THE BOOK--Playing The Percentages In Baseball

I’ve always felt the same way about Carlos Pena, and couldn’t find a nice analogy.  With Pena there will be less raw number of game changing hits, but each time he does impact the game will be a much more meaningful play.  Fans seem to find it easier to remember the vast number of occurrences where he couldn’t come through while ignoring that when he does his impact is much more game changing.

Tango, I think you’ve missed the really interesting discovery in your clutch exercise.  Recall that in another tread—a dissection of WPA/LI I think—we discovered that not all wOBAs are in fact equal in high leverage situations.  What you really want then is a high BA, while BBs and HRs are relatively less valuable.  And that is exactly the kind of hitter fans are saying they want to hit in the clutch.  So the fans’ intuition uncovered something that saber analysis had missed.  It’s true that the fans only found a small amount of “clutch” talent in the sense you were looking for, i.e. hitters’ performing above their own mean in clutch situations (although the fans found more than I expected).  But their selections are in fact exactly the kind of hitter you want up in those PAs.  If you matched the fans’ choices with random players of equal wOBA, I bet the fan picks would deliver a bigger WPA over time.  The fans were on to something very important, even if it wasn’t the kind of clutch performance you were looking for.

Guy: I agree that it is a fascinating point.  For example, Tony Gwynn turns out looking great in the clutch while Mike Schmidt doesn’t.

I think Nate and Keith made a point of this in their book a few years back as well.

It makes sense guys like Adam Dunn (and Carlos Pena) perform worse in “clutch” situations. Hitters reliant on the three true outcomes (and vulnerable to splits) should perform disproportionately worse against the league’s better pitchers compared to those who make a lot of contact.

I think the biggest problem with clutch is the assumption that it has the same shelf life as other skills.  Clutch skills probably fade the fastest.

I think clutch skills may be visible for only a few years, mainly in a players peak. 

Also, due to SSS, it’s foolish to look at Y2Y comparisons.  You probably have to look at several years in a players peak to get a large enough sample.  Start with Papis 2003-2006.

Not every player is a clutch hitter.  In fact, for every clutch hitter there are probably as many, if not more anti-clutch guys you dread seeing up at the plate in a clutch situation.  And of course others who don’t hit any better or worse in the clutch.

pft: read Baseball Between The Numbers, since they do exactly that.  Maybe you can see it on Amazon’s Look Inside.  Do a search for Mark Grace.

” Hitters reliant on the three true outcomes (and vulnerable to splits) should perform disproportionately worse against the league’s better pitchers compared to those who make a lot of contact.”

Dalton, it has nothing to do with that. And I don’t know of any evidence that 3TO outcome hitter do worse than they “should” against good pitchers.

The reason that some hitters have lesser or greater impact in clutch situations (as measured by, say, WPA) is not because of a “clutch skill set” (performing better or worse than their usual stats in clutch situations). It is because certain hitting profiles (e.g., more contact, fewer walks) are more valuable in clutch situations. 

In other words, let’s say that we have 2 players and one has more K than the other, but all their other stats have the same rates. Overall they are equally as valuable, using our standard, overall lwts values.  But, those lwts values change depending on the situation. In clutch situations, if the K is a lot worse than the batted ball out, whereas overall, they are around the same, then the low K guys will be more valuable in the clutch when we use something other than standard, overall, lwt values, to evaluate them.

Which brings up the $64,000 question:

Is it correct to use overall lwts values to value players?  If there are different lwt values for high and low leverage situations, don’t we have to use some kind of weighted (with more weight given to the high leverage situations) lwt values for the offensive components?  Wouldn’t that make things like walks have a lower lwts value and K’s have a higher negative value, etc.?

Have we been valuing players with regular lwts wrong all along?

The Book has win values.  I forget which Table, but it’s in Chapter 1.  Table 5 or 6 or something.  You don’t really see much different than random, except for things like steals, bunts, and IBB.

So then Guy is not correct?  Wasn’t that his point, that players with certain hitting profiles indeed do create more value (WPA) in clutch situations because the lwt values of the various events change?

And then why did you respond with this:

“Guy: I agree that it is a fascinating point.  For example, Tony Gwynn turns out looking great in the clutch while Mike Schmidt doesn’t.”

MGL - “The reason that some hitters have lesser or greater impact in clutch situations (as measured by, say, WPA) is not because of a “clutch skill set” (performing better or worse than their usual stats in clutch situations). It is because certain hitting profiles (e.g., more contact, fewer walks) are more valuable in clutch situations.”

That was literally the exact point I was trying to make. Maybe I did a poor job of that.

Put differently, I’d much rather an Adam Dunn (in his prime) over a full season than Placido Polanco, but if facing Mariano Rivera for a single at-bat, I’d prefer the latter.

And I absolutely agree we probably haven’t been valuing skill sets optimally because of this. Not that I have any kind of solution for it though.

Table 11 has the win value (and run value) for all the events.

And you are right MGL that the relative value of the win and run values remain pretty constant.  So, I guess what I said doesn’t make sense, but I’ll have to think about it.

We do see that HBP are not issued nearly as much when it can cost the win, so that even though the run value is higher than the BB, it’s win value is almost the same.

BK, PB, WP, and especially the SB have more win value than expected.

Defensive indifference have almost no win value, as expected.

I think we can reconcile these findings.  In high-LI PAs, singles are very highly leverage (on average).  The value of a single then is quite high, relative to its average value.  While the value of HR is also magnified, it’s not magnified as much.  (Tango:  could you estimate average leverage for each offensive event when LI is, say, 3.0?  That would be interesting.) However, in low-leverage PAs we will see this pattern reversed.  To take a simple example, at bases empty and 2 outs a single is worth about .13 runs, or about one-fourth its average value.  A HR, however, retains about 70% of its average value then.  So this will net out at close to average for each event, except for events which can really be timed based on the situation (sac, IBB, SB). 

*

While I do think high-BA hitters are more valuable in the clutch, it doesn’t look like this proved true in Tango’s experiment.  His hitters produced about 10 more wins than the fan choices.  I’m not sure, but that seems like even more of an advantage than we’d expect given the difference in OBP/SLG, after accounting for LI of about 3.0.

What we clearly have to do is to weight the lwts values in each situation by the proportion of time they occur.  Those are the “real” lwt values ((in terms of valuing a player) that should be used in wOBA, lwts, RC, etc.  Anyone up to that?

MGL: that’s what I did in The Book, Table 11.

This all touches on the value of situational hitting.  Logic would dictate that hitters alter their approach in “clutch” situations (e.g. runners on 2nd and 3rd, one out), but I’m skeptical about whether most hitters can increase run expectancy in this way.  In other words, I want Adam Dunn doing his Adam Dunn thing, and not expanding his strike zone trying to put the ball in play.  If this results in a low-value walk, so be it.  Are there studies that have looked at this issue?