THE BOOK--Playing The Percentages In Baseball

Baseball orthodoxy and the Rockies - yeah, that fits…

Incidentally, I am using Hit Tracker to calculate HR park factors based on the geometry of the fences and accounting for weather conditions (which is all that really should matter if you truly want to isolate the effect of the park), and I find myself wondering how the propensity to bunt may impact conventional park factor calculations.

A park that is conducive to homers may show fewer than expected if the manager has his guys laying down a lot of bunts (or doing a lot of hit & run or other “small-ball” plays, for that matter).  Not as many double or triples, either, if the batters are swinging away a lesser percentage of the time (or giving away outs to cuaght stealings).  This would only come into play if small-ball was practiced more at home, or more when away, of course…

Probably a minor effect, but possibly not.  Any thoughts on this?

Fences (distance and height), weather conditions, and altitude.  Perhaps how pitchers pitch in that park for whatever reason.  For example, in a home run park, they might try and keep the ball down, so that there are fewer homers than the fences, altitude, and weather might suggest.

To answer your question, home run and other component park factors should be per some denominator that is fairly constant across all parks.  For example, first I eliminate all IBB’s.  Then I eliminate all bunt attempts.  That takes care of your problem.  Then I figure foul out park factors (which vary a lot among parks).  Then I factor out the foulouts and figure K and BB factors.  Then I do HR per ball in play.  You could also do HR before factoring out the K and BB.  The danger in that is that if a park affects BB and/or K then the HR per non-IBB and non-bunt attempt PA could change even if the HR per BIP does not.  Since K and BB do not vary that much from park to park, it is not that big a deal to compute HR per PA, but, as I said, you probably want to factor out bunt attempts, IBB, and bunt attempts first.  For example, there are more bunt attempts in grass parks than in turf parks.

In addition to MGL’s list, the profile of the players.  I’ve blogged this before, but Coors impact on Dante Bichette and Juan Pierre is not, and cannot, be the same. 

If you based your park factor solely on the home batters, and if those batters were guys like Pierre, then you’d conclude that Coors is a neutral-HR park.

***

Also, you need to know the handedness and FB tendency of the batter/pitcher.  Those also affect the park impact.

I think he means, “What is it about a park that actually affects players’ HR rates?” and not how it affects players differently or how it might be computed.

Once we know how a park affects HR’s independent of the players, how it is computed, etc., then theoretically we can compute how it indeed does affect every single player.  For example, if we know that on fly balls to LF, the weather and altitude increased distance by 5 feet and the fences were 5 feet closer than an average park, we could take all of Pierre’s fly balls to left field, and all of Bichette’s fly balls to left field and figure out their “virtual (theoretical) home runs” in a neutral park.

I was fast forwarding through the game this morning and when I saw that I knew what was going to happen. Hawpe has never been able to bunt and if Hurdle really wanted a bunt he should have pinch hit at that point. All he ended up doing was putting Hawpe in a position where he was in an 0-2 hole. But against Stanton it was pretty likely that Hawpe was going to K anyway. Iannetta and Baker were still on the bench…

Sure, if he was going to bunt, which he did, absolutely why not pinch hit?  There is noithing uglier trying to watch someone bunt who is not good at it.  I could try and run the numbers, putting in some poor success rate, but I imagine that the overall results of a sac bunt attempt by someone who bunts poorly would be atrocious.  Surely ANY manager must know that.  It is almost like a manager wants the batter to go to 0-2 and then he can yell at him and blame him for not being able to get the bunt down.  Regardless of whether a major leagueer should or should not be ABLE to bunt, why would you ask someone to do something that he is clearly not good at when there is a perfectly good alternative (hitting away or pinch hitting and bunting)?

I would love to have someone ask the manager in a situation like that, “Exactly what were you thinking?  Do you think that the overall effectiveness of a bunt attempt versus swinging away has SOMETHING to do with how good a bunter the batter is?  If yes, at what point (how good/bad does the batter have to be at bunting?) do you think that you should not ask a batter to bunt and that hitting away would be a better overall alternative?”

IOW, like in school, teach a manager to be a critical thinker and a more effective decision-maker by encouraging him to THINK about his decision rather than operating on “auto-pilot,” “by the book,” according to some “conventional wisdom,” by some “hunch or intuition he may have,” or “what he always does,” or “what he THINKS is right, without really thinking about it rationally and logically.”

IOW, while I don’t think that ANY manager can figure out the correct deicision in situations where the alternatives are close in terms of WE, I think that most managers ought to be able to figure out the correct decision in situations where with a little critical and logical thinking, the answer should be fairly obvious (like whether to sac bunt when the batter is a poor bunter).

MGL, you are right, I am looking strictly for physical characteristics of the parks at this point; I think your list is good, I hadn’t thought about the surface influencing bunt rates…

BTW, do you think the K and/or BB park factors are ever real, or simply noise?  To be real, you would have to believe that a pitcher K’s more of the same opponent at home than away, and vice versa.  Maybe that would be the case in a larger park where the pitcher pitches more confidently, or in a smaller park where the pitcher is looking over his shoulder at the Green Monster or the Crawford boxes…

Tom, I am going to use a complete set of test fly ball trajectories that covers everything from the slap-hitters down-the-line jobs to the high and far blasts, and everything in between - so I don’t think the team makeup should matter for my calc.  And as far as what I compare to, most park factors include home and visitor results at the park, so the results from a team of sluggers at Coors would be analyzed along with that same team’s performance on the road.

To complete the thread-jack I executed here, there are some other things I’ve analyzed on park factors, including the impact of inter-league play and the unbalanced schedule that have a suprisingly significant impact.  I’m going to try to describe all the warts of the current park factor methods…

Greg, where is your work going to be presented?  Yes, the BB and K factors are “real” but the year to year correlations are very low for the BB and fairly high for the SO so that the regressions are a lot for BB (70% for one year) and not so much for the K factor (around 50% for one year). Lighting has a lot to do with K rate and to a small extent BB rate. BB rate is mostly noise, but as you say, in hitters’ parks, pitchers tend to nibble and walk more batters. I use only regressed component park factors to factor out the random noise, of which there is a lot, not to mention that caused by fluctuating weather from year to year and different personnel on the home team.  As Tango says, even if you are using the same players in your home and road data, which you pretty much are, if one year or for several years you have a bunch of Juan Pierre’s, the HR factor is going to be different than if you had a bunch of Barry Bonds’.

Of course for park factors, you don’t want to use inter-league data at all, and because of the imbalanced schedule you want to weight your data according to which parks each home team is playing in and how often.  You can’t just compare home and road stats and call that a park factor relative to the whole league.  It is a PF relative to the parks that each team plays in.

Greg - Welcome back. I hope The health problems you had in April are over.  I have taken your HitTracker data from last year and matched it with the MLB hit location data to see how close they were.  I expected a fairly large difference in distance since you weren’t really trying to measure the same thing as MLB.  The average discrepancy was around 30 feet.  The angular measurements were very close, about +- 2 to 3 degrees; even closer near straight away center and the foul lines, and a little more in the alleys.  I expect the non home run balls will be much closer in agreement in distance and a little better in angular measurement since reference points will be clearer.  I am looking forward to analyzing that data.  Any chance you will start posting it again soon?  Also, is there any chance that you will be including ground balls as well as fly balls?

MGL - In your “Irony” thread you have been a pretty harsh critic of the biomechanics “experts” who give us analysis after the fact rather than real studies that predict behavior and can be tested for validity.  But isn’t that exactly what you have been doing in this thread?  I haven’t seen your list of players that a manager should never have bunt, and yet here you are telling us that because Haupt is a bad bunter it is always a bad play to have him bunt and that this is something any manager should know.  You may be absolutely correct, but it doesn’t seem to measure up to the rigorous scientific predictive analysis that you are requiring of others.

MGL, I haven’t decided where to present the data yet, but certainly I’ll be posting it on my site, and I’ve floated the idea with Studes of posting it in THT.  We’ll have to see if it’s solid enough for that venue… I would appreciate some peer review before that, though, if you and any others here would be willing.

Peter, 30 feet sounds like a fairly large discrepancy, can you provide me a link where I can look at the data you’re comparing with?  I’ve never seen a database of MLB-approved hit locations that covered outside the fence.  I have looked at the home runs distance-only data posted by Stats Inc., though, and have generally found it to be hit or miss compared to mine, sometimes similar, sometimes quite different…

And as for the inside-the-fence hits, you ought to be careful, because I believe the data recorded by MLB (or its designates) is the fielding location, not the landing location.  I came across this when I did an analysis of J.D. Drew’s hits from 2006, and a triple he hit at Great American BP was plotted in deep left center by the wall, whereas the video showed a sinking liner that ducked just under a diving Griffey’s glove and skated all the way to the wall, after which Griffey ran it down and nailed Drew at the plate.

The back is better, thanks for asking, but not yet back to normal.  Now that my volunteers are through spring finals, we should be able to get some data up on all hits.  It will include grounds balls as well as everything else (except popups, which will be location and time of flight only), but the amount of data to cover is so huge, we will be nibbling at it rather than covering completely, which is beyond our capabilities.

Greg - I used the hit locations given in mlb’s gameday XML files.  Its what they use to generate the graphics that show all a players hits and outs superimposed on a plan of the ballpark.  I’m sure its not meant to be superaccurate but its all you can get for hit location for free.  I was aware that mlb only gives the location of where the ball was picked up and was planning only to compare balls caught in the air and the angular component of ground ball outs.  I was mostly curious about the limits of observational data.  It doesn’t make sense to design a fielding metric based on super small zones if independent observers can’t place a ball on the field closer than 10 feet and 2 degrees of angular measurement.  I hope we can convince mlb to give us speed off bat and vertical and horizontal hit ball vectors off the bat from the same set up that they use to generate pitch data for enhanced gameday.  It would be one more chunk of information to provide a confirmation of the data that you will be providing.  What does your program do now to include a factor for Magnus effect?

Peter,

Have you read my chapter on sac bunting in The Book?  In it, I clearly explain what happens when a poor versus a good bunter attempts a bunt.

Even if I didn’t, if we KNOW, through the scientific process that one alternative is about equal to another one, then, when something detracts from one of those alternatives, we AUTOMATICALLY know that it becomes the WORSE alternative.

MGL - I have read your chapter on sac bunting in the Book.  I even reread just before I wrote my post.  But your analysis about what happens with an average pitcher and an average lineup is not really a blueprint for action in any real world situation is it?  Its certainly a far cry from your statement at the beginning of this thread that its always a bad play to have a bad bunter bunt.  Its also a bad play to have a bad hitter hit away.  But to have a testible hypothesis you have to define how bad a bunter and how bad a hitter and than see how such an individual actually faired in real world situations when he did or did not bunt.  In the Book you made a start at defining a hypothesis, but you didn’t finish defining it in any way that would give it real world applicability for a manager and you certainly didn’t test it.  Which puts you in about the same situation as the biomechanical experts that you criticize.

Peter, there are things that I don’t mind having a legitimate argument about.  This ain’t one of them.  It appears to me that you are arguing for the sake of argument.

If you are saying that if I made the statement that, “It is 100% NEVER correct for a bad bunter to bunt,” then you are right my friend.

If you are saying that, “We must first define what we mean by a ‘bad’ and ‘good’ bunter,” then you are also 100% correct.

If you are saying that, “It depends where the defense is playing (for example, it is probably OK for a poor bunter to bunt if the defense is playing as if the batter were Barry Bonds),” then you are also correct.

I don’t remember exactly what I said initially, and if I said anything to contradict the above three things, then I take them back and concede that you are right.  If I said or implied that, “Almost all the time that a poor bunter tries to bunt, and the defense is at least considering the possibility of the bunt (which they did when Haupt squared early, and continued to square until he got 2 strikes), the bunt is WRONG, regardless of the quality of the non-pitcher batter,” and you disagree with THAT statement, paying particular attention to the words “ALMOST ALL THE TIME,” which means NOT 100% of the time, then you are dead wrong, and I can prove it, but you can do that yourself easily using the data in the book.

Hi.