THE BOOK--Playing The Percentages In Baseball

Why am I not surprised that Scrappy Doo was second in most H&R attempted AND below-average in success rate?

One bias that he noted early on, but that it doesn’t look like he adjusted for, is that a hit and run is in neutral or moderate hitter’s counts.  So, we expect more runs on that basis.  This will probably account for .03 or .04 runs of bias.

Tango - I read the below quote from Mike’s article that he was comparing the runs from the same counts that he had defined as hit and run situations.  Doesn’t that correct for the potential bias that you noted in the above quote? 

Teams that attempted the hit-and-run play scored 0.11 runs on the play and 0.69 runs in the remainder of the inning on average, compared to 0.17 runs scored on the play and 0.70 runs in the remainder of the inning by teams that did not attempt the hit-and-run play in hit-and-run situations.

It wasn’t clear to me that he controlled for the count, or at least weighted by the frequency.  There was alot to digest, and maybe I didn’t read it as well as I should have.

I don’t have a hit-and-run sign, and I believe it’s the worst play in baseball…

This quote from Weaver on Strategy, probably my favorite sabermetric book of all time, should be put in the context of its following paragraph.

“Now I do have something I call the run-and-hit. That is far different from the standard hit-and-run.  With this play, the runner is going full-speed to second.  He wants to steal the base and isn’t counting on the hitter to protect him by making contact.  The batter swings at the pitch only if he likes it; he can take it if its out of the strike zone...”

With the number of Weaver’s former coaches and players who have managed in the majors and the influence that he has had on how baseball is played I would be very surprised if the traditional hit-and-run where the batter has to swing and the runner has to run outnumbers the Weaver defined run-and-hit.  This makes separating out and evaluating the actual hit-and-run more difficult.  However, I like that Mike has taken a very conservative approach in his definition and still has found a significant positive value to the strategy.  A more inclusive definition that includes more run-and-hits will likely show an even greater positive value.

There is really no reason to have any other strategy than letting the runner decide when he has the best opportunity to steal and the batter decide when he has the best pitch to hit.  The key is to educate all players about the relative value of attempting to steal at various baseout situations, lineup positions, counts and game scores.

By the way, when I write up these posts, I write them as I read them.

So, I as I read Mike’s article, I put in stuff like: “Mike didn’t consider that...” I had a few sentences written.

Then I continue reading it, and Mike then DID talk about that.  So, I had to delete what I just wrote.

I did it a second time, and the same thing happened.  Again, I deleted what I wrote.

That’s when I decided I’m in for a good ride here, and decided to read the whole thing straight through, and then double-back with my comments after.

Peter/2, correct.  I compared batter quality in the same set of five ball-strike counts.

I did not, however, calculate the effect of the hit-and-run plays disproportionately happening in some of those five counts more than others (i.e., mostly at 2-1).

@4: The use of the run-and-hit is probably the biggest reason for the sharp increase in league stolen base percentage over the last two decades.

The best benchmark I could find was that the runner was noted as going on about 90 percent of pitches resulting in a stolen base or caught stealing. So we are probably missing at least 10 percent of the hit-and-run plays.

A number of SB and CS occur on pickoff throws where the runner was not actually running on the pitch.  Retrosheet no longer gives an explanation of what “*” means in the pitch sequence or even what “>“ means.  If I am remembering correctly that “*” indicates a pitch in the dirt, then some of the other SBs where there is no “>“ may be correct also.  This may mean that the stringer is recording running on the pitch with a higher accuracy than the 90% that Mike guesses in the article.

I just started reading the article, which is a great approach thus far.  I wish he had isolated more things in order to identify possible H&R situations, though.

I think he should have looked more at the identity of the batter and runner.  Batters who have high K rates are not likely to be part of a H&R. Same thing with premium batters, like Pujols, ARod, Votto, Fielder, etc.  Those batters rarely are asked to hit and run by the manager for obvious reasons.  Also, true hit and runs are rarely executed with a good base stealer on first.

So I think Mike could have and should filtered out a lot more situations.  I think he should have at least filtered by lineup order, which he chose not to do.

That being the case, I think that a good proportion of his hit and runs are straight steals where the batters swung, which means that the batter likely got a good pitch to hit and the runner was fast.  On those occasions, it is not surprising that there would be a large positive result - fast runner going on the pitch and a batter in a hitters count swinging at a pitch he obviously likes.

The downside of the hit and run is two-fold:  One, the batter is forced to swing at bad and sometimes terrible pitches, and two, the runner is often not fast and does not get a great jump.

So unless your likely hit and run situations are indeed mostly (90+%) true hit and run situations, you are not getting a fair and accurate representation of the offensive value of the hit and run.  And even if 90% of your situations are true hit and runs, the other 10% are very favorable offensive situations (again, fast runner, good jump, batter likes the pitch enough to swing in a hitters count), such that the overall results will be inflated (toward a more positive result) by quite a bit.

Anyway, let me read more of the article later and I’ll comment again…

Pujols has been involved in his share of hit&run (or run&hit) plays. 

Probably makes more sense for him than some of the other top hitters, with his low strikeout rate and high GIDP rate.

Peter/8, it’s about 88 percent in the raw data, but getting a little better over time, particularly in 2011.  I looked at the video for some of the SB/CS plays where no runner was marked as going in 2011, and a sizable number of them were delayed steals or pickoffs or things like that where the runner wasn’t actually going with the pitch, but many of them were also just misses by the stringer.

So I said about 90 percent, but there’s no good way to know for sure without reviewing a lot of video.

MGL/9, I didn’t filter the situations by the lineup order, but I did adjust for the quality of batter and frequency of the hit-and-run attempt by the lineup order.

In the situations I did use, the stolen base success rate dropped from around 78 percent in non-hit-and-run situations to around 58 percent in hit-and-run situations (numbers off the top of my head, but they should be close).  I wish I had an independent way to measure the SB success rate in true hit-and-run situations.  If I did, I could estimate how many straight steals were polluting my data.  As it is, I can’t think of a reliable way to estimate that other than guessing.

the other 10% are very favorable offensive situations (again, fast runner, good jump, batter likes the pitch enough to swing in a hitters count), such that the overall results will be inflated (toward a more positive result) by quite a bit.

I’m not sure why you think those situations would be biased toward good jumps, but I agree with your other two observations.  How much does a faster-than-average runner and a pitch good enough to entice a swing bias the results?  Remember that I adjusted already for it being a hitter’s count.

I actually looked into the speed-of-runner issue a little bit but didn’t have time to finish that analysis or space to include it.  However, nothing from that was suggesting to me that there was a major bias from the speed of the runner (as judged by his stolen base rate and stolen base success rate).  It’s probably worth looking at further.

I should also look at the run outcome difference between swinging and taking in these counts when the runner is not going.  That would give me an idea of the amount of bias from the batter getting a good pitch to hit.  That’s one that I didn’t think of controlling for and should be investigated.

Mike #11 - After I wrote my post #8 I thought of including the baseout situation and the Event_Text info to explain the situations further.  Many of the situations appear to be where the runner breaks early and the pitcher steps off the rubber and throws to the next base, missed bunts where the runner is caught too far of the bag but is not actually trying to run to the next base, delayed steals where the runner goes on the throw back to the pitcher, the aforementioned pickoffs, and whatever is indicated by the “*”.  Video confirmation is best but a lot of plays could be filtered out in advance.  Eliminating pickoffs, stepoffs, and missed bunts brings the 2005-2011 data up to about 94.3% correct which I think is much closer to the current state of stringer accuracy.  I can believe that stringers missed a lot more during the first couple of years.

Peter:

http://www.retrosheet.org/eventfile.htm

+ following pickoff throw by the catcher
* indicates the following pitch was blocked by the catcher
.  marker for play not involving the batter
1 pickoff throw to first
2 pickoff throw to second
3 pickoff throw to third
> Indicates a runner going on the pitch

B ball
C called strike
F foul
H hit batter
I intentional ball
K strike (unknown type)
L foul bunt
M missed bunt attempt
N no pitch (on balks and interference calls)
O foul tip on bunt
P pitchout
Q swinging on pitchout
R foul ball on pitchout
S swinging strike
T foul tip
U unknown or missed pitch
V called ball because pitcher went to his mouth
X ball put into play by batter
Y ball put into play on pitchout

Thanks Tom. That’s what I have used before, but I couldn’t find the link to it when I went to the Retrosheet site today.

"I’m not sure why you think those situations would be biased toward good jumps, but I agree with your other two observations.”

I just meant that when there is a straight steal, there is generally a better jump from first as opposed to a hit and run where many runners get a lazy jump assuming that the batter will make contact. It doesn’t really matter.  You can ignore that part of my post.

"The batters who attempted the hit-and-run play ended that plate appearance with an average batting line of .294/.323/.403, as compared to the batting line of .267/.324/.419 for the batters who did not attempt the hit-and-run play.

For the hit and run, you have a BA of .294 and an OBA of .323.  That seems like a lot of walks.  The only way a walk can occur is when the batter swings and misses and subsequently walks, right?  Again, that seems like a lot of walks.

Mike, I think you have to eliminate almost all of the base stealers from your hit and run opportunities and only look at the non-base stealers. Here is why:

Let’s say that we don’t know the results of your analysis and we transport ourselves into a game, and there is a hit and run opportunity (runner on first only, 0 outs and a hit and run count). Let’s even say that there is a below average, low strikeout, low power hitter at the plate.

Now, we ask ourselves if the hit and run is a positive play or a negative one.  Now, you come along and say, “I found out that it is generally a postive play, as long as the count is not 2-1 and we don’t have the 3rd or 4th batter at the plate (or whatever the criteria was for negative results).”

Then I turn around and say, “Wait a minute. I am going to have two categories of runners on first base.  The first are my base stealers. If they go, they pretty much are going to be safe greater than the BE point, so I know it is OK to send them.  I don’t need you to tell me whether I should hit and
run though. I know it is not correct because I know that just sending them is OK.  Certainly if I force my batter to swing at a bad or terrible pitch, that can never be correct. I don’t need my batter to take a good pitch to protect my runner either.  I simply need him to do what he always does at that count - swing if he likes it and don’t if he doesn’t.  On top of that, he is going to get more fast balls anyway because of the runner on first.”

“Now, the other category is the non-base stealer at first.  NOW I need you to tell me whether the hit and run is correct.  I have no idea.  I suspect that it isn’t, but I’ll wait for your research.”

You see, Mike, I don’t think you can combine the two types of base runners.  If you do, of course you are going to come up with a postive result (for the hit and run) as long as there are enough good base stealers in your sample (even if those were indeed hit and runs with those runners at first).

But, rather than compare the hit and run to no hit and run, you should be comparing the hit and run to:

1) a steal by those base runners and no hit and run.

2) no hit and run with a non-base stealer at first.

So again, the ONLY thing we care about is what happens with a non-base stealer at first!

We already know the answer with a good base stealer at first. The answer is that a hit and run is better than no hit and run with the base runner not going, BUT a steal with no hit and run is better than that!

So if the answer is that a hit and run with a non-base stealer at first is wrong AND a straight steal (not all the time) with a good base stealer at first is better than a hit and run, then you would still get the results you are getting, but a hit and run is never correct! Does everyone understand that?

A really good base stealer, 80% success, will add .19*.8 - .45*.2 = +.062 runs

Mike is showing that the H&R is adding around that many runs. 

So, MGL is right that we have to take out the good basestealers.

I’ll look into that more, but as I said earlier, I really don’t think it’s biasing the results nearly as much as MGL claims.  The good basestealers are not stealing at anywhere near an 80 percent success rate in these situations, nor are they making up anywhere near a majority of the attempts.

To put it another way, I agree that some of what I am picking up is baserunning skill (which contrary to what you say, I believe it is appropriate to include) and some of that is being used on straight steal attempts (which I would like to exclude, but I believe are a fairly small fraction).

Also, to the extent that more good baserunners are involved in the hit-and-run attempts, I need to bias my expectation for the non-hit-and-run outcome to what would happen if a higher percentage of good baserunners were on base (in line with with the attempts, rather than all the situations).  That’s going to slightly reduce the advantage from the extra baserunner advancement on hit-and-run attempts.

I’m guessing based on the numbers I’ve seen that it’s going to move the overall run advantage from .06 runs to .05 or maybe .04.  But I’ll put it on my list to check for sure.

I agree that it’s going to have a limited effect, in the grand scheme of things.  But, that’s what we’re dealing with here, peeling away layers and layers of bias.

***

The .06 runs just seems way too high.

For example, I think in The Book I showed that a LHH with a runner on 1B showed a big jump in wOBA, as you’d expect (to some degree anyway).  He’s got the hole with the 1B holding the runner.  If you have a GB hitter, he’s dreaming of that situation.  I don’t remember now, but I think the jump in wOBA was around .020 points (I may be misremembering).  It wasn’t that much more than that.  Anyway, 20 wOBA points is around .017 runs.

Now, if the (non-fast) runner takes off, the 2B will be pulled away (maybe, at least a little, even though the SS should cover the throw with a LHH).  That helps the batter.  But the batter also has to swing more than he’d want to.  That’ll hurt him a little too.  It would seem that it should be kind of a wash at best for the hitter, while the runner gets a modest value.

So, that’s why I have my doubts at the degree of numbers we’re seeing.

Mike’s research here is important because it gives us plenty of research material, as well as providing a central discussion point that tackles all the potential biases.

Even if we end up that the H&R will add +.02 runs over not H&R (holding all other parameters constant), that would be a huge finding.  Even +.01 runs would be great new knowledge.  If it was a break-even play, it would be a pleasant surprise, and a feather in the cap of traditionalistas.  -.01 runs is I think the expectation of saberists.  -.02 runs would mean that the H&R should die.

I think that’s the kind of range that we should be seeing, -.02 to +.02.

If we get something outside that, either there’s biases still to uncover, or, if we’ve exhausted all that, we’d have to rethink our entire premise toward the H&R.

This is exciting research, made frustrating by the lack of quality stringer data.

Could we do a breakdown by caliber of the basestealer (SBA/Times on First)? That might help us understand what is going on here.

If by “we” you mean “Mike”, then I would support tie-ing him up to his chair until he gives us what we need.

I agree that the number I found is much higher than I expected.  And MGL identified one possible bias that I hadn’t thought of, which was good (that I need to compare to batter results when swinging at a pitch in these counts), and one possible bias that I had explored, but incompletely (looking at the pools of baserunners).  These things will probably inch the results down somewhat.  I’d be surprised if they push it all the way down to +.02 runs, but I’ll run the numbers and check to see (probably do that after Christmas).

In terms of expectations for impact on the infield defense, though, I would think it should be greater than that for the first baseman holding a runner.  The hole opened on the infield defense by the hit-and-run is in a location (the middle) where more batted balls go than toward the first-base side.  In addition, the fielder is moving away from the hole on the pitch in the hit-and-run situation, so his momentum is carrying him in the wrong direction, whereas the first baseman starts moving back toward the hole with the pitch.

The gain of 27 points of batting average for the hit-and-run attempt seems in line with that.  The batting average gain when a hit-and-run ball is put in play is about 40 points.  Those are mostly going to be singles, so something like .03 runs.  That’s a little more than the value of the first baseman holding the runner, so that makes sense to me.

This is exciting research, made frustrating by the lack of quality stringer data.

I’m not sure what you mean by that.  It’s simply not possible for a stringer to discern a hit-and-run from a straight steal, I don’t think.  And if they miss some small percentage of the runners going, I’m not too worried about that, unless we think there’s a reason for those missed notations to occur in a particularly biased set of circumstances.  (They didn’t appear to do so in the set of SB/CS plays that I checked.)

Btw, in case I haven’t made it clear, I appreciate the input on the possible sources of bias I might have overlooked.

This is the most selection-bias-laden topic I have ever tried to research.

One other thing that bugged me about the data was how positive the Yankees came out.  Either they were doing something nobody else was doing (possible), or I missed a source of bias that disproportionately affected them (also possible).

Mike: it would be interesting then to do a home/away split, or a park split (i.e., both teams at Yankee stadium).

Mike, it’s not the bias I am concerned with really. Again let’s assume that all of the possible hit and runs were indeed hit and runs and that you had no straight steals in your sample.

It’s still a no brainer that a hit and run with a good base stealer is better than no hit and run and no stolen base attempt.

So the only 2 interesting questions are: One, is the hit and run good with a non-base stealer on first, and two, is a hit and run with a base stealer on first better than a stolen base attempt?

Mike doesn’t answer either of those questions.

If nothing else I think we want to know whether the hit and run is good for those two categories of base runners separately. I mean does anyone mind the hit and run with a fast runner on first?  When we cringe is when there is a hit and run with Jason Varitek on first. 

One more thing. Mike, do you have the swing and miss rates overall or on the various counts/outs situations (when the hit and run is on of course - ie when the runner is going)?

It’s still a no brainer that a hit and run with a good base stealer is better than no hit and run and no stolen base attempt.

That’s not the comparison, though.  I’m comparing a hit-and-run with a good basestealerto no hit-and-run with a good basestealer (possibly including a stolen base attempt).

I won’t be doing that perfectly because I’m not separating the hit-and-runs perfectly from the straight steals, but if your assumption, for the sake of argument, was that I got that separation perfectly correct, then all the straight steal attempts will be in the “no hit-and-run” bucket.

I just don’t agree with you that it was a no-brainer prior to doing this analysis that the hit-and-run attempt was a positive play with a good runner on base.

Secondly, I’ve checked, and the hit-and-run does not seem to be put on disproportionately with good runners on base. 

The runners on first for the hit-and-run attempts had an out-of-sample SB attempt rate of 14% of the times they were on first base, with a success rate of 74.4%.  The runners on first in hit-and-run situations where no hit-and-run attempt was made had an out-of-sample SB attempt rate of 10%, with a success rate of 73.8%.

So, there is some difference between the two groups, but not dramatically so.

I can check for the effect in the reverse direction, too, i.e., how the hit-and-run rates vary based upon baserunning skill.  I’ll do that later.

One more thing. Mike, do you have the swing and miss rates overall or on the various counts/outs situations (when the hit and run is on of course - ie when the runner is going)?

Yes, I have that data.  I will look it up and get back to you.

I think what MGL was saying is that if you limit your choice to H&R with a good base-stealer, or no H&R, and no SB attempt, then you do the former.

In the latter case, keeping a good base-stealer from trying to steal would be a net negative.  So, you want the good base-stealer in motion.

That even if a H&R from the batter’s perspective was a slight negative, having a fast runner in motion is a big positive.  As I showed, a really good basestealer will add +.06 runs.

Is this what you are talking about, MGL?

I think what MGL was saying is that if you limit your choice to H&R with a good base-stealer, or no H&R, and no SB attempt, then you do the former.

Right, I do think that is what he was saying, and if so, that’s a faulty comparison.

The choice is not between doing a hit-and-run and keeping the runner tied to first base.  It’s between a hit-and-run and whatever you normally do when you don’t hit-and-run, which includes steal attempts.

It’s only to the extent that straight steals by good baserunners are polluting my hit-and-run population that MGL’s comparison makes sense.  The fact that SB success rates are dropping 15-20% in these situations is a strong suggestion to me that the straight-steal-by-good-basestealers pollution is fairly low.

It’s not non-existent, of course, so I agree that MGL has a point at the margins.  I don’t see how it re-frames the debate, though.

Very interesting article, Mike.  A few quick thoughts (I should give it a 2nd, more careful read):

Could the poor outcomes on 2-1 count be a function of mainly weak hitters being asked to H&R on such a (relatively) good count?  I would guess that the better the count, the weaker the pool of H&R hitters becomes.

One factor that will reduce the estimated value of the H&R is that the runner will sometimes leave too early and be picked off, but this won’t be counted since no swing occurs.  And if the defense pitches out, again this will be missed by the analysis (with presumably bad results for the offense).

That said, I don’t doubt at all that the H&R has a significant positive value on balance.  That might not be true if you remove the “run and hit” plays.  But in that case you need to add the value of those plays when you estimate the value of SBAs.  Historically that value (fewer DPs, more advancement) has largely been ignored, and Mike is providing an important correction here.

Yes, that is what I am saying.

“I’m comparing a hit-and-run with a good basestealerto no hit-and-run with a good basestealer (possibly including a stolen base attempt).”

Yes, I realize that some of your non hit and runs include the base runner going (and no swing).

But that sample is also leaving out all the times when it is a straight steal and the base runner is going.

So, isn’t your not and hit and run sample going to be biased in favor of deflating the overall run value since it doesn’t include times when the runner is going the batter swings (presumably at a good count and with a good pitch)?

Sorry to bold that sentence again, but I think that is a fatal flaw in your non-hit and run sample.

Mike, how about you just give us the results with a non-base stealer at first.  Isn’t that what we are most interested anyway.  As I said, if a good base stealer runs, we don’t care that much whether the batter is forced to swing, although obviously we might prefer the runner to be safe on a steal a high percentage of time.

I mean, even if we find out that the extra hits by opening up the hole plus the extra advances by the base runner more than makes up for having to swing at some bad pitches, don’t we want to know the BE rate for the runner when the batter swings and misses.

Apparently the swing and miss rate is fairly high, otherwise the difference between BA and OBP for the hit and run situations would be a lot lower than they are (again, did anyone else find it curious that there was such a large difference?)…

MGL, Guy, and others have raised some good questions.  My plan at the current time is to address these questions after the Christmas holiday.

Sure thing.  A couple more things:

If you want to make sure that most of your straight steals are filtered out, only use the runners who never steal.  After you get the numbers with them, then you can incorporate how the success percentage on swing and misses (with the hit and run on) changes the overall numbers.

You can also estimate how many straight steals you have by looking at the swing and miss percentage.  The swing and miss percentage should be lower on hit and runs, I would think (even though they are forced to swing at some pitches that are hard to make contact with).

For example, if the swing and miss percentage with non-base stealers on first is 10% and with base stealers, it is 20% (assuming you control for the identity of the batter of course), then you likely have many straight steals with those base runners on first.

But again, if you want to eliminate the straight steals why not just eliminate the PA with a base stealer on first.

I would also do the whole thing another way. Instead of comparing hit and runs to no hit and runs having to worry about sample biases in both samples (the hit and runs and the non-hit and runs), why not just look at the costs and benefits of the hit and run.

Identifying the likely hit and run situations is 95% of the battle and not being able to has kept us from analyzing the hit and run in the past.

So once you identify mostly hit and runs (again, you might want to only look at non-base stealers at first), then compute the run value of the difference in batted ball outcomes between with a hit and run and a non-hit and run.  Presumably you get more singles and fewer extra base hits, more base runner advances, and a high percentage of CS on a swing and a miss.  Then add that, proportionally, to the value of a generic PA after a batter swings and misses.  Then compare that to a theoretical normal PA (no hit and run).

Basically I am suggesting looking theoretically at the value of a non-hit and run PA rather than actual ones, which are likely biased because you are eliminating some very favorable contact where the runner is stealing.

Apparently the swing and miss rate is fairly high, otherwise the difference between BA and OBP for the hit and run situations would be a lot lower than they are (again, did anyone else find it curious that there was such a large difference?)…

The runner is safe more than 50% of the time when the batter swings and misses the pitch.  Since that opens up first base with a man on second or second and third, many of the walks are pitch around walks or intentional walks that would not occur without the hit-and-run attempt.

One possible filter that could be used to differentiate true hit and runs might be the number of times a pitcher throws to first.  More than once or once with a pitchout would indicate that the defense was expecting the runner on first to attempt a steal.

Incidentally, Mike, I just wanted to say thank you for the first study of the hit-and-run that was worthy of any criticism. We’re going to learn a lot from all this.

I agree with Charlie.  You know you have a great hit and run article, if you have tons of (constructive) criticisms of it.  I would have been disappointed if the saber community did not step up with its opinions.

A hit and run article with NO criticisms would either be the perfect article, or a useless article.

***

By the way, Mike made a note that there was little he could find on the web.

I seem to remember Tippett doing something in his online newsletter.  One of his last entries if I remember.  I’ll see if I can find it…

Bad memory.  Tippett did something about sending the runner on a 3-2 count.  I created a new thread on this.

So, forget what I said otherwise.

The only way a walk can occur is when the batter swings and misses and subsequently walks, right?  Again, that seems like a lot of walks.

Or if the batter fouls the ball off and subsequently walks.

One more thing. Mike, do you have the swing and miss rates overall or on the various counts/outs situations (when the hit and run is on of course - ie when the runner is going)?

The swing and miss rate is 17.7% of swings when the runner is going and 18.3% of swings when the runner is not going, in the hit-and-run situations that I defined.  I don’t think that will work as a reliable indicator of hit-and-run plays.

Could the poor outcomes on 2-1 count be a function of mainly weak hitters being asked to H&R on such a (relatively) good count?  I would guess that the better the count, the weaker the pool of H&R hitters becomes.

The effect is there, but it is fairly small.  The pool of hitters asked to hit and run on 2-1 is about .004 runs/PA worse than the hitters not asked to hit and run on 2-1, as compared to the difference in quality of the overall pools of hitters in the other hit-and-run counts. That is still not nearly enough to bring the hit-and-run result positive on the 2-1 count (from -.026 runs/PA to -.022 runs/PA).

One factor that will reduce the estimated value of the H&R is that the runner will sometimes leave too early and be picked off, but this won’t be counted since no swing occurs.  And if the defense pitches out, again this will be missed by the analysis (with presumably bad results for the offense).

Do you have any ideas of how to identify the pickoffs that occurred on what would have been hit-and-run plays vs. pickoffs that occured normally?  I would guess it’s a fairly rare occurrence, but I don’t know, and if I could attempt to measure, that would be nice.

My understanding of the basic difference between a straight steal and a hit-and-run play is that the runner waits just a fraction of a second longer in order to make sure the pitcher is committed to the plate.  I thought he does that in order to avoid pickoff plays, with the hope that it won’t matter if the batter makes contact.  He won’t be perfect at it, of course, but I’d think the pickoff rate would be much lower on hit-and-run attempts.  I don’t know how I’d measure that, though.  With only about 5% of hit-and-run situations turning into hit-and-run attempts, I can’t look at the pickoff rate in hit-and-run situations and expect to be able to measure anything even if it happens disproportionately on hit-and-run attempts.

Adjusting for baserunner quality turns out to be a fairly important adjustment, along the lines that I thought it would be.  Differences in baserunner quality between H&R attempts and H&R situations with no attempts inflated the value by about .016 runs/play.  If make that adjustment, it drops the overall hit-and-run advantage to about +.045 runs/play.

I divided the runners on first base into four equal groups based upon their stolen base attempt rate over the period 2003-2011.  I defined the stolen base attempt rate as (SB+CS)/(singles+UBB+HBP).

The fastest runners do get more advantage out of the hit-and-run play than do the slowest runners.

Why not include IBBs as well? And if you can, wouldn’t it be better to look at SBA% in non H&R situations when determining who the good basestealers are (and, of course, limiting for sufficient sample size to make the determination, too)?

Why not include IBBs as well?

What do you think the stolen base attempt rate is for intentionally walked batters?

And if you can, wouldn’t it be better to look at SBA% in non H&R situations when determining who the good basestealers are (and, of course, limiting for sufficient sample size to make the determination, too)?

There are probably a number of things I could have done to get a more accurate measure of baserunning/basestealing ability/tendency.  But I don’t think your suggestion would move the needle much given that failed hit-and-run attempts make up only a small percentage of stolen base attempts.

I think it’s worth making a reasonable effort to determine whether the hit-and-run is a +.06 or a +.02 or a -.05 run play.  If someone else wants to do the analysis to move the needle from .045 to .043, they are welcome to do so.  I’m willing to assist anyone who is interested in replicating my research, assuming they have access to Retrosheet data.

SBA% for IBB victims is likely really low. I doubt that will make much difference; I just don’t see why not to include it. The rest, I don’t know if it will make much difference, but it seems reasonably probable that it could.

It’s ridiculous to consider adding IBB as an opportunity, certainly at the same level as a single and walk.

When I do it, I put in the denominator:
80% of singles
60% of hit batters and walks

If I included IBB, I’d put 1% if not less.  There’s more chance of a SB following a double than a SB following an IBB.

C’mon, let’s not get so picky, that you end up making it worse.

Keeping the denominator as 100% for singles and walks and hit batters is certainly acceptable.

If Mike wants to improve it, he can use Retrosheet and find out how often he steals 2B (and 2B was open).  But, I doubt that’s needed for what he’s doing.

One other thing someone mentioned, I think maybe it was MGL, was that batters swinging at pitches in hit-and-run situations probably included some portion of batters swinging at really good pitches to hit on straight steal attempts.

There may be some of that going on; however, when I looked at all hit-and-run situations where the runner was not going, batters taking pitches did about .01 runs better than batters swinging at pitches.  So the fact that a batter swung at a pitch does not in and of itself indicate that the batter was at an advantage.

Do faster runners have a big advantage in SB% too? With the hit-and-run going, obviously.

Also, there’s going to be a selection bias with the runners. Those runners who have already attempted a stolen base have removed a set of plate appearances from the sample, since they are no longer on first base. Does including these change anything?

Charles/49, I have those numbers, but I’ll have to look them up and post them later.  The results are included in the numbers I posted earlier, but I can break them out separately.

Those runners who have already attempted a stolen base have removed a set of plate appearances from the sample, since they are no longer on first base. Does including these change anything?

Those plate appearances are included in the numbers I used.

Neat. You really have thought about this study.