How Many Pitches _Should_ Pitchers Have?

[RonCo]

Just because I needed to put it someplace, here's a breakout of pitchers FB velocity, frequency, and linear weights quality based on role (SP=starter only, SW=swingman, R = Relief Only) and selected on having thrown at least 300 pitches in games:

Descriptive Statistics: FBv, FB%, wFB/C (300+ Pitches)

Variable  Role    N  N*     Mean  SE Mean    StDev  Minimum   Median  Maximum
FBvel     SP     135   0   90.711    0.204    2.376   83.900   91.000   95.000
          SW      98   0   90.378    0.293    2.902   72.900   90.800   96.000
          R      219   0   92.248    0.175    2.596   83.800   92.600   98.000

FB%       SP     135   0  0.56990  0.00735  0.08538  0.23200  0.57400  0.74200
          SW      98   0   0.5600   0.0145   0.1438   0.0240  0.58950  0.84300
          R      219   0  0.61289  0.00898  0.13292  0.07100  0.62600  0.86700

wFB/C     SP     135   0  -0.1584   0.0872   1.0130  -4.4100  -0.1900   1.8300
          SW      98   0   -0.341    0.151    1.491   -7.540   -0.095   2.1400
          R      219   0   0.2757   0.0843   1.2476  -4.4200   0.3000   3.7400

[RchW]

Originally Posted by TribeFanInNC

Nice analysis Ron. I agree with your premise - the game doesn't really feature enough pitches. It has always seemed to me that pitchers in OOTP don't have enough fastballs in their repertoire. I would think a lot of pitchers show more than one movement on their fastball (like a 4-seamer and a 2-seamer).

I always have trouble relying on pitches thrown data though. Justin Masterson is a perfect example. Sure by pitch tracking, he throws 90% fastballs. But his fastballs have at least 3 distinct movements - a more straight 4-seamer he uses up in the zone, a tailing fastball that runs away from lefties and back doors righties, and a sinker that dives down in the zone.

I was going to post something similar. Many really hard throwers have to be coached into taking something off the fastball to improve command and generate some movement. A lot of pitchers talk about throwing two change ups; one harder and up in the zone and one slower and in the dirt if necessary.

OOTP has "sinker" as a stand alone pitch. I have only ever thought of (and thrown) it as a fastball and when it didn't sink it got hit hard.

[Nutlaw]

I wonder if many of the two-pitch relievers were LOOGYs? If a pitcher throws regularly to hitters from both sides of the plate, he'll usually need a fastball and then an off-speed pitch for each type.

[OutS|der]

part of the problem i think is allot of the 2 pitch pitchers are lacking an actual rating at fastball, they all have a "fastball type" pitch but lack the rating for the fastball but if they had that rating since all throw fastballs that would eliminate allot of the 2 pitch pitchers

[ike121212]

Originally Posted by RonCo

Just because I needed to put it someplace, here's a breakout of pitchers FB velocity, frequency, and linear weights quality based on role (SP=starter only, SW=swingman, R = Relief Only) and selected on having thrown at least 300 pitches in games:

Descriptive Statistics: FBv, FB%, wFB/C (300+ Pitches)

Variable  Role    N  N*     Mean  SE Mean    StDev  Minimum   Median  Maximum
FBvel     SP     135   0   90.711    0.204    2.376   83.900   91.000   95.000
          SW      98   0   90.378    0.293    2.902   72.900   90.800   96.000
          R      219   0   92.248    0.175    2.596   83.800   92.600   98.000

FB%       SP     135   0  0.56990  0.00735  0.08538  0.23200  0.57400  0.74200
          SW      98   0   0.5600   0.0145   0.1438   0.0240  0.58950  0.84300
          R      219   0  0.61289  0.00898  0.13292  0.07100  0.62600  0.86700

wFB/C     SP     135   0  -0.1584   0.0872   1.0130  -4.4100  -0.1900   1.8300
          SW      98   0   -0.341    0.151    1.491   -7.540   -0.095   2.1400
          R      219   0   0.2757   0.0843   1.2476  -4.4200   0.3000   3.7400

Interesting thread. I really hope it leads somewhere productive.

Number of pitches and development aside, I hope the fact that there are 98 swingmen in your stats isn't lost. In OOTP, they are exceedingly rare. Bullpen's are stronger than rotations. If you someone can't crack the rotation, they usually end up in AAA, not the bullpen.

I'd love to see a patch that gave us greater control over the types of pitchers generated, or if someone had the ambition to write a utility to manipulate the draft pool to get there.

[RonCo]

I'm looking at lots of weird math right now and I'll post if things get interesting...but for now here's a small bit of "obvious" data.

I took all pitchers who have thrown at least 30 IP, split them into groups based on how many pitches they throw at least 5% of the time. Then I averaged their fastball velocity, quality, and usage frequency:

N5%	Vel	wFB/C	FB%
---	---	-----	---
2	92.47	+.661	64.8
3	91.82	+.098	60.6
4	90.69	-.113	54.6
5	89.70	-.084	48.4

Bottom line here seems to be that If a guy doesn't have a good fastball he tends to throw more pitches. If a guy has a good fastball, he uses it quite a bit more often.

As I said, obvious. At least it's good validation of the data.

[Pdubya64]

Nice analysis RonCo. It also mirrors what our resident Major Leaguer Curt had to say about the # of pitches issue.
Reading through this got me thinking- what if part of the problem lies with ALL the fastballs being lumped together in one rating? As a couple others have mentioned, there are enough types of fastballs that it may prove statistically relevant to break out some of them (Fastball Primary)/(Fastball Secondary)/(...) due to the fact not all fastball types are created equal when it comes to speed and movement.

I imagine this creates its own set of problems that would have to be worked through to program properly.

[RonCo]

Originally Posted by OutS|der

part of the problem i think is allot of the 2 pitch pitchers are lacking an actual rating at fastball, they all have a "fastball type" pitch but lack the rating for the fastball but if they had that rating since all throw fastballs that would eliminate allot of the 2 pitch pitchers

This is a problem, but maybe not in the way you're thinking. Since OOTP views a sinker as a what you're terming a "fastball type" (which is how it would appear to this data I'm working with), but separate pitch type, you see two-pitch pitchers in OOTP with a fastball and a sinker (here's one: Player Report for #1 Shinji Watanabe).

The problem is that, per the data I'm working with, a real pitcher with this repertoire would show up as a 1-pitch pitcher (of which, there are obviously none per the 2011 MLB data).

[RonCo]

Originally Posted by Pdubya64

Nice analysis RonCo. It also mirrors what our resident Major Leaguer Curt had to say about the # of pitches issue.
Reading through this got me thinking- what if part of the problem lies with ALL the fastballs being lumped together in one rating? As a couple others have mentioned, there are enough types of fastballs that it may prove statistically relevant to break out some of them (Fastball Primary)/(Fastball Secondary)/(...) due to the fact not all fastball types are created equal when it comes to speed and movement.

I imagine this creates its own set of problems that would have to be worked through to program properly.

The fact that all fastballs are lumped together in this fangraphs data is still fine by me. It does cause some issues, but this data basically says "in addition to all the different types of fastballs one might have, you also will find pitchers throwing at least 2-3 other fundamentally different kinds of pitches."

As I noted in my last post, the fact that OOTP separates these fastballs and treats them as separate entities means it's actually making the situation worse (meaning less realistic).

[Pdubya64]

Originally Posted by RonCo

The fact that all fastballs are lumped together in this fangraphs data is still fine by me. It does cause some issues, but this data basically says "in addition to all the different types of fastballs one might have, you also will find pitchers throwing at least 2-3 other fundamentally different kinds of pitches."

As I noted in my last post, the fact that OOTP separates these fastballs and treats them as separate entities means it's actually making the situation worse (meaning less realistic).

Hmmm...

Color me confused. Since you are closer to the data RonCo, what do you make of it in terms of how OOTP might treat pitches? Seems obvious to me that things are a bit off right now. But I don't mess with it enough to make more than very opinionated observations.

[RonCo]

Don't get me wrong. I don't know that all those fastballs should be grouped or not grouped. I'm more commenting on the applicability of the data I've got, which suggests pitchers have 3 or more pitches, not counting all the different fastballs they may have off the basic.

[RonCo]

Originally Posted by Pdubya64

Hmmm...

Color me confused. Since you are closer to the data RonCo, what do you make of it in terms of how OOTP might treat pitches? Seems obvious to me that things are a bit off right now. But I don't mess with it enough to make more than very opinionated observations.

You ask how should OOTP handle pitch types. My answer is as follows (sorry if this is a bit of a bore, but what the heck, if you're not interested in design neep that will never come about, just stop here. Seriously. It's not worth it. [grin]):

1) Each pitch should be rated for velocity, control, and movement (or command?).
2) Each pitch should contribute to a repertoire
3) The quality of a repertoire defines performance and how long they can be effective
4) A pitcher's stamina should be dynamic and keyed to their physical make-up and maybe mechanics (and almost every pitcher should be physically capable of throwing 90+ pitches.

Beyond that, there's a ton of design work.

But to give an idea wof what that might look like, this fangraphs data lets use pitch quality (based on linear weights wXX/C) to discuss how a repertoire's quality might be assessed.

To begin with, we probably need to map a baseline performance quality for the league in question for K-rates, BB-rates, and HR-rates among others (OOTP does this already, though the mechanism may need to change). These will be our baseline performance rates that wil get adjusted by a pitcher's repertoire.

Then let's look at pitch quality and its influence. We'll use a simple two-pitch repretoire of fastball/slider for simplicity's sake. First, let's find out how much the quality of a pitch should change a pitcher's performance.

To do this, I took all pitchers with 30 IP and ran correlation tests on the pitch quality values (wFB/C and wSL/C) on rate performance:

	Fastball	Slider
Stat	r	p-value	r	p-value
K/9	.377	.000	.207	.000
BB/9	-.199	.000	-.019	.742
HR/9	-.511	.000	-.270	.000
HR/FB	-.503	.000	-.216	.000

This suggests the fastball rating might influence all performance rates and the slider rating influences K and HR rates, but doesn't do much to BB rates beyond random. In all cases, the fastball's influence is 1.5-2.5 times greater.

This means that in OOTP performance parlance,

"Stuff" = K/9 Rating = Baseline + Fastball Adjustment + Slider Adjustment
"Control" = BB/9 Rating = Baseline + Fastball Adjustment
"Movement" = HR/9 Rating = Baseline + Fastball Adjustment + Slider Adjustment

I would also probably then go through an adjustment process to take into account repertoire size. There are several ways to do this, and I would need to do more number crunching to make it happen well.

so, that's a high-level look at how this data and pitch quality could/should influence ratings. However, the real magic probably happens at a level lower, as more granular data matches together to create intertwining influences across pitch types--for example, "how should velocity affect performance?", or specifically, how should each pitch be rated, and then how whould those ratings be converted to Plate-appearance-based results? The data I've been playing with could be used to answer these questions. Here's a framework for how it could work...

Let's look at the effect of velocity on a pitcher's K-rate, BB-rate, and HR-rate (the three main elements of the OOTP DIPS-based results engine). I'll use the fastball and slider for now to keep it moderately simple. I ran a correletion test on pitcher's FBvelocity and SLvelocity to their performance rates and got the following (please realize this just scrathes the surface of the design process):

Fastball:

	r	p-value
wFB/C	.291	.000

This says that velocity (alone) should adjust the quality of a fastball by a moderate amount.

As a test base, if this is doen right, FBv influence on FB quality should result in the following kinds of correlations...so you tweak these equations until you get something in these ballparks.

Stat	r	p-value
K/9	.421	.000
BB/9	.221	.000
HR/9	-.279	.000
HR/FB	-.167	.000

Velocity should influence a pitcher's BB-rate downward about twice the amount it adjusts the K-rate upward. Likewise, a pitcher's fastball velocity should reduce home runs at roughly the same rate as it adjusts walks.

Slider

	r	p-value
wFB/C	.061	.288

This says that velocity (alone) should not adjust the quality of a slider by much if any (because the .288 p-value says the correlation is not statistically significant.

Same test baseline for vSL and it's invluence on performance:

Stat	r	p-value
K/9	.138	.015
BB/9	.010	.816
HR/9	-.172	.002
HR/FB	-.105	.064

On the other hand, the velocity of a slider works about like that of a fastball, except note that the BB and HR influence is of questionable significance (the control value says it's clearly not significant). So slider velocity does seem to adjust the K-rate directly, and maybe you give velocity of the slider a very small adjustment on HR-rate...in other words, you'll need to get more data and or play with it a bit in testing.

Note that in the case of multiple fastball types (or multiple changes or curves or whatever), I could see a situation where the algorithm rates each individually, then merges them together in some fashion to create a multiplicative index that leverages them all together into a greater whole.

Admittedly, this is a complex discussion for a message forum.

[RonCo]

And semi-finally:

1) Each pitch should be rated for velocity, control, and movement (or command?).
2) Each pitch should contribute to a repertoire
3) The quality of a repertoire defines performance and how long they can be effective
4) A pitcher's stamina should be dynamic and keyed to their physical make-up and maybe mechanics (and almost every pitcher should be physically capable of throwing 90+ pitches.

In addition to the above:

5) Almost all pitchers should get 3+ pitches (or, 1 or more fastballs and 2+ others)
6) Pitchers should be selected for roles based basically on a "best pitchers start," with a leaning toward the best three-pitch repertoire.
7) Bullpen use should be designed to get the pitchers with the best two-pitch repertoire (with other pitches in the weighting, but of smaller affect) the most work, with the possible exception of the closer, which should be chosen based on two pitch repertoires slanted toward guys with big fastballs. The AI should not need a SP/MR/CL assignment to do its work.

[ike121212]

Originally Posted by RonCo

And semi-finally:



In addition to the above:

5) Almost all pitchers should get 3+ pitches (or, 1 or more fastballs and 2+ others)
6) Pitchers should be selected for roles based basically on a "best pitchers start," with a leaning toward the best three-pitch repertoire.
7) Bullpen use should be designed to get the pitchers with the best two-pitch repertoire (with other pitches in the weighting, but of smaller affect) the most work, with the possible exception of the closer, which should be chosen based on two pitch repertoires slanted toward guys with big fastballs. The AI should not need a SP/MR/CL assignment to do its work.

Where can I pre-order your game?

[OutS|der]

Originally Posted by RonCo

And semi-finally:



In addition to the above:

5) Almost all pitchers should get 3+ pitches (or, 1 or more fastballs and 2+ others)
6) Pitchers should be selected for roles based basically on a "best pitchers start," with a leaning toward the best three-pitch repertoire.
7) Bullpen use should be designed to get the pitchers with the best two-pitch repertoire (with other pitches in the weighting, but of smaller affect) the most work, with the possible exception of the closer, which should be chosen based on two pitch repertoires slanted toward guys with big fastballs. The AI should not need a SP/MR/CL assignment to do its work.

I agree and I'd like to add one more addition

8) players should get rated on their abilities to hit the different pitches and the bullpen use should reflect that with breaking ball pitcher used over a fastball one based on the batters skill against that pitch

[Cryomaniac]

Originally Posted by RonCo

Admittedly, this is a complex discussion for a message forum.

It is, but I for one appreciate that you take the time to do things like this, even though they have a negligible chance of being implemented (although if Markus does ever re-write the engine, I sure as hell hope he reads your posts first).

[Hazza]

Just my observations from some light research:

Pitches used is not random, everyone had fastball, a breaking ball (or two) and an offspeed pitch. Ergo every starter should be created with a fastball, breaking ball and an off speed pitch.

It seems (eyeballing) that only 19 pitchers don't use a slider of some sort, only six don't use a change up, the rest use sinkers. 37 (about a third) use the slider as a primary breaking ball (used it > 5% more than the curveball). And most of the pitchers use the breaking ball more than the change up, (but obviously there are some exceptions, like Felix Hernandez).

I think that if I was doing this I would calculate the pitching ratings based on K/9 BB/9 and HR/9 as per usual. I would then use the pitch ratings to determine the frequency of the pitch, and then use the in built scouting to give me color ie: "Throws a great 2-seam fastball", "Has a plus breaking ball (slider)"

That way you can program the pitcher to use the right pitches the right amount of time in game, and over all they should spit out the right number of ks/bbs/hrs over the course of a year, which should be the goal of the game because I think making it show that each game each pitcher gets the right number of ks per pitch might be a bit too minute.

Sorry if something here was blindingly obvious, or blindingly incorrect.

[ike121212]

I just hope Markus is reading this and gives us something in the last patch, however small, to improve this. Just a little more control over pitcher generation would go a long, long way.

[injury log]

Originally Posted by RonCo

You ask how should OOTP handle pitch types. My answer is as follows (sorry if this is a bit of a bore, but what the heck, if you're not interested in design neep that will never come about, just stop here. Seriously. It's not worth it. [grin]):

1) Each pitch should be rated for velocity, control, and movement (or command?).
2) Each pitch should contribute to a repertoire
3) The quality of a repertoire defines performance and how long they can be effective
4) A pitcher's stamina should be dynamic and keyed to their physical make-up and maybe mechanics (and almost every pitcher should be physically capable of throwing 90+ pitches.

Beyond that, there's a ton of design work.

Originally Posted by OutS|der

I agree and I'd like to add one more addition

8) players should get rated on their abilities to hit the different pitches and the bullpen use should reflect that with breaking ball pitcher used over a fastball one based on the batters skill against that pitch

There's a point where adding granularity no longer adds to the user experience. I mean, OOTP could, in theory, calculate the velocity, release point and spin (etc) of every pitch, and the swing plane, bat speed (etc) for every swing made, and via physical modelling work out the result of each at bat. I don't see what that would do for the user that the current game fails to do.

So my question is: if the game is going to add new ratings for individual pitches, or for a batter's ability to hit each pitch type, what gameplay problem does that solve? If the goal is to move to a pitch-by-pitch results engine, I'm sure you both appreciate how absurdly complicated the math gets if you want to maintain the statistical integrity of the game output.

As for RonCo's initial post, about the distribution of pitches among MLB pitchers, I've been running a few tests recently to see how OOTP would work if it created more 3-pitch pitchers. I ran one league where I only allowed myself to acquire (in draft, trade, etc) pitchers with three pitches and the stamina to start. Many of these pitchers evolved into pure relievers by the time they reached the Majors (they either lost Stamina, or never adequately developed their third pitch), some became obvious rotation candidates (because they were good), and some ended up in a certain role because that's where they first had success. The decisions I had to make about pitcher usage were far more interesting, and seemed more realistic, than when I drafted a mix of 2-pitch and 3-pitch guys. An interesting consequence of playing in this way: star relievers still exist, even without needing to generate a lot of star RPs in the draft pool. I'd have guys the game considered 0.5 or 1-star pitchers in my bullpen (rated that low because the stars are based on comparisons with the rest of the league, and the rest of the league was full of those 5-star RPs you see in the draft pool), and I'd still have guys posting ERAs of 2.50 every year in a league with an average ERA of 4.00.

Still, the game engine doesn't yet properly support that kind of player creation model. It would happen in my league that I'd use a guy as a reliever for five years, he'd leave as a free agent and immediately become an SP. That's not a common occurrence in real life. So for the game to work in this way, it needs to be more aggressive about cementing pitchers into roles; pitchers need to lose Stamina more often through injury so some are forced to be relievers (e.g. Kerry Wood), and they need to lose third and fourth pitches when not used as starters for a long time (since no thought would ever be given to switching most real life veteran relievers to a starting role, Braden Looper excepted).

In beta, I've suggested the following changes be made to the pitcher creation and development model:

* in the draft pool, all High School pitchers should be generated with at least 3 pitches (though often with one pitch with very poor potential), and with at least the minimum Stamina to start. That is, every HS pitcher should appear to have at least a chance to become an SP.

* College pitchers should sometimes be 'pure relievers', though the number of obvious 'superstar' relievers in the draft pool should be very small.

* the development engine should be much more aggressive about developing individual pitches. So a guy with 5/5/1 pitch ratings should sometimes, because development would be laser-focused on individual pitches, turn into a 9/5/1 guy, and thus a reliever, and sometimes turn into a 5/5/5 guy, and thus a starter

* players would need to often become cemented into roles. So Stamina would need to be affected by serious injury, independent of other ratings, forcing guys into relief. Further, a guy not used as a starter would need to lose the ability to throw a wide repertoire. These things happen already in the game, but not to the degree required.

* Pitchers with elite Stamina should probably always be able to start, regardless of their repertoire. The game fails to model two types of pitcher now - the guy who spends his career as a swingman (Mark Hendrickson, say), and the guy with very poor Stuff who can still start (Nate Cornejo, Carlos Silva). This might also allow the occasional 2-pitch starter to exist.

* On draft day, one factor the user would need to consider is the risk of a pitcher not developing into a starter. This creates interesting decisions: pick a guy with borderline Stamina who might be one injury away from becoming a reliever, or pick a guy with only two good pitches and one rudimentary one who might not start, or instead pick a lower-rated guy who is more certain to become an SP. Real life teams seem to grapple with this issue all the time, and it seems to me the reason why 'polished' college SPs (Deck McGuire, Mike Minor, etc) are premium picks on draft day.

______


I think those changes would address a lot of gameplay issues without requiring any fundamental revision to the game engine. Nothing will happen until next version of course, but based on our discussions in beta, I think there's a reasonably good chance some changes will be made to this area. The main difficulty is that even a small change here would require a lot of testing to make sure the game still produces realistic stats output and career arcs, etc.

[ike121212]

Originally Posted by injury log

Still, the game engine doesn't yet properly support that kind of player creation model. It would happen in my league that I'd use a guy as a reliever for five years, he'd leave as a free agent and immediately become an SP. That's not a common occurrence in real life. So for the game to work in this way, it needs to be more aggressive about cementing pitchers into roles; pitchers need to lose Stamina more often through injury so some are forced to be relievers (e.g. Kerry Wood), and they need to lose third and fourth pitches when not used as starters for a long time (since no thought would ever be given to switching most real life veteran relievers to a starting role, Braden Looper excepted).

Great post and I hope all of your ideas are considered.

I'll just add that I'd gladly pay for a patch that gave the ability to generate 100% 3+ pitch pitchers. It has a few shortcomings, but I'd rather play with those than the current model. If there were a way to export/import a draft pool, I'd do it myself. Manually editing every draft pool is just too much work.