THE BOOK--Playing The Percentages In Baseball

Right on, MGL! Even Obama supporters get one right sometimes .

And how foolish was it for the Phils to let Pedro Martinez pitch into the eighth inning with that pitch count.

One advantage of the bunt is that the pitcher doesn’t have to run the bases. Do pitchers tend to do worse in innings right after they spent a lot of time on the basepaths? I would not be surprised. Does anyone know if this has been looked at?

Also sometimes pitchers get injured running the bases, which is a big deal. That said, I see no way these considerations could erase a .30 run gap.

What about a squeeze play?

I’m not one to think I know what I (or anyone else for that matter) am talking about when it comes to pitch counts, but yes, that seemed awfully irresponsible given that they have the division all but locked up and they need Pedro to be healthy for the post-season.  130 pitches?  Ouch!

From 2005-2008 PBP data.  There were 258 times that a pitcher was allowed to come to the plate with 1 out and men on 1st and 3d.  He bunted 135 times and averaged a Run Value Added -.405 runs per bunt.  The pitcher hit away 123 times and averaged an RVA of -.229 per PA.  So bunting lost .176 runs, not .30 runs.  Given that there is a selection bias as to which pitchers are allowed to hit away and which pitchers are asked to bunt, the amount of runs actually lost is probably quite a bit less.  But bunting is still probably an overall run loser.  However, since a pitcher bunting in that situation is only happening an average of 34 times a season or about two or three times per team, the loss of runs of .2 to .5 runs a year per team from this decision doesn’t seem to qualify as a major strategic blunder.

I don’t consider whether something is a major strategic blunder or not, Peter, based on how many times it occurs. I suppose it is a matter of perspective, but if you told a batter to strike out on purpose for no reason, but you only did that once per year, I suppose that would not be a major strategic blunder either, using your logic.

dcj, I never really understood the “I don’t want my pitcher running the bases” logic.  Your pitcher is only going to be on the bases around 20% of the time when hitting away, and even then, he may not even have to run once he is on the bases.  He is going to strike out 37% of the time.  Would you rather have that or have him bunt the ball and have to run to first almost every time, risking injury?  And how often does a pitcher end up on first after a bunt, either because the runner on first is forced or the pitcher is safe on an error or hit?  I don’t think the “running the bases” argument holds any water.  He is probably going to be on first base around 10% after a bunt and 20% of the time when hitting away.  And 80% of the time, he is going to have to run to first after a bunt, even if he is going to be out.

By the way, I vaguely recall someone doing s study on the next inning after a pitcher is on the bases.  I don’t remember the results, but I would bet a lot of money that it had no effect.  Again, how often does the pitcher actually have to do anything on the bases?  Not very.

The running the bases argument doesn’t make sense.

All a pitcher has to do is run like he’s an Upton brother.  This way, he will not get spent at all.  And he MIGHT be able to move up a base now and then.

Othewise, by intentionally not trying to get to first base, he’ll never have a chance to score.

very nice post...I never knew that this mistake was so -EV.

Didn’t see the game and you don’t mention what inning this was, but my answer regardless is “Pinch Hit”.

This is Tim Redding, not a good pitcher, and your bullpen is fully staffed because we’re into September roster expansion.

Rally - It was the top of the 2nd. 

If you are going to have the pitcher bunting in this situation the man on third should be running for home on non pop up contact.  An out at home leaving men on 1st and 2nd with 2 outs is only .16 runs worse than a successful sacrifice that leaves runners on 2nd and third.  If it is a good bunt and the runner on third scores the gain averages .88 runs.  The runner on third scores on around 23% of pitcher bunts in this situation.

The runner was Brian Schneider, so there was no chance for a squeeze.  What’s even funnier is that Luis Castillo tried to bunt for a hit as the next batter, and Schneider didn’t even try to run on the hit.

"The pitcher hit away 123 times and averaged an RVA of -.229 per PA.”

As a point of reference, in a random PA, a pitcher is roughly -.14 runs per PA.

For the 1b, 3b, 1 out situation, we can try to use this:
http://www.insidethebook.com/ee/index.php/site/comments/leverage_index_by_base_out_states/

The LI is 1.77.  So, the -.14 balloons to -.25 runs per PA.  Peter reporting -.23 runs sounds reasonable.

(What I did was a sanity check.  You’d have to work out the expectations of various hitting events like MGL did.)

So, I am NOT surprised that Peter and MGL are in agreement on this point.

***

The standard RE:
http://www.tangotiger.net/RE9902.html

for that situation is 1.243. That’s with a 5 RPG run environment.  Also with a 5 RPG run environment, men on 2B, 3B, and 2 outs is .634 runs.  So, the difference between the two is .61 runs.

As we noted above, Peter is putting the RE of the pitcher swinging away at -.23 and I’m saying the sanity check is at -.25.  So, 1.243 minus .243 or so is 1.000.  MGL is going further lower still to .909, which is the better number, because of the run environment, and he looked at it piece-meal.

However, Peter is saying that with the bunt, the run change is -.405 runs.  That must mean that there were a few occasions where the pitcher ended up on base on an error.  That is, if Peter were to do a breakdown of his numbers, yes, we would see the “-.60” or so runs that is the result of ending up with 2b, 3b, 2 outs, but that this frequency would be say 70%.  And then, the other 30% would be broken up to “+.35 runs” or so (no outs) and “-.90 runs” or so (at least one out, and/or no bases gained).

And the resulting frequency is such that it would come out to -.40 runs or so.

Peter… can you show us that breakdown?

Zack - I knew Schneider was the runner.  It makes very little difference to the success of the squeeze.  It is not a force play.  Whether it succeeds or not is mainly dependent on the placement of the bunt, the fielding of the ball and the accuracy of the throw, and whether the catcher can apply the tag and hold onto the ball.  There is very little downside to his attempting to score.

Tango:  If I’ve followed this correctly, I think the bigger factor MGL missed was the runner on 3B scoring 23% of the time.  Presumably in some of these a throw was made to home, so the pitcher is also safe.  I don’t think MGL’s .60 post-bunt estimate incorporated that.

Right, that’s why I wanted Peter to break it down.

It seems to me that MGL was showing the “100%” frequency as a proxy when he said this:

If the pitcher bunts successfully and makes out at first, which is a reasonable proxy for a bunt attempt by a pitcher (actually, I think a bunt attempt by a pitcher is worse than that, but no matter), you have 2 and 3 and 2 outs.

And I think he jumped the gun on that presumption.  Or at least didn’t offer the evidence for it.  And Peter’s evidence, albeit with a limited sample, shows that it’s not correct.

Which is why I proposed that we see what actually happens, like the following:

Freq deltaRE BuntEvent
70% -.25 batter out, 2b to 3B, 3b stays (standard)
10% -.60 worse than standard
20% +.30 better than standard

(Above for illustration only.)

Or whatever it is.  And then we can debate that perhaps the observed frequency of bunts of 70% standard outcome should really be 75% in “true talent”.  That 20% maybe should be 10%, and that 10% should be 15%.  Or whatever.

But, the 100% proxy is not a good starting point.

Tango:

BaseOuts----Runs_on_Play-----N-----
----0-----------0------------6-----
---32-----------1------------1-----
--202-----------1-----------22-----
--231-----------1------------1-----
--232-----------0-----------81-----
-1031-----------1------------3-----
-1032-----------0-----------14-----
-1201-----------1------------3-----
-1202-----------0------------3-----
-1231-----------0------------1-----

Average RE following the bunt attempt is .80 based on Peter’s data.

That would suggest the outcomes on hit away PAs (RE of .975) were also a bit better than MGL estimated.  I’d guess that pitchers hit a bit better than usual with runner on 1B and one out, since the pitcher really doesn’t want to issue a BB (and the pitcher-as-hitter knows that).

You gotta be careful using retrosheet data to figure the average result (RE+runs scored) after a bunt attempt.  You may get all the positive results (like a bunt hit, error or reached on a FC), but not some of the negative ones, like a K on a bunt foul or a called or missed strike 3, or even when the pitcher switches to swinging away with 2 strikes (which is rare).  You also are not going to get the walks on a bunt attempt.

The squeeze may in fact be the best play, depending on the pitcher (how good a batter he is and how good a bunter he is).  It is almost no doubt better than the sac bunt.

You also have to distinguish between the suicide and safety squeeze.  For the suicide squeeze, the speed of the runner is not an issue of course.  The runner is either safe on a fair bunt or out on a missed bunt.  With a slow runner, occasionally a fast come-backer to the pitcher or a bunt right at home plate might get the runner at home.

With a safety squeeze, where the runner breaks as soon as the ball is bunted on the ground, but not right at the pitcher, the speed of the runner is definitely an issue.  But, as Peter says, leaving runners on 1 and 2 and 2 outs is not that much worse than 2 and 3 with 2 outs.

Rally, there is no doubt that pinch hitting for Redding, who is a replacement level pitcher anyway, is 100% the correct move there, but no manager is going to do that in the 2nd inning.  And of course the Mets are not playing for anything right now, other than not pissing off their fans any more than they have, although this year, injuries have been by far and away their major problem.

Peter, thanks.

Ok say taking his data, and we have this as the standard play:
--232-----------0-----------81-----

This as the “better than standard”:
---32-----------1------------1-----
--202-----------1-----------22-----
--231-----------1------------1-----
-1031-----------1------------3-----
-1201-----------1------------3-----
-1231-----------0------------1-----

And this as the “worse than standard”:
-1032-----------0-----------14-----
-1202-----------0------------3-----
----0-----------0------------6----- (I’
m going to figure this means “double play”, since I would expect to see a “3” at the end, but you are showing it as the “start of next inning”?)

So, that’s 81 standard outcomes, 31 better-than-standard, and 23 worse-than-standard, or a frequency of 60%, 23%, 17%, respectively.

I’d have to apply the RE to each of those outcomes to see what to expect, but I’m on my way to lunch, so I’ll make some quick head calculations and guesstimates and say:

.634 standard outcome
1.3 better than standard
0.5 worse than standard

If we focus on the 54 times a non-standard outcomes happened (31 plus, and 23 minus) that gives us an RE of .96.

Clearly, MGL thinks that that is not what we expect, since he used the .634 as the proxy for all outcomes for this situations.

He’s also thinking there’s alot of squeeze plays in there that shouldn’t count.  Perhaps he’s thinking that 6 are successful squeezes and 13 unsuccessful?  (No idea, just throwing numbers out there.) That leaves us with 25 better-than-standard outcomes in the regular bunt, and 10 worse-than-standard.

The RE of those 25 plays is .79.

So, it really makes a difference in isolating the squeezes from the non-squeezes here (and suicide and not).

Great data Peter!

MGL - 30 of the 123 PAs that I had classified as “hit away” had some bunt attempt in the series of pitches.  Two ended with a missed bunt attempt K.  No fouled third strike bunts.  18 regular strikeouts,one of them a K+CS, 1 walk.  9 hit balls,8 for outs, 6 for double plays, 2 resulting in scores, one of them a force of the runner at 2B.  1 single scoring the man on third and leaving men on 1st and 2B. 

18 of the 30 plays sarted with a bunt attempt of some sort for the first strike.  7 with called first strikes which we have no idea whether they were on bunt attempts or not. 3 on foul swinging strikes and 3 on swing and miss.

I don’t like the idea of classifying a PA as “hit away” if it had a bunt attempt in it.  If, for example, you had a position player, and you classify all PA’s that ENDED in a swing as a “hit away” you would have a disproportional number of 2 strike counts in your “hit away” bucket, and the hit away bucket will underestimate the number of runs scored when hitting away.  Not so much with pitchers of course because they tend to keep bunting with 2 strikes.

There is of course a gray area when classifying a PA as hit away or bunt attempt, and sometimes the batter changes his intent in mid-stream and you have a “hybrid” PA.  As long as you try to make sure that you don’t have too much of a bias in either bucket, it is OK.  There are potential problems though and you have to be careful and thoughtful when creating the buckets.

I realize that you are not doing a careful, rigorous analysis.  And of course, there are huge sample size issues if you are only tracking 100 or so PA’s.

When I was doing my bunt research, I had to, for example, estimate how often I thought a walk would occur when the batter definitely intended to bunt but there was no bunt attempt recorded in the PA (such as when the first 4 pitches were balls or there were 4 balls and 1 or 2 called strikes). Of course, I wish that the retrosheet or even the STATS or BIS data would include “squared to bunt and took the pitch,” but, alas, they don’t, I don’t think.

Anyway, iff I didn’t do that (estimate the number of walks with no bunt in the PA and put those walks in the “bunt attempt” bucket) I would have walks under-represented in my bunt attempt bucket and the resulting number of runs scored would be too low.

So I am just saying that in trying to figure the actual number of runs scored after a bunt attempt or hit away, at least from the retrosheet data, you have to very careful and you have to make some assumptions and estimations when classifying a PA as a “bunt attempt” or a “hit away.” It is easy to underestimate or overestimate the number of runs scored because of mis-classifying some of the PA’s causing the buckets to be biased (if it is a random or non-biased mis-classification, it is not that big of a deal).

Fredi Gonzalez says I see your stupidity and raise you. In the second inning tonight, down 4-1, he has his pitcher Nolasco try to bunt with 1st and 3rd and one out. Even worse, aside from the decision to bunt given the score, is that he let Nolasco attempt to bunt with two strikes. Baffling.

If you are going to bunt, it is generally correct for a pitcher to bunt with 2 strikes.  But, given that the sac bunt with runners on 1 and 3 is grossly incorrect in the first place, of course bunting with 2 strikes is not correct either.

But, given that he thinks that the bunt is correct in that situation, which most managers do, why would you think that he wouldn’t bunt with 2 strikes?  IOW, why is that a worse mistake?

And down 4-1 in the 3rd inning, if you are fighting for a wild card birth, the only correct move is to pinch hit.  Nolasco is probably better than a league-average starter, but he is no ace.  I don’t think that a pitcher would have any right to be pissed if gets yanked for a PH after giving up 4 runs in 2 innings.

Fredi is the one with the bizarre lineups, right?  Stupid is as stupid does!