In the series I've been running on the Plate Approach for batters (and pitchers), one thing that I've created is a run value chart by pitch location. This one is for the 0-0 count, but I have published them for all 12 counts. (click to embiggen)
Focus on the dotted black line. We see that pitches in the Shadow-In region is where the run value is the best for the pitcher (and worst for the batter), at 2 to 3 runs per 100 pitches relative to average. And naturally, any pitch in the Waste region is where the run value is the best for the batter (and worst for the pitcher) at 4+ runs per 100 pitches.
All we have to do is look at each pitch, one at a time, and see where it lands in the strike zone, and add up the corresponding run value of that pitch. A pitch that hits exactly on the border of the strike zone, for the 0-0 count, is worth -1.5 runs (per 100 pitches). A pitch that is right down the middle is about -0.5 runs (per 100 pitches). So, wherever that pitch lands, that's the Location Run Value for that pitch.
We repeat this for every pitch, in each of the 12 plate counts, for every pitcher, every year, since 2018. We add them up and we find our leader. The best pitcher since 2018 in locating his pitches is, very unsurprisingly, but very comforting, Kyle Hendricks, at 43 runs better than average. Here are the leaders:
-43 Hendricks, K
-37 Mikolas, M
-34 Gonzales, M
-33 Eovladi, N
-32 Nola, A
-31 Matz, S
-31 Wheeler
-31 Verlander
-29 Stripling
-29 Scherzer
We can also look at the pitchers with the worst location. Tanner Scott leads the way at 23 runs worse than average. Among the league trailers is Blake Snell at 21 runs worse than average. However, his overall run value is 74 runs better than average. So if we split up his location impact (21 runs worse than average) from the rest (95 runs better than average) we see that Snell is a very impressive pitcher, basically overpowering or deceiving or otherwise beating batters at something other than pinpoint location.
Typically, you will find pitchers with great sliders with "poor" location: this is intended by the pitcher. Patrick Corbin for example provides the ideal example: in his best seasons his location was among the worst. In his worst seasons, his location was among the best. This is because batters would chase his sliders and so, he had no reason to throw it in the strike zone. But in his last couple of seasons, the batters were no longer being fooled. And so Corbin had to throw the pitches more in the strike zone. While his location improved, batters feasted on the low quality pitches in the Heart of the Plate.
As much as we can separate location from the rest of a pitcher's skillset, the location itself is not necessarily a goal unto itself. We'll work on getting this on Baseball Savant in the coming months.
I looked at all batters who faced at least 750 pitches in each of 2021 and 2022. I broke up their performances based on whether the pitch was a fastball (4-seamer or sinker) and non-fastball using Run Values (per 100 pitches). An example of a fastball hitter is Luis Garcia of the Nationals: in 2021, he was +1.5 runs per 100 pitches on fastballs and +2.1 in 2022. On non-fastballs, he was -2.6 in 2021 and -1.7 in 2022. Indeed, the split is even more pronounced if you looked at 4-seamers only, and put sinkers in with "other". I don't know WHY he has those splits. Maybe he's sitting fastball all the way, and can't react to non-fastballs? I really don't know. But his split is very pronounced and is the largest in the league.
The question however if this is a true split, or if it's like a split you might find based on the day of the week: Random Variation. When we look at all batters and compare their 2021 fastballs-only performance to their 2022 fastballs-only performance, we get a correlation of r=0.28.
When we include both their 2021 fastballs-only and their 2021 non-fastballs performance, in predicting their 2022 fastballs-only performance, our correlation goes to... r=0.29. In other words, we get little new information on their non-fastballs performance. The correlation of the non-fastballs is r=0.09. So, the split is real.
How about non-fastballs? Well, here it gets trickier, because non-fastballs means curves and cutters and sliders and changeups. It's a mish-mash of pitches. I can tell you that the correlation is ALSO r=0.29, but in this case, both splits are contributing a good amount to the prediction. So, we really want to split up the non-fastballs into their specific pitch types. The problem: we don't have objective pitch types the way we do with fastballs. One pitcher's slider is another pitcher's curve, or their slider is another pitcher's cutter. Until I actually implement my Pitch Palettes, there's really no point to going too deep here.
That said, given the very very strong split we found with fastballs v fastballs, it certainly looks like some batters are of a certain type. Ideally though, there should not be ANY split. Why? Because if a pitcher knows that Luis Garcia is a big-time 4-seam batter, they will simply stop throwing him 4-seamers. Suddenly, Garcia will stop sitting on 4-seamers (if that's what he is doing). And then he'll be just as prone to missing 4-seamers as he is to missing other pitches. He'll get into a sort of equilibrium point.
It's also possible that a batter is simply is who he is. Take the lovely Jackie Bradley Jr. He destroyed 4-seamers in 2020 and in 2022. But in 2021, he was completely eaten up by 4-seamers. Why? I don't know. But if I were to speculate wildly, maybe he changed his approach. Maybe he used to sit on 4-seamers, but in 2021 he tried to pick out the ball differently. And he guessed wrong alot. Again, just wild speculation here.
Anyway, the splits are real at the league-level, and it'll be interesting to see if batters learn, or they simply am-who-they-am.
I am not asking if he happens to face pitchers who throw a high number of fastballs, nor pitchers who are speed demons.
Instead, I am following this process: For each pitcher, I take all their 4-seam fastballs, and place them into one of three buckets:
+1: That pitcher's 25% hardest fastballs
-1: That pitcher's 25% slowest fastballs
0: That pitcher's remaining 50% of fastballs
In this way, whether you average 98 or 99 like deGrom or you average 88 or 89, it doesn't matter. Each of those pitchers has their own set of +1, 0, -1 fastballs. And therefore, since each pitcher has 25% in their "+1" bucket, that means we'd expect each batter to also face 25% of all fastballs in the "+1" bucket. And naturally, 25% in the "-1" bucket.
I repeated the same process for sinkers.
So, what happened with Aaron Judge? Well, in 2021 he faced 1168 fastballs (4-seamers and sinkers). We'd therefore expect that 25% of those, or 292, would be his opposing pitchers hardest fastballs. And similarly, 292 of those would also be the opposing pitchers slowest fastballs. Instead, he faced 460 "+1" fastballs and 168 "-1" fastballs. That is a lopsided total. In 2022, it was just as lopsided:
+1: 548
0: 581
-1: 201
Our expectation if every pitcher randomly threw their hardest and slowest fastballs to Judge was: 332/665/332. So, pitchers indeed tried to throw their harder fastballs when they were facing Aaron Judge.
***
How about performance against the three groups of fastballs? For that, we turn to run values, per 100 pitches. League-wide, performance against the weakest 25% of fastballs resulted in +0.45 runs per 100 pitches, while against the 25% hardest was -0.20 runs per 100 pitches. Directionally, this is what we expected. The fastest fastballs are harder to hit, and so are an advantage to the pitcher. And vice-versa. The gap is 0.65 runs between the fastest and slowest fastballs. Speed matters.
How about Aaron Judge? In 2021, he destroyed the weak and average fastballs, while he was barely above average against the fastest fastballs:
+1: +0.1 runs per 100 pitches
0: +2.7
-1: +3.8
As you can see, his gap in performance was much much wider than the league-wide gap. His gap was 3.7 runs per 100 pitches of performance, compared to the almost 0.7 that the league sees. He simply destroyed the pitcher's weakest (and average) fastballs.
How about 2022?
+1: +3.3 runs per 100 pitches
0: +3.0
-1: +2.6
Well, he demolished each pitcher's fastest fastballs. In 2022, that ranked him 8th league-wide. The top two batters in demolish the league's fastest fastballs, even more than Judge: Bryce Harper and at #1 is Joc Pederson.
One thing that interested me is if older players start to lose it faster against the fastest fastballs. In 2021, Miguel Cabrera was indeed below average against the fastest fastballs. But also terrible against the slowest fastballs. In 2022, he looked like this:
+1: -1.5 runs per 100 pitches
0: -1.6
-1: +0.7
That is more what I'd expect from an aging hitter. Is this actually true, or it's just n=1? Here's Pujols in 2022:
+1: +0.2 runs per 100 pitches
0: +3.5
-1: +0.8
Not really what I was expecting. So, this question really deserves more analysis. And of course, this is only for fastballs. We can look for and ask the same question for curves and sliders and changeups and so on.
And then other questions is to ask how reliable it is to look at the data in this manner. Do we learn anything about batters, or is this like splitting up the data by day of the week: you end up with Random Variation carrying the day, and so there's no true split. I don't know, that needs to be looked at too.
Once we flag the data in this manner, all this research becomes possible. Stay tuned.
(Editor's Note: original article had data only for 4-seam fastballs.)
Since 1999, there have been 23 batters(*) with at least 3000 times batting in an 0-1 count. By the time their plate appearance ended, their wOBA was 33 to 59 points below their career wOBA. In other words, none of these players ended up performed better when passing through an 0-1 count than 0-0.
(*) For purposes of this study, I broke up the players based on their bat-side and the opposing pitch-hand.
The league average drop is 47 points. What should we have expected in terms of range, based purely on Random Variation given 3000+ plate appearances? One standard deviation in wOBA is about 0.5/sqrt(PA), meaning 9 points. Two standard deviations is therefore 18 points. And so, we'd expect to find 95% of batters to have a difference of 47+/-18 or 29 to 65 points in wOBA difference. We ended up with 33 to 59 points of drop. In other words, that Nick Markakis had a wOBA drop of only 33 points in an 0-1 count and that Alex Rodriguez had a drop of 59 points doesn't actually mean anything about them specifically: we expected to see those kinds of drops, just by Random Variation.
Ok, how about 2000+ plate appearances? Chris Davis has a 93 point drop (or 46 more than the average) and Marco Scutaro has only a 9 point drop (or 38 less than the average). All 147 batters still show a drop overall, as we'd expect. But what about the range? Well, with 2000 PA, one standard deviation is 11 points. Here we see changes of 3 to 4 standard deviations. Does that mean that what Davis and Scutaro did is real? No! What it DOES mean is that we can't ascribe the ENTIRE difference to Random Variation. We still have a very healthy amount of that deviation that is nothing but Random Variation. But not all of it.
How about 1000+ plate appearances? Now we have Pedro Alvarez (against RHP anyway) who drops 100 points (or 53 more than average) on an 0-1 count, while Khalil Greene (against RHP) who drops only 2 points (or 45 less than average). One standard deviation for 1000 PA is 16 points. We are again around 3 standard deviations. With 589 batters, we start to expect to see things happening beyond 3 SD just by Random Variation to ~two batters.
What we can do is calculate the z-scores for each batter. A z-score is simply the number of standard deviations from the mean that the player's performance is. These extreme batters that we've noted have a z-score around 3. What matters is the distribution of all these z-scores. If the standard deviation of these z-scores is exactly 1, that's proof that what we observe is purely Random Variation. If there is an "0-1 skill", then we'd expect a spread wider than normal, and so, the standard deviation of the z-scores would be more than 1. And the more the skill, the much higher the standard deviation of z-scores.
And what do we see? The standard deviation of the z-scores is... 1.0. In other words, no 0-1 skill.
I'll go through the other 10 counts to see what kind of results we get. My expectation is that maybe at 3-2 we will see something, because that is such an extreme count. Of course, even if there is a 3-2 skill, no batter will intentionally go from a 3-1 to a 3-2 count because he can "handle" a 3-2 count better than the typical batter. Whatever skill he may have at 3-2 can't possibly overcome the inherent advantage he'd have just by being in a 3-1 count.
You can read The Book, as we spend alot of time on that topic. But, for those who don't want to do that, I repeated and expanded that study. I did this because they were talking about it on the Statcast Broadcast on the MLB Network. This is what I did:
I looked at all batter-pitcher matchups from 1982-2019, split based on whether they had or didn't have the platoon advantage. I limited it to those where they faced each other at least 15 times as part of the "to-date" and at least once as part of "rest of career", and a pitcher had to have at least 20 such batters. For each pitcher, I ranked the batters-to-date based on their wOBA matchup. The 25% highest wOBA matchups goes into the "hot for batter" and the 25% lowest wOBA matchups goes into the "cold for batter".
There were 634 pitchers in the platoon group. The hot batters were .513 wOBA (aka better than Trout) and the cold batters were .171 wOBA (aka about as good as a pitcher batting). How did they do the rest of their career? The hot batters were .351 while the cold batters were .330. That's a 22 (rounding error) advantage for the hot batter. It's not the 342 points advantage we observed, but 22 is not 0. Or... is it?
The hot batters are, in their careers slightly better than the cold batters. Their career average is .354. In other words, we observed .354 hitters who were hot at .513, and when we looked at it AFTER the observation period ended up at .351. Similarly, the cold batters had a career average of .339, and we observed them when cold at .171. AFTER the observation period, they hit .330. In other words, the 22 point advantage that the hot batters had over the cold batters can be explained almost totally by the 15 point advantage that the biased sample would suggest.
The 342 point advantage we observed shrunk all the way down to a 7 point advantage in reality.
As for the matchups when they were same-handed, it's more of the same:
Hot batters, who in their career are .338, were observed at .502 during their hot period, and .339 after the hot period
Cold batters, who in their career are .327, were observed at .165 during their cold period, and .324 after the cold period
The 337 point advantage we observed shrunk all the way down to a 4 point advantage in reality
In other words, about 98-99% of what you observe as being hot or cold is Random Variation. The other 1-2% is real. At least as long as you use outcomes, like walks, hits, and homeruns.
?In a brilliantly simple method, Hareeb compares batter performance to whether the prior pitcher was the same handedness of the current pitcher or not.
The theory is that you can mess up a batter's approach if you can throw a fast reliever after a slow starter, or a reliever with a very different arm slot than a starters, etc. Basically, the more the reliever is different from the starter (or the previous pitcher), the easier it is to beat the batter.
Hareeb took the first simple approach of simply taking the obvious: pitcher handedness requires a very different release point and incoming trajectory. That, more than anything, is going to upset a batter's approach. One would think anyway. And Hareeb found nothing at all.
It will be interesting to see if something more refined will lead to different results, but given the above, I'd expect: nope.
?Good stuff from Craig, showing how balanced Strasburg is with the pitches noted (4 seamer and curve). One of the interesting things is that we'd expect (per pitch) that the run values to be identical. Or, if it's not identical, it would be somewhat close. If it's not, then the batters have simply decided to sit on one pitch more than the other, which should make them a bit exploitable. Kluber for example has an impressive breaking pitch (whatever you want to call it), and so, batters are getting nothing out of it, while they are having reasonable success on his fastball. They've likely decided to give him the breaking pitch and sit on the fastball. Which is why he throws his breaking pitch so much in comparison. (Though really we'd want to look at 4 seamer and sinker.) Anyway, fascinating topic as the batter and pitcher are dancing around the strike zone.
?We were just discussing this at the office, and if you want to know what we were saying, well, Ben Clemens essentially repeated word for word what we said, without him actually being here! Terrific stuff from Ben.
The main cause, I was suggesting, was the dilution of talent. I'm a little concerned with how Ben approached it, since that will cause a selection bias. In order to get around that bias, Ben can look at first-half performance, establish the identity of pitchers at that point, then look at second-half performance.
The other cause, as Mike and Matt in the office were suggesting, is that if you know you will pitch less, you will then pitch more like a reliever. Ben tackles this as well.
And of course, we always have the "Third Times The Harm" effect for starting pitchers, which Ben also tackles.
Note that looking at the first 7 batters in the lineup is a clever way to get around pitchers-as-batters. Note that you'd want to split between home/away, for reasons we've talked about many times in the past: the visiting pitcher is "cold" against the home batters in the 1st inning.
A home run of an article for an Opener on the subject.
?Data from 2018 season, includes first 8 innings, of games that lasted at least 8 innings. Excludes bunts.
"P100" is "per 100 pitches". Red is more runs. Blue is less runs. Hopefully everything in there is decipherable.
Sliders continues to be the revelation as we've learned recently. Batters not swinging at first pitches when they are down the middle has got to be a huge inefficiency in batter approach (you would think).
?This is something that might appeal to a small audience out there. MGL for one. It SHOULD appeal to a very wide audience. But it's a hard topic to blog about.
The code numbers in the column corresponds to the strike zone map you see on the right. 1-9 is the heart of the strike zone, 11-19 is the "shadow" zone, roughly one ball width on either side of the official strike zone, and 21-29 are outside the strike zone ("ozone").
The plate counts at the top are ordered from batter to pitcher counts, from 3-0 to 0-2.
The two sections show whether the batter swings at the pitch, or takes.
The values you see are run values.
Virtually all the time: you want to swing at any pitch in the heart of the strike zone,regardless of count, and take any pitch in the ozone. For the Shadow Zone, even though it comprises partly of the strike zone, you USUALLY want to take... except in a few spots when you should swing.
I marked in purple one such occasion (but there are a few others). So, at 2-2, if the ball lands at the 14 or 16 zone, swinging will cost you about .01 to .03 runs. But TAKING the pitch will cost you .10 runs. That's because at 2-2, the cost of taking is very high, as a pitch near the edge, called a strike is a strikeout.
Don't pay too much attention at the 3-0 counts, since they are VERY low quantity.
***
Anyway, hope you like it. And I have an idea for something better. I kinda hinted at it above. If you see it, great. If not, then wait until the offseason.
These two choices were actually exactly even, as you can tell here, by focusing on the 0-1 and 2-2 lines, helpfully right next to each other:
with the wOBA from that point of the count to the end of the plate appearance virtually identical. But HOW the plate appearance ends, and so the path to the end, is different.
This is like on a game show, where you are offered 10,000$ or a 50/50 shot at 20,000$, you "should" be indifferent, since the expected values are the same. But if you have no money, then you would likely prefer the 10,000$. But there's always a break-even point. You might for example risk the 10K for a 50/50 shot at 25K. It all depends on your personal risk aversity.
So, look at the chart and you see that by being at 2-2, you will get 0.07 more K/PA... and 0.07 more BB/PA. In other words, preferring being in the 0-1 count as a batter, which the voters overwhelmingly favored, means that you rely on batted balls being in your future. Basically, the as-batters wanted the control.
The "as pitcher" voters were indifferent to the two, as they should have been. Presumably, they are just as happy to smell a 2-strike count, as being early/ahead in the count.
?This was a topic that was front-and-center in The Book. And the conclusion, then, was that it was just another PA to add to the PA pile. Since then however, I've done research that suggests something a bit more. I'm not ready to talk about it, but until then, let's talk about Pizza's interesting piece here.
What he did was try to forecast a future matchup, using: (a) their actual past matchup and (b) their season-to-date totals. And he found that BOTH added value. To which I say: well, obviously, both HAVE to add value. I'll explain in a sec.
He also said that the season to date gets weighted at twice their past matchup. To which I say: well, I don't know what that means unless I know the average number of PA in each group. I'll explain in two secs.
First the first one: adding data is always good. So, no one is saying that matchup data should get tossed. It just needs to get added to the pile, basically, unweighted.
The second one: When you use "season to date", you are using 0 to 600 PA for each hitter and 0 to 900 PA for each pitcher, more or less. The average for a starting hitter would mean about 300 PA, but since he includes ALL hitters, the average is probably... I dunno... 150 PA? Similarly, for SP, the average is probably 400 and for relievers it's 50 or something, so maybe it's 200 PA as the average? I dunno. Let's say the harmonic mean of the average number of PA for each pairing being forecasted is around 175 PA.
In addition to that, we have something that has 50 PA. From my standpoint therefore, using these pure-guesses, we'd expect season-to-date to be weighted at 3:1 or 4:1, simply because one sample has 3x to 4x more than the other sample. Instead, we find it's only 2:1. That is, matchup data gets "overweighted" by a factor of 2 relative to non-matchup data. That's a significant finding.(*) But, I'd like to see Pizza fill in all the numbers I'm guessing.
(*) Note that in my unpublished research, with a methodology that is different from Pizza, I can confirm that the weighting is definitely higher than 1x. So, if anyone wants to keep researching this, you will find something.
Two points: as MGL brought up in the comments, you have to account for the handedness issue (batter and pitcher). That's an easy variable to include. The other is that Pizza uses "season to date", which means he starts at game 1. Which means it's kind of a mess in terms of some samples having only 1 PA in season to date (actually, it starts at 0) and others at 600 PA.
Really, in order to do an apples-to-apples comparison, he should simply limit it to match each player's actual number of trials. If let's say Rice/Guidry faced each other 53 times, then I'd look at Rice's first 53 PA in the season against LHP (which by the way, MUST exclude Guidry), and Guidry's first 53 PA in the season against RHH. What this does is now make clear we have two equal sized samples, and we've neatly handled the handedness issue. Then we just figure out the weights.
And if Pizza does this, we may find we need to weight matchup data at 2x to non-matchup recent-data.
Indeed, way back when I introduced the Pitch Count Estimator, I used a Markov Plate Count to develop my model. I then used that concept to came up with a quick model as noted in the above link. So, I'm glad that Jim is able to shine his light on this very underdiscussed concept.
?This was one of the key findings in The Book, that the more times you go through the order, the higher the offensive output. It's gotten to the point that this has become accepted conventional wisdom.
The question is whether it was due to pitch fatigue or the number of times the batter comes to bat (sees pitcher's arsenal). MGL took my general idea from The Book and turned it into a great research piece a few years ago, enough that he is now the leading voice on understanding the TTOP. As Bill James once noted, when we publish our work, that work is now an orphan. We can't really control or contain it, and someone will get inspired enough by it that they'll eventually take it to places you didn't. This is what happened here.
MGL responded to a claim from Pizza Cutter with additional research and theory. But what I liked beyond everything MGL did was the little nugget about pinch hitters. This would be a terrific offshoot project for an aspiring saberist. We always look for "controlled" studies in our line of work, and a pinch hitter offers that. What a great idea to look at whether pitch counts impacts performance than to look at how pinch hitters respond to pitchers who have, say 60 or fewer pitches and 90 or more pitches (or whatever might be a good demarcation point).
First and third with Trout, two outs bottom of the 9th. A hit wins it, a walk gives the next batter a chance. So, one of the very few times we care about batting average. Trout will get a hit 30% of the time there's no walk, and he'll get a walk 12% of the time. If Trout walks, then it's on Pujols who will get a hit or walk 32% of the time.
But, this is against Kimbrel. So, maybe Trout gets a hit 25% of the time when there's no walk, and he'll still walk 12%. And Pujols will get on base 25% of the time against Kimbrel.
So, Trout hitting means: (1-.12) x .25 + .12 x .25 = 25% of the time Angels win, 75% goes to extra innings.
If Pujols hits: 25% they win, 75% goes to Extra innings.
So, same thing.
***
Interestingly, and historically, 1st and 3rd 2 outs has been a 69% chance of winning, compared to bases loaded of 68%. That's in the 9th inning.
But in extra innings, it's 66% with 1st and 3rd, and 68% with bases loaded.
***
All to say, it's pretty much a pick-your-poison kind of situation, and you can look at Trout and Pujols specifically against Kimbrel, a very extreme kind of pitcher. It's possible that one guy matches up very differently against Kimbrel than the other.
***
Anyway, that's enough to get some aspiring saberist to take this to the next level. Either that, or we get MGL to run his simulator against these very extreme matchup types.
In The Book, Andy wrote about protection, and as luck would have it, it was excerpted at The Hardball Times back in 2006. It exists in terms of changing the approach of the pitcher, but the question is what is the net impact.
?I did an informal peer-review of this paper. The author was a charter student of mine. I told him it was terrific, he gave me an idea I never had earlier considered, and I made some minor point, that I don't really recall. Anyway, I don't know how long it will be up, so, grab it while you can.
Some day, I'll really quit my day job and start back that online school. I'm not even sure why I haven't done that yet.
I saw this a few months ago, and I had comments then. So as to not influence anyone, I'll see if someone can see the same issue I saw. So, read it, and then tell me why Pedroia's league-best of more balls than strikes in this metric could just as well be a league-worst more strikes than balls.
As a quick experiment, I compared their split performances. Overall, in the league, hitters perform a few points of wOBA better with runners on base. FIP goes up .13 points, and xFIP goes up .25 points. With runners on, the decreased-fastball group averaged a wOBA increase of seven points, with an FIP increase of .20 points and an xFIP increase of .22 points. With runners on, the increased-fastball group averaged a wOBA increase of 12 points, with an FIP increase of .15 points and an xFIP increase of .33 points.
So, the two extreme groups, those that decided to throw alot fewer fastballs and those that decided to throw alot more fastballs, did worse than the overall league average. This would be consistent with the league being aware that they are changing their otherwise optimal approach, and the league responding to that change. There were a couple of big names in there (Doc, J Fernandez, Price, Cueto, Kimbrel, one or two others out of 48), otherwise not a collection of pitchers that makes you think that the approach is something that will pay off.?
Recent comments
Older comments
Page 1 of 150 pages 1 2 3 > Last ›Complete Archive – By Category
Complete Archive – By Date
FORUM TOPICS
Jul 12 15:22 MarcelsApr 16 14:31 Pitch Count Estimators
Mar 12 16:30 Appendix to THE BOOK - THE GORY DETAILS
Jan 29 09:41 NFL Overtime Idea
Jan 22 14:48 Weighting Years for NFL Player Projections
Jan 21 09:18 positional runs in pythagenpat
Oct 20 15:57 DRS: FG vs. BB-Ref
Apr 12 09:43 What if baseball was like survivor? You are eliminated ...
Nov 24 09:57 Win Attribution to offense, pitching, and fielding at the game level (prototype method)
Jul 13 10:20 How to watch great past games without spoilers