s WPA/LI was 3 SD from the mean relative to his LW wins.  Posada was the worst at -2 SD from the mean.

I figured the z-score for each player and took the SD.  That SD was 0.95.  If we got exactly 1, it means it’s random.  So, it seems to be pretty much random, and my guess of r=.10 to .20 is way too high.

Again though, it all depends on how Sean calculates the two metrics.

#2 is Larry Walker, and he seems like the kind of guy who is a good situational hitter.  Then again, #3 and #8 hitters are also Rox hitters (Castilla, Helton), so maybe we’ve got park issues to contend with.  Juan Pierre is 10th, Bichette is 12th, Galaragga is 11th.

Then again, Erstad, Polanco, Counsell are all in the top 30 (out of 200), the kind of guys who you’d see there.

Anyway, we’re talking about half a win per season.

The more important finding is that all players improved as a group, showing that perhaps hitters can leverage their situation more than pitchers.

More interesting would be a breakdown by the 24 baseout states.  Maybe it’s only a few of the states where the leverage is happening.

It would not be surprising if the batters out leveraged the pitchers, since they get “last licks” in the PA.

On the other hand, I would not be surprised if it is due to WPA/LI not being leveraged.

The z-score of .95 surprises me.  I would have thought that it would be more of a skill.  Even if all players “tried” equally hard to be situational hitters, it seems like some players are much more likely to be successful given their hitting style.  For example, some RH batters simply cannot hit a ball to the opposite field (or they don’t try) with a runner on second and no outs, whereas others are really adept at it.

Interesting stuff, though…