THE BOOK--Playing The Percentages In Baseball

I can’t get the formatting right in the comments section, but if you want to see the results in an easier to read format, just go here

Sorry guys. Feel free to post the chart in here if you can get it to work.

If there is interest, I can give the result by spray and launch as well.

Thanks

The problem is with blogspot.  They don’t make linking easy.

You would do this:

But, that url is a temporary URL blogger creates.  You need a permanent URL.

Here are the weights I’m using - first the rollup of values by year. As you can see, 2009 is running on the high side. The inputs were based on 2008 MLB totals, I’m not (yet) park adjusting or separating out by season.

Now the weights, slightly different than the BA version, using different inputs as well as more column to keep the look-up query simple.

Using Harry’s updated run values, here are the average run values by spray location according to hit f/x for April. I have taken out the bunts.

45 degrees is down the left field line and 135 degrees is down the right field line. The highest average run value is RHB that go opposite field.

The four main dips in value for LHB is where the infielders play, if I had to guess. I know there has been a lot of discussion in this blog recently about fielding position, so i’m going to stay away from that.

Good job!

Actually, if you can show the LHB with runner on 1B and with 1B empty, you will likely see a somewhat sizable difference.

Harry, those values...what is the ValF column?  (I get all the others).

Never mind, value of a foul ball.  Kind of a silly thing to include, no? (It’s the same as a strike except with 2 strikes, and then it’s Zero)

Here are the values for a left handed batter with a runner on first vs no one on.

As expected, the run value drops down the first base line with the fielder holding the runner. The gap between first and second open up while the value drops up the middle, probably because teams playing for the double play/steal attempt.

Jay: I thought that 45 degrees was down the first base line from Harry’s article on THT.  Did you convert the angles?

I would suggest that 0 degrees be up the middle, and then go -45 at 3B and +45 at 1B.

Loran, as I noted, the extra columns I added are to simplify/optimize the queries.  Also, I’m not assuming that we won’t, some day, assign or discover a value to two strike foul balls.

Tango: I would agree that 0 degrees up the middle would simplify everything.  I was questioning some of the “conclusions” from the plots if it is actually flipped.  To me the plots make more sense if 45 is actually the first base line, i.e large spike between 1B/2B for LHB and between SS/3B for RHB.

Harry - Your linear weight value for the HR looks way off.  It should be around 1.40 just as Joe gives it in the cited BA article.  As a matter of fact all of Joe’s are closer to what I have for the last 4 years than yours.  What time period are you basing yours on?  Just the recorded Pitch f/x pitches that you mention?  And why are you giving linear weight by count numbers to Jay when he doesn’t have count information to work with?  Isn’t that just confusing things?

Tango - There are almost as many systems of designating the angles of hit balls as there are analysts working with the data.  Why are you selecting yourself to be the arbiter of how it should be done when (a), you don’t even work with the data yourself, at least publicly; and (b), the system you suggest is the least used of all the systems out there?  Over the past week you have suggested changing the terminology of vXY, Horizontal angle, and Vertical Angle for no real benefit that I can see.  Why, since as I mentioned before you don’t even work with the data?

JBrew: flickr is blocked at work, so I can’t see the images you posted.  I’ll have to check at home tonight.

***

Peter: I said:

I would suggest that 0 degrees be up the middle, and then go -45 at 3B and +45 at 1B.

You said:

Why are you selecting yourself to be the arbiter of how it should be done

Since I suggested something, that doesn’t make me an arbiter, does it?

To me the -45 to +45 makes sense, since we get to see mirrors.  For example, just as an illustration, the RHH will pull the ball as much as a LHH.  If I say the RHH spray angle is an average of -30 degrees, and the LHH spray angle is an average of +30 degrees, doesn’t that make it crystal clear that they pull to the same extent?

On the other hand, if I say 60 degrees for the RHH and 120 degrees for the LHH, is that just as apparent?

To me, the scale should mean something, other than choosing whatever arbitrarily makes sense to whoever devised the system.

***

Horizontal angle, and Vertical Angle for no real benefit that I can see

You don’t see it for yourself, but I see it for myself.  I’m proposing it, and if that means more people think about it and adopt it, all the better.

***

And, I don’t need to be working with the data, publicly or otherwise, in order to make suggestions.  Indeed, Mike Fast has taken at least one of my suggestions, and others have taken them as well.  If the limit of my public contribution is having other PITCHf/x-ers adopt some tiny ideas of mine, I’m ok with that.

BIS used to have some really strange way to plot trajectory.  I said “Why don’t you use angles”.  Boom.  They made the change immediately, which made processing so much easier.  And I didn’t even need that part of the data (at the time).

JBrew: You are correct, the chart was flipped so throw out the above analysis. I’ll post the updated one:

Peter: I do have the count information. I was just using linear weights from an early 2008 article, Harry was just updating the values for me.

Trying the -45-0-45 approach:

While we are here, this is the average RV by launch angle:

11.5 degrees is the top average launch angle, by run value.

I’ll take all the feedback I can get guys. It’s the only way I’ll figure all this stuff out. Most of my knowledge of this stuff has come through comments in these blog threads, so I find them invaluable.

Having limited experience to rely upon with this analysis, it’s a learn as you go process. For me at least.

Tango - I disagree with you.  I use 45 degrees for the left field foul line and 135 for the right.  HitTracker and Sportvision and Alan Nathan all use 135 for left and 45 for right.  I change all of their values to mine when I add their data to my files for consistency in all my files which I created before Sportvision and HitTracker began reporting data.  But if I were to report research based on Sportvision or HitTracker data I would use their conventions.  I would do that because it would avoid confusion and is the respectful thing to do since I am using their data.  Since they are the two major public sources of hit ball angular data I imagine that their convention will, and should be, become accepted.  I can understand why you might want to have consistency in the way horizntal angular measurement is reported.  But why would you suggest method that NOBODY uses?  That, to me, seems disrespectful of Greg and Sportvision and just like your ego trip to me.

The terms horizontal and vertical angles have all been used consistently by HitTracker, Sportvision and me to describe the flight the ball takes off the bat.  Why would you CREATE confusion by suggesting alternative terms when there was no confusion?  More ego trip?

I introduced the term vXY to represent the velocity of a hit ball in the XY plane at last years Pf/x Summit.  I used that specific term to make it consistent with the terms Vx, Vy and Vz that Sportvision was already using and to differentiate it from SOB and because velocity rather than speed is the more correct term when a direction is specified.  Why would you feel the need to change that term when, again, there was no confusion about what it represented?

Peter, the weights are based on 2008 MLB, not split by league. Any tuning or improvements that can be made, well, that’s why I posted them here, to get some feedback.

But I’m not sure I understand what you mean. The home run values are around 1.4, varying based on the count. 3-0 is 1.25 and 0-2 is 1.57, which sounds about right.

Jay - I thought that the count information was not included with the pitch information in the Hit f/x file, but I now see that it is.  You do realize that you should only be using the values given in Harry’s file for hit balls at the 0-0 count, don’t you?

Peter, re 21, please explain why you would do that.

re 21, I don’t understand why you would do that either.

Harry - The only values that are meaningful for Jay’s research are those at count 0-0.  There you have a HR value of 1.47.  The correct value for 2005-2008 empirically derived linear weights is 1.40.  It looks like you derived your values from only looking at HRs that happened on a 0-0 count.  That is incorrect and limits and biases your sample.  You must calculate the 0-0 hit values for all events at all counts and then adjust the other count lines for the value of the balls and strikes that have come before.  The reason for this is that a HR or any other batting event result) is going to have the same ultimate average run value if the BaseOut situation is the same, no matter what the count is.  All you should be adjusting is the expected run value of the particular count you are in.

That’s not the case. Using 2008 MLB’s aggregated batting line, I found event weights, then, using wOBA by count (all found within the resoures on this site) and adjusted the run values. There’s no selection of events by count whatsoever.

If you derive the LW for a HR in 2008, league/park neutral, what value do you find?

I believe what Peter is getting at is that the count shouldn’t matter in the analysis that Jay is presenting.  What we should be concerned with is the runners on base and out state instead.  To a batter, his run values shouldn’t matter whether he hit the HR on a 0-0 count or 3-2 count the outcome of the PA is still the same, but a bases load HR is greater than a bases empty HR.

Jay should be using these instead:  http://www.tangotiger.net/lwtsrobo.html

HR 2008 = 1.40265477654779, but I think I’ll round to 1.40.  You used Tango’s generator to generate your linear weights?  I believe it is a Markov generator using average values for each batter.  As such, it will overestimate HR values because a greater percentage of HRs are hit with no one on base.  At least I think I have the reasoning correct on that.  I calculate my linear weights from the actual Run valued added for each particular event during the season.  That way accounts for the different distributions of BaseOut states that actually occur for each event.

Peter/27
Ah, I think we’ve hit the nail. Thank you.

I’ll re-run my weights using the run values, not the markov values. I chose those, along with the “through count” not the “at count” - in addition to switching from the markov values, would you suggest a change from through to at?

JBrew - You are correct that I am saying that the count shouldn’t matter.  But Jay shouldn’t be using linear weights based on BaseOut state either.  He is only calculating the linear weight of specific categories of hit ball speed, not hit ball speed as it relates to BaseOut state.  Therefore, he should be using the generic linear weights of hit ball events, 1B = .48, 2B = .78, 3B = 1.06, HR = 1.40, Batter reached Error = .51, batted out = (-.288).  If he were ambitious he could use separate values for air outs and ground outs and infield singles and outfield singles.

Peter: I would prefer that you accept what I say on its face, and I am offering what I think are helpful suggestions, and it’s not some sort of ego trip as you keep repeating.

I also don’t think it’s disrespectful if I suggest my own terms.  As I said, I find it more useful and helpful to have -30 and +30 be mirrors of each other.  The common reader would surely find those points more convenient than 60 and 120.  Or, what if it’s -27 and +27.  Is the reader going to instantly realize that 63 and 117 degrees are mirrors of each other?  I completely disagree that Sportvision and whoever else is going to work on this will be able to set the standard.  For me, it’s a silly standard to have 63 and 117 be mirrors when -27 and +27 are obviously mirrors.

Not to drag Alan into this, but he continually makes reference to launch angle in several different threads:
http://www.insidethebook.com/ee/index.php/site/member_posts/?name=Alan%20Nathan

I’ll keep saying launch angle, and you can say vertical angle at point of contact or however else you wish to describe it.

You have made your point clear, that you are offering your perspective as to how you see how I am proceeding.  However, I will continue to operate as I am doing.  If your suggestion to me is that I change my opinion or keep my opinion to myself, that’s not going to happen.  If your objective is to let me and everyone here know that you think I am on an ego trip who is being disrespectful to pioneers, then your objective has been satisfied.  Otherwise, I really don’t know why we are having this discussion.

If you feel it necessary to continue discussing this topic, I will be glad to open a separate thread and move these comments there.

But why would you suggest method that NOBODY uses?

Peter, I have been using, publishing with, and advocating for a angular convention of center field = 0 degrees for a while now, well prior to seeing any of the HITf/x data from Sportvision this past winter.

I may be forced to adapt my convention for the very reason you give:

But if I were to report research based on Sportvision or HitTracker data I would use their conventions.  I would do that because it would avoid confusion and is the respectful thing to do since I am using their data.  Since they are the two major public sources of hit ball angular data I imagine that their convention will, and should be, become accepted.

However, I will continue to do my own analysis using my convention because it’s easy so much easier to account for handedness differences in a system where center field is the reference point. 

Btw, there is a third major public source of hit ball angular data--the MLBAM Gameday stringer data--that for all its flaws is still by far the most complete public record of hit balls.

I introduced the term vXY to represent the velocity of a hit ball in the XY plane at last years Pf/x Summit.  I used that specific term to make it consistent with the terms Vx, Vy and Vz that Sportvision was already using and to differentiate it from SOB and because velocity rather than speed is the more correct term when a direction is specified.  Why would you feel the need to change that term when, again, there was no confusion about what it represented?

In some cases like this, there may be a divergence between terms that are best for science/research and terms that are best for wider public understanding.  For research purposes vXY is perfect because it says concisely and specifically what it is.  For public consumption, it’s a little bit tougher to use that term because it requires an understanding of the PITCHf/x coordinate system, as well as typically being denominated in feet/second, whereas something like “horizontal speed off the bat” has the advantage of being clearer to the layman because it is in a coordinate system he understands and in units he can relate to.

Harry - I wouldn’t bother calculating the value for hit ball events at different counts unless you have a specific project that requires them, and then whether you chose through counts or at counts depends an what that project requires.  I calculated them both ways several years ago when I was looking at pitchers BABIP at different counts, but I have not had a use for them before or since.

For people who are looking for empirically-based Linear Weights values, you might also consider using one of these:

1. Ruane data, binned
http://www.insidethebook.com/ee/index.php/site/article/linear_weights_by_run_environment/

As you can see, even using many leagues, you get data that jumps around a bit.  Even the run value of the single does not have a smooth run.

2. Retro data, totalled
The empirical data for all the Retro years is here (at post #4)
http://www.insidethebook.com/ee/index.php/site/comments/actual_wins_retrosheet_years/#4

***

Alternatively, you can use the Markov-generated values from The Book.

And for those who have the Lahman database, you can use these values:
http://tangotiger.net/bdb/lwts_woba_for_bdb.txt

Mike - Sorry, I didn’t realize that you had been advocating for a change to -45_0_45.  To my knowledge MLBAM stringer data doesn’t provide angular hit ball locations.  Those numbers are created by the analysts that convert MLBAM X and Y pixel hit locations to angular data and distance.  When I wrote about conversion factors to normalize that raw data I think I specifically avoided a specific convention for the horizontal angular data.

I don’t particularly care about what term comes to represent vXY.  If HSOB conveys the concept better that’s fine with me.  I just thought it was a little early for Tango to start suggesting a new term for something where there was not much evidence of confusion.  And then he seemed to want to change the terminology and conventions of every term that was currently being used for Hit f/x.

For most of the studies that I have done using handedness I have been interested in the absolute value of the difference in hit ball horizontal angle from dead center so it has been no more trouble to write a query that looks at ABS(90-HA) than it would be to write ABS(HA).  Personally, I would find the negative numbers of a -45_0_45 system would make my query writing more prone to error without an offsetting benefit.

As long as we don’t end up adopting BIS’s 225_180_135 system which I find totally counterintuitive, I will be happy.

Peter - The weights I use are for pitch analysis, not batted ball. I used the LWBIP values that Tom suggested recently when applying such a concept to batted balls.  Posting my weights here, in this thread, was based on jay’s use of the BA data. I agree, however, for the purposes of Jay’s work, the count should be irrelevant.

Tom/33 - I’m curious as to what approach would be best for coming up with the count-based LWs for pitching analysis. Can you suggest an approach? Or possibly discuss some of the pros/cons of the options we’ve been discussing in this thread? Even a pointer to another thread/article/book chapter would be sufficient.

Peter/19:

For what it’s worth, as an onlooker in this debate, I disagree with you on which scale to use.  To me, the -45 to 45 scale is far more intuitive than 45 to 135.  On the 45-135 scale, if you say a ball is hit at 62 degrees, I really have to think about what that actually means.  “Ok, 62 minus 45, that’s 17 degrees, so that means it’s 17 degrees away from the foul line.  And 90 is the midpoint, so 90 minus 62 gives 28, so it’s 28 degrees from being straight up the middle...”

Why make it so difficult?!  Wouldn’t “-28” be far more intuitive?

I don’t care what scale HitTracker and Sportvision may use.  Respecting their work does not necessitate using their flawed scale.  Why compound something that really doesn’t make sense?

Since my name has been brought up in this discussion, I might as well get in my opinion.  Actually, I don’t believe that Sportvision is locked into any given definition of angles for hitf/x.  They have chosen to make the horizontal angle the angle with respect to the x axis (1B=45, CF=90, 3B=135).  They could have just as easily chosen to make it with respect to the y axis, so that 1B=-45, CF=0, 3B=+45 (or perhaps even the opposite sign).  The issue is not one of coordinate system, it is one of angle definition. 

I actually agree with those of you who like the mirror system, where CF is 0 and 1B/3B have are equal and opposite.  It makes more sense from a baseball analysis point of view.  And it makes absolutely no difference from a “science” point of view.  I also like the terms “launch angle” (for the vertical angle) and “spray angle” for the horizontal angle.

So, if there is a consensus that the mirror definition of horizontal angles is best, we should simply tell Marv White.  It is an easy fix for them.  And the data are not public in the same way that the pitchf/x data are public, so making a change would not (hopefully) cause too much confusion.  In fact, one thing they could do is simply expand the data base by adding a new angle “spray” = “horizontal angle” - 90 (or the negative of that).  That would make everyone happy.

Tango - We disagree, but we have both stated our points clearly so I see no more need for further discussion.

I’d just like to thank Harry and everyone else for helping me figure all this out.

With the new information I am reading, I’m fairly certain Harry has already covered some of the things I’m trying to get at with my charts and he referred to that work in post 35:

http://www.beyondtheboxscore.com/2009/6/9/903793/optimal-launch-speed-and-angle-for

Looking at the moving charts at the bottom of the comments section is particularly mesmerizing, so great work as always Harry.

So just to clarify:

If i’m trying to do LW’s for pitch type analysis, hitter or pitcher perspective, use the LW’s by count.

If i’m trying to do LW’s with regards to base state and outs, use: http://www.tangotiger.net/lwtsrobo.html

And finally for batted ball LW’s, regardless of base state, count and outs, use: 1B = .48, 2B = .78, 3B = 1.06, HR = 1.40, Batter reached Error = .51, batted out = (-.288).

I think that should be correct. Slowly backing away now......

To all: I have updated Harry/3’s chart so that it is readable (formatted to 3 decimal places).  If you need to see his original values, they are visible if you do “view/source” from your browser.

***

Harry/35: I generate LWTS by count, using a Markov pitch-count matrix.  I used the pitches as found on Retrosheet, 1993-2008.

Mike Fast posted his LWTS by count values, which we have combined with Hale’s heat charts here:
http://tangotiger.net/halejon/allcounts.html

Mike then updated them a few times in this thread:
http://www.insidethebook.com/ee/index.php/site/comments/pitch_analysis_of_eric_bedard/

Let me compare Mike’s values to mine, to see if there’s any noticeable differences.  Hmmm… his post 25 is very similar to what I have.  However, he said he had to correct his post 25 because he didn’t handle the two-strike fouls, and so updated it in post 31.  However, I DO handle two-strike fouls, and I get a close match to post 25.  I’ll have to do a bit of investigating because I thought I was pretty meticulous with the data.

I’m also looking at Harry/3 now, and as Peter (?someone else) has noted, the run value for each event (other than ball/strike) should be static across counts.  The run value of the HR is the same, regardless of whether it came from an 0-2 or 3-0 count.

If you were to treat the 0-2 count as its own universe, then you could make the HR value relative to the average run value of that count (say it was -.10 runs), and so, the HR run value would be +1.50.  This perspective does have some limited application.

I do not think this is what Harry is after however.

Alan - If Sportvision changes its terminology, I will use whatever they chose to report any analysis I do on their data.  If you can get Greg and Marv to agree on a common terminology and convention for angular measurement that would be a good thing and I would accept whatever they decide.

OK, I have suggested to Marv White that they define “spray angle” = 90-"horizontal angle”, so that 1B/2B/3B have a spray angle of 45/0/-45.  I like the idea of having the positive angle on the right side, since that conforms to the convention that x is positive on that side.  I will make a similar suggestion to Greg (who will eventually read this thread anyway).

Tango #41 - In Harry’s chart in post #3 the run value for hit events is the ADDITIONAL run value generated from the hit event occuring at that count over the run value that that count has changed from a 0-0 count.  So the run value of a HR at 1-0 is the run value of a HR at 0-0 plus the run value of a ball at 0-0, and the run value of a HR at 0-1 is the run value of a HR at 0-0 plus the run value of a strike at 0-0.  Presenting the data in this way may have value for some analysis (like Harry’s pitch type analysis), but is confusing for what Jay was trying to do.

Re: the run value by count.

Mike Fast shows, in part, the following, at the aforementioned link in post 31:
2-0 0.100
1-0 0.038
0-0 0.000

Peter in the same thread shows this:
Count Total
2-0___0.092
1-0___0.032
0-0___0.000

Basically a match.

If we look at this data:
http://www.insidethebook.com/ee/index.php/site/comments/hitting_by_count/

We see this:
Through Count wOBA BABIP AVG OBP SPC PA
Through 2-0 0.443 0.301 0.290 0.494 0.493 25,514
Through 1-0 0.371 0.300 0.280 0.382 0.459 72,841
Through 0-0 0.332 0.299 0.268 0.328 0.430 175,638

We can convert the wOBA numbers easily enough into runs.  Simply subtract .332, and then divide by 1.15.

So, for the “through 2-0 count”, the .443 wOBA converts to .097 runs, which is smack in the middle of Peter and Mike’s numbers.

The 0-0 is obviously zero.

For the 1-0 count, that’s +.034 runs, and again, right in the middle of Peter and Mike.

My 1993-2008 numbers are +.096, zero, and +.035 respectively.  So, we are all in agreement.

Now, the question is on the 3-2 count.  The wOBA is .403.  That converts to +.062 runs.  My 1993-2008 data says +.060 runs.  Peter and Mike are saying it is a bit below +.020 runs.

So, I would say that my numbers are right. 

If I had to guess as to what the problem is, I would say that they may have treated the two-strike foul as worth 0 runs, which really they should treat it as worth the same as the run value of the count they are in.  Since this is recursive, you can get around it by doing: total runs divided by (1 minus two-strike fouls).  It’s a neat little trick.

Above all, however, since we know what the wOBA is at each count, we need to be able to create an equation that matches to that.

Peter, I’m going to move your post to a related thread.  Look for it in a minute.

Tango Post #35 - I went back and looked to see what I might have done wrong in my calculations on the 3-2 count.  Turns out I multicounted the run values for events that got to a 3-2 count and then had additional foul balls.  Removing those and recalculating I got a value of .059.  Things work best when we check each others work.

I don’t particularly mind rotating the horizontal angle so the zero point is towards CF, my original orientation was selected, I think, so that the sines of the angles would always be positive.  And truthfully, I can’t even remember why that was necessary.  Why I had the numbers increase counterclockwise, I can’t recall either.  I think I did this at the very, very beginning, before I had any idea how far I’d end up going, otherwise I can’t imagine why I did something so counterintuitive…

So, the HitTracker horizontal angle system is the quintessential “cowpath” that eventually ended up becoming Main St.  Or at least one block over from Main St.

One advantage to switching the angles so that +45 is 1B and -45 is 3B is that this will conform more naturally to the way wind directions are expressed, which is 0-359 degrees counting clockwise, with 0 degrees at true north.  Well, actually you usually just get NW, SE, SSW, etc., but you know what I mean: this is the direction convention used in navigation worldwide.  East is 090, South is 180, etc.

I think HitFx will eventually incorporate, or at least connect to, weather information in some way, so it would be a good idea to set it up to do so in the most natural way.

In any event, there’s no way I can convert Hit Tracker in the near term, so it will have to stay the way it is, at least until the off-season, and perhaps until I turn it into a stand-alone application.

I also prefer a cf=0 mirrored system, and have been using that in my own data for a couple years.

These were mine for 2008..am I close to what everyone else has:

0-0 0.00
1-0 0.036
2-0 0.099
3-0 0.208
0-1 -0.042
1-1 -0.015
2-1 0.032
3-1 0.131
0-2 -0.099
1-2 -0.078
2-2 -0.039
3-2 0.054

My 3-2 count seems a bit off.

FWIW, I also used a cf=0 system for my own data. I’m relieved that it seems to be the consensus here.

While we’re at it we might as well discuss other HITfx properties - namely the orientation of the bat on contact with the ball.

It seems like “tilt” is already in use to describe the levelness of the bat. I would suggest that 0 deg be a perfectly level bat and that negative numbers would describe the end of the bat tilting toward the ground.

The other bat angle we need to describe is where in the swing the bat is on contact. I’m not even sure what to call this. pull angle? Anyway, I would also propose that the value is zero when the bat it perpendicular to the cf=0 axis. positive numbers would describe a pull swing and negative numbers an inside-out swing.

I’m not sure if the signs should be reversed for a opposite handed batter or not, but it might be nicer if negative pull angles tended to produce negative spray angles and positive tended to positive. In that case a positive pull would be for left handed batters and righties would be reversed.

This may seem pretty esoteric, but I think swing analysis will become important as our understanding of HITf/x improves.

John - The numbers that you are showing for count state linear weights are very close to the revised numbers that I got after I rechecked my calculations for 2004-2006.  Several things that I discovered on my rechecking.  I had not removed intentional walks as I had intended to do, so I did that.  I also had to remove non-batting events, since I was making my calculation from Retrosheet data and leaving them in would count their count states twice, and I also wanted to have the count state linear weight only represent batting event runs.  These two changes made some very minor changes in the count state linear weight values, but they also had the effect of making the count state linear weight value for 0-0 count not actually equal to 0 anymore.  Mostly because I had eliminated intentional walks the 0-0 count state was now equal to -.004.  Since I wanted the count state linear weights for other counts to be relative to a 0-0 count state of 0, I had to add .004 to all the count state linear weights. 

As for your 3-2 count state value.  I found that my value for 3-2 changed when I took into consideration that I needed to limit counting 3-2 counts to 1 per at bat.  You actually need to limit counting occurences of any particular count to 1 per at bat, but the particular method that I was using for calculating count linear weight values had already eliminated the double counts for every count except 3-2.  When I eliminated the double counts of the additional 3-2 counts my value for 3-2 rose to the .006 range. 

I hope this information helps.

Thanks.  That was extremely helpful.

So for my 3-2 count, in other words, I need to eliminate any two strike fouls I guess.

"rose to the .006 “

Peter must mean .060, to be consistent with other posts on the matter.

***

John: right.  We don’t want to count that you passed-through the 3-2 count twice on a foul, since that event is really a “let” or mulligan, for our purposes here.

Tango - Yes, .060. Misstyped again,. Thanks for correcting that.

If I had to guess as to what the problem is, I would say that they may have treated the two-strike foul as worth 0 runs, which really they should treat it as worth the same as the run value of the count they are in.  Since this is recursive, you can get around it by doing: total runs divided by (1 minus two-strike fouls).  It’s a neat little trick.

I don’t have time to recalculate my LW right now, but I agree with your identification of my mistake and your solution.

Maybe I’m missing something here, but what I’m after is the change in run expectancy based on the outcome of a pitch. If I foul off an 0-2 pitch, my RE has not changed, right? Therefore the runvalue is 0, as RE2 = RE1

BTW, I’ve been rebuilding my weights, starting with wOBA by count (Tippet’s data is from 2003, that’s part of the problem).  I’ll try and share those sometime today

Harry: if you want to say “GIVEN that I’m in a 0-2 count, then the HR adds +1.5 runs” and that “GIVEN that I’m in a 3-0 count, then the HR adds +1.2 runs” to the state I’m currently in, then fine.  Is this EXACTLY the question you are asking?  If not, then what is exactly the question you are asking?

As for the Tippett data being from 2003, note that I’m using 1993-2008, and my wOBA (relative to the 0-0 count) numbers are very close to his.  I don’t see this as a “problem”.

Maybe I’m missing something here, but what I’m after is the change in run expectancy based on the outcome of a pitch. If I foul off an 0-2 pitch, my RE has not changed, right? Therefore the runvalue is 0, as RE2 = RE1

Harry - That’s exactly why you have to ignore two strike fouls when calculating the run value of a particular count.  Including them would double count the event value for that AB since you already included it the first time that count appeared.

Tango/58 - Yes, that’s what I’m getting at with my pitch value analysis.

Harry/#60, that may be your ultimate question (as it was mine) and the answer for that is indeed that the two-strike foul has a value of 0. 

However, when you are calculating the value of the 0-2, 1-2, 2-2, and 3-2 counts, you need to exclude two-strike fouls from that calculation (or really, value them at the value of that count, which has the same net effect, as Tango indicated).

Harry, ok.

Then the only thing you need to ensure is that the HR run value at countX plus the average run value of all events at that countX is equal to 1.4 for every count.  Is that what you’ve got?

Maybe an example would be helpful here.  I’ll make some simplifying approximations for linear weight values of the various events, but hopefully the point will still be clear.

Let’s say our hypothetical league saw 100 pitches at a 3-2 count.  The results were as follows:
35 fouls (not caught)
30 strikes
10 balls
20 ground outs
5 home runs

The 30 strikes and and 20 ground outs have the value of an out, which we’ll approximate at -0.3 runs, for a total of -15 runs.

The 10 balls have the value of a walk, which we’ll approximate at +0.35 runs, for a total of +3.5 runs.

The 5 home runs, valued at +1.4 runs each, total +7 runs.

The 25 foul balls return us to the 3-2 count, which currently has an unknown value.

So, in order to find that unknown value, we will ignore the foul balls for the moment (and check later if that worked).  So far we have a total of -15 + 3.5 + 7 = -4.5 runs.  Divide that by 65 pitches to get a value of -0.0692 runs per pitch for the 3-2 count.

Now add back in the 35 foul balls, valued at -0.0692 runs each, for a total of -2.42 runs.  Now add up the run totals -15 + 3.5 + 7 - 2.42 = -6.92 runs.  Divide that by 100 pitches to get a value of -0.0692 runs per pitch for the 3-2 count.

So we see that when calculated the per-pitch run-value for a 2-strike count, you should ignore the foul balls.

Tom/62 - I should, and I think I do, I don’t have all my numbers with me (I emailed myself some files so I’d have them w/ me at work, but I missed one). I won’t be “done” until that’s the case, though.

Mike/61 - I’m calculating “through” count values as such:

1) I built a view that has the count, the ab_id and the event for the ab (not the pitch, each pitch in the ab becomes associated with the final event of the at bat)
example:
count,abid,event
0-0,123,HR
0-1,123,HR
1-1,123,HR
1-2,123,HR

Say, in this example, the hitter fouled off 3 pitches at 1-2 before going yard. I don’t care, since I’ve grouped by the count and ab_id, repeats of the same count are collapsed that way.

2)Once I have all the outcomes through each count in each ab, I calculate the wOBA for that count, based on all the events

2a) I’m calculating lwOBA too, and the 0-0 wOBA matches the lwOBA exactly. I’m not counting intentional walks when determining the value of a walk, but I am including IBB for the total # of PAs (is that wrong? Should I drop it altogether?)

So, is this approach correct? Am I properly dealing with 2 strike fouls by doing it this way?

Another question - what is the base value of a walk , HBP or a strikeout (we’ve agreed on the batted ball values to use, but not those two non-batted ball events)?

Mike/63 is correct.

Harry would take the +1.4 for the HR subtract the -.07 runs in Mike’s illustration, to get +1.47 for the HR, GIVEN that you are in a 3-2 count.

So, Harry ends up like this:
HR +1.47
BB +0.42
out -0.23
foul 0

1.47*5 +.42*10 -.23*50 +0*35 = 0

And this is what Harry wants.

Say, in this example, the hitter fouled off 3 pitches at 1-2 before going yard. I don’t care, since I’ve grouped by the count and ab_id, repeats of the same count are collapsed that way.

Yes, you are doing it correctly.  Grouping by count is the key part there.

Harry, exclude IBB altogether.

Another question - what is the base value of a walk , HBP or a strikeout (we’ve agreed on the batted ball values to use, but not those two non-batted ball events)?

Do you mean in wOBA?  I use 0, .7, .9, 1.3, 2.0 for out, BB/HB, 1B, 2B/3B, HR.  As for distinguishing between K and a non-K out, it’s not worth the trouble, as the difference is around .01 runs, and as you can see by my list above, having things to one decimal place is appealing.

The wOBA coefficients year-by-year, for use in the Lahman database is here:
http://tangotiger.net/bdb/lwts_woba_for_bdb.txt

(Go to the far right of the page.)

As you can see, the HR coefficient has hovered around 2.0 in the current 1993-present era.  You could try to make it perfect for each season, or you can simply go with the quick coefficients described in the previous post.

I guess I can get there from wOBA, but I’m talking about the LW, such as these noted by Jay in an earlier commment

1B = .48, 2B = .78, 3B = 1.06, HR = 1.40, Batter reached Error = .51, batted out = (-.288)

So, any out I’ll keep @ -.288, but what value should I assign a walk/hbp? I’m thinking it’s around .35 or so.

You might be talking about this chart Harry which was referenced in posts 29 and 33:

http://www.tangotiger.net/customlwts.html

Harry, the link in post 68 also shows you the LWTS values for each event.

Thanks, I don’t know how I missed that!

I really appreciate the feedback and help, going back into the cave to emerge tonight with an update, probably at Cubs f/x. I’ll drop a link over here in any case.

http://www.cubsfx.com/2009/06/updating-linear-weight-values-by-count.html

OK, get our your red pens and hole punchers.

Thanks again for everyone’s input. There is a very impressive group of people loitering around this joint.

I have put both the vertical and horizontal components together with a basic linear weight at BtB.

http://www.beyondtheboxscore.com/2009/6/18/913774/zones-of-scoring-using-hit-f-x-to#comments

I get similar drops like Jay sees for IF positions along with seeing what angles the IF doesn’t matter and the outfield has to play the ball. 

Next I will be dividing the data up and creating multiple charts.  I am thinking of doing 3 with < 75, from 75 to 95 and greater than 95 to keep the number of hits somewhat even.

Harry: if you haven’t, you can consider sorting by wOBA.  It’ll make the chart easier to read I think.

Your 3-0 wOBA looks a bit low to me.  Though maybe I’m thinking that one of my old charts didn’t remove the IBB.  I’ll have to check that.

Jeff, great job!

As you noted, a breakdown by RHH and LHH is essential.  It’s not the absolute degrees, but the relative pull/opposite degrees that we care about.  So, you can either show two charts, one for RHH and one for LHH.  And, if they look symmetric, you can merge them by taking the negative angle for the LHH and then adding them to the RHH.  This would mean that “-45” means “pulled down the line” as opposed to “pulled down the 3B line”.

Next I will be dividing the data up and creating multiple charts.  I am thinking of doing 3 with < 75, from 75 to 95 and greater than 95 to keep the number of hits somewhat even.

Jeff, hopefully what you find lines up with what I posted in the comments here:

http://www.insidethebook.com/ee/index.php/site/comments/batted_ball_distance_of_ortiz_is_way_way_down/#20

Harry: I have the 1993-2008 wOBA on a 3-0 count at .569 (based on 150,000 PA), and 3-1 at .496, and 2-0 at .450.

I think I excluded all pitchers-as-batters, and batters-as-pitchers PA, not that it would really matter much.  So, there is a sizable gap between the three, consistent with the Tippett data.

I did not exclude anything actually.

#77 Mike,

Did you pick <55, 55-80, and >80 for a reason?

Jeff/#80, the reason I chose those values is because they were the approximate inflection points in the graph in Post #18:

If you hit it less than 55 mph (as measured in the plane going up at 11 degrees), it has a >90% chance of turning into an out, unless you hit it so slowly that the infielders can’t get to it in time. (That graph might include bunts--I don’t remember.)

If you hit it between 55-80 mph, your chances at a hit aren’t dependent so much on how hard you hit it as whether you can get it over the infield and not hit it right at an outfielder.  If you put it on the ground, the infielders will get to it whether it’s hit at 55 mph or 80 mph.  Similarly, if you loft it high enough in the air (>30 degrees or so), the outfielders will get to it whether it’s a shallow fly or deep fly.

If you hit it harder than 80 mph, the harder you hit it the better your results will be.  At this speed, you have a significant chance of poking a ground ball between the infielders, and your fly balls find the gaps or the wall a little quicker.  You’re trying to get it on the outfield grass before a fielder can get to it, so every additional bit of horizontal speed improves your chances.

Btw, these are generalizations based on the chart above and other one in Post 20 of the Ortiz thread.  I might be wrong about a few of the conclusions I’ve drawn from them.

Mike,

Thanks, I think I will actually look at 4 points (all depending on sample size)
< 50 (5% hit chance)
50 to 80 (20% hit chance)
80 to 95 (ramp of 20% to 75% hit chance)
95 and above (75% hit chance)

Here’s what I put in the 2008 THT Annual for data on Torii Hunter and Andruw Jones:

Striking similarity, no?

After Tom noted some numbers were off, I re-checked my queries and found I was under-counting plate appearances. Here are the updated wOBA by count values. My numbers are little higher than Tom’s, but I’m only using 08-09 Gameday data.  Clearly, I can’t understate the value of sharing your work and its impact on the quality of the final product.

count wOBA
0-0 0.332
0-1 0.281
0-2 0.212
1-0 0.378
1-1 0.319
1-2 0.245
2-0 0.454
2-1 0.378
2-2 0.295
3-0 0.585
3-1 0.498
3-2 0.408

I’m not sure if I posted these somewhere, but here are mine for 1993-2008:

B S wOBA
3 0 0.569
3 1 0.495
2 0 0.449

3 2 0.407
2 1 0.380
1 0 0.379

0 0 0.339

1 1 0.322
2 2 0.297
0 1 0.290

1 2 0.248
0 2 0.220

Harry and Tom, has anyone figured out the run value of a double play on average (averaging across all base/out states)? 

It seems to me that my rv and rv/100 analysis undervalues extreme ground ball pitchers...it takes into account the lack of home runs they give up, but not the large amount of double plays (valuing them the same as single outs, when they’re clearly worth more.) Given that different pitchers (such as extreme ground ball pitchers vs extreme fly ball pitchers) can have consistantly different GIDP (and other forms of DP) rates, I was wondering if anyone had found the proper run value for these outs so i could incorporate them into an analyis (for example, of Met Mike Pelfrey).

garik - GDP run value 2005-2008 is -.845.  Adjusted ground balls outs without DP is -.246.  Air ball outs -.28.  You may also want to include value for infield single, .40.  Value for outfield GB single .48.  Value of Outfield FB or LD single .47.

This is pretty easy to figure, if you look at the RE chart:
http://www.tangotiger.net/RE9902.html

Let me know how much experience you have in those charts to figure out the run value of a single, or HR, etc, and then we can walk you through the DP value.

I have a small amount of experience with these charts (I actually was looking through this stuff and then just finished The Book last week because i was looking to do more analysis).

I could figure out how to compute for each state the value of a DP.  That said, I don’t know where to find data on the amount of DPs per each base/out state in a year so as to properly average the thing…

In other words, I can compute the RV of a DP with a man on first and no outs (1 out is easy) but i don’t know how to properly average that with the results from other states without the ratios of the occurance of these events (I’d imagine most DPs occur in a no out or one out X _ _ base state followed afterwards by XX_ base states, where XX_ = men on first and second). 

Help would be much appreciated. 

Thanks a lot.