Kim

I want to start by pointing out that I love every single one of our commenters here. Every one. No exceptions. That said, I do a lot of work around here, and that gives me the privilege (I’ve decided) to occasionally make a riposte here rather than in the comments.

So, Roger, how do you propose to get Kim the time he needs to take on MLB pitching? What’s the point of a cushion (and a great team) if you don’t have a little leeway to make your team a little worse today for a sufficiently large return later in the season? I get the sense you’d prefer to cut him loose. I think that’s premature. Way premature. I certainly agree with you that some combination of Mateo and Dubón is clearly superior of producing today what Kim is capable of producing. So playing Kim will lower your chances of winning each game by, let’s say, 5%. (And that’s a lot.) A productive Kim raises your chance of winning a game by, say 2%. But 2% x 80 games is a lot better than -5% x 20 games…. about half a game better Obviously that assumes that playing Kim for three weeks, even if he sucks, will pay off. But if you don’t play him, you’ll never know. And if he still sucks after that, well, we’ll have sacrificed a couple of games off our lead. I can live with that.

AJ, CJ

I joked last year that April 28th is two-initialed Polish ex-Braves day, for that was the day that CJ Nitkowski pitched to AJ Pierzynski, striking him out in their only head-to-head confrontation. I missed it this year, but watching Gaudin work with Pierzynski tonight on Fox reminds me that these two have been warily circling for years. Watch your back, CJ.

As color commentators, though, I really like both of them. One brings a veteran pitcher’s perspective and one a veteran catcher’s. Between them, I learn a lot.

What Did You Do Today, Jonathan? Nerd Alert!

I got to thinking about my analysis of pulling pitchers yesterday, and I decided to do a better job. First, I got every situation in which a starting pitcher is tied or in the lead with two outs in the fifth inning, and I looked at whether the manager pulled the pitcher or not. That’s 284,699 situations. In those 284,699 situations, the starting pitcher was pulled 2296 times, a minuscule 0.8% rate.

But the base rate is misleading. Many times a manager has more than one of these in a game. Last night, Holmes had two outs leading by two with a man on first and Weiss gave him one chance. When JJ Bleday singled he pulled him. But of course the situation had changed: it was now first and third with tying run on base and a pitcher who had just given up a single. As we’ll see, that stuff matters a lot.

What if I subdivide the data by the size of the lead? That turns out to matter a lot:

Stay In Hook % Hook
lead
0 82555 901 1.1
1 65477 593 0.9
2 47791 393 0.8
3 33018 211 0.6
4 21661 114 0.5
5 13792 51 0.4
6 8577 13 0.2
7 5126 10 0.2
8 2955 5 0.2
9+ 3747 5 0.1

And what about the situation on the bases? Guys in scoring position matter a lot:

Stay In Hook % Hook
basestate
123 6661 372 5.3
13 10244 257 2.4
23 7213 167 2.3
12 24562 558 2.2
2 32042 296 0.9
3 12388 105 0.8
1 59290 370 0.6
Empty 132299 171 0.1

It clearly matters what the pitcher did on the previous batter as well:

Stay In Hook % Hook
Previous Hitter
hr 3055 108 3.4
triple 1402 47 3.2
hbp 1654 52 3.0
double 9234 273 2.9
walk 23321 681 2.8
single 40011 815 2.0
Other 162327 246 0.2
k 43695 74 0.2

Finally, we know that the bottom of the fifth meant something very different in 1926 than it means in 2022. In the first case the pitcher is getting his second wind, and in the second he is sucking wind. So let’s look at the hook rate by a division into eras:

Stay In Hook % Hook
Era
2013 and later 39610 765 1.9
1960-2012 153679 1060 0.7
Pre-60s 91410 471 0.5

We can look at combinations of all these, but the combinations quickly get really hard to understand. That’s when you reach for a trusty statistical model, the logit, to give you a concise view. I’m not going to reproduce the logit coefficients here, because I don’t want to give statistics lessons about the logarithm of the odds-ratio (to be honest, I actually do, but this probably isn’t the right time for it) but it produces a prediction of the probability of a hook as a function of lead,base state, previous batter result, and era for every one of the 284,699 situations. Since I’m not giving you the coefficients, you’re going to have to trust me, but everything works as you’d expect.

OK… so now what can I do with these 284,699 numbers? I can look at managers and see how their predicted hook rate lines up with their actual hook rate. The ones with a high ratio of actual to predicted are Captain Hooks, while the low ratios are the “Give the guy a chance to earn a win” managers.

Now to get stability of the Actual number, you want to face this situation a lot. You don’t want to conclude that someone has a quick hook after 50 situations because he pulled one guy. This is just the small sample problem we deal with all the time, except where the expected number is really small (.010 here instead of say 0.250 for batting average) you need much larger samples. Conveniently, there are exactly 100 managers who have faced 913 such situations or more. These are essentially the top 100 managers by games managed, though I didn’t check at the bottom margin whether that was true. The exception would be managers who managed a disproportionately small number of leads in the middle innings despite careers long enough to hang around. Even 1000 opportunities is a pretty bare minimum for judging decisions this rare. An underlying 1 percent rate is only 10 hooks. 12 would be twenty percent more likely than average, but could easily just be noise.

Here then, ranked, are the hook ratios adjusted for situation:

Cases Predicted Actual Ratio
Manager
Craig Counsell 1112 0.016890 0.030576 1.810277
Johnny Oates 1167 0.007203 0.011997 1.665405
Steve O’Neill 1417 0.005552 0.009174 1.652517
Casey Stengel 2724 0.006185 0.009912 1.602570
Gil Hodges 947 0.006686 0.010560 1.579298
Billy Southworth 1384 0.004586 0.007225 1.575479
Phil Garner 1426 0.006687 0.010519 1.573032
Felipe Alou 1465 0.007833 0.012287 1.568521
Lou Boudreau 1700 0.005316 0.008235 1.549102
Don Zimmer 1277 0.006094 0.009397 1.542028
Bill Terry 1086 0.004822 0.007366 1.527718
Earl Weaver 1916 0.006376 0.009395 1.473375
Whitey Herzog 1736 0.005874 0.008641 1.471064
Al Lopez 1799 0.004964 0.007226 1.455594
Jimy Williams 1246 0.008840 0.012841 1.452676
Jerry Manuel 994 0.006926 0.010060 1.452602
Burt Shotton 1061 0.006539 0.009425 1.441367
Tommy Lasorda 2226 0.005939 0.008535 1.437090
Frank Robinson 1539 0.007333 0.010396 1.417811
Roger Craig 1012 0.006300 0.008893 1.411621
Bill McKechnie 2639 0.005127 0.007200 1.404343
Kevin Cash 1011 0.018005 0.024728 1.373362
Charlie Grimm 1773 0.005422 0.007332 1.352401
Joe Cronin 1600 0.004663 0.006250 1.340197
A.J. Hinch 1256 0.017055 0.022293 1.307089
Gene Mauch 2658 0.006632 0.008653 1.304835
Joe Girardi 1496 0.013344 0.017380 1.302397
Bill Rigney 1710 0.005980 0.007602 1.271258
Lou Piniella 2710 0.007718 0.009594 1.243011
Dave Roberts 1137 0.016370 0.020229 1.235739
Joe Maddon 1878 0.012964 0.015974 1.232243
Ralph Houk 2097 0.005937 0.007153 1.204775
Bill Virdon 1488 0.006401 0.007392 1.154835
Jim Tracy 1176 0.006002 0.006803 1.133452
Chuck Tanner 1859 0.005759 0.006455 1.120811
Billy Martin 1752 0.007369 0.007991 1.084391
Al Dark 1398 0.006033 0.006438 1.067029
Bucky Harris 3113 0.005138 0.005461 1.062781
Leo Durocher 2712 0.005568 0.005900 1.059567
Mike Matheny 1077 0.018440 0.019499 1.057408
Bud Black 1714 0.015509 0.016336 1.053323
Jim Riggleman 1070 0.007148 0.007477 1.046010
Rogers Hornsby 1029 0.006560 0.006803 1.037001
Fred Hutchinson 1206 0.005614 0.005804 1.033954
Danny Murtaugh 1510 0.005893 0.005960 1.011376
Dusty Baker 3080 0.009447 0.009416 0.996639
John McGraw 2395 0.004649 0.004593 0.987858
Birdie Tebbetts 983 0.005279 0.005086 0.963492
Jim Leyland 2579 0.008065 0.007755 0.961540
Torey Lovullo 961 0.017391 0.016649 0.957380
Ron Washington 1089 0.010613 0.010101 0.951720
Walter Alston 2791 0.005682 0.005374 0.945874
Don Mattingly 1246 0.015465 0.014446 0.934144
Joe Torre 3235 0.006622 0.006182 0.933667
Paul Richards 1275 0.005044 0.004706 0.933025
Terry Francona 2675 0.011305 0.010467 0.925915
Clark Griffith 1118 0.003875 0.003578 0.923280
Dick Williams 2223 0.006823 0.006298 0.922992
Jim Fregosi 1512 0.005747 0.005291 0.920703
Brian Snitker 1036 0.018993 0.017375 0.914792
Charlie Manuel 1380 0.006434 0.005797 0.901023
Mike Hargrove 1751 0.007728 0.006853 0.886771
Joe McCarthy 2891 0.005148 0.004497 0.873523
Connie Mack 4056 0.005466 0.004684 0.856980
Bobby Cox 3596 0.006847 0.005840 0.852897
Wilbert Robinson 1909 0.004924 0.004191 0.851082
Lee Fohl 996 0.004786 0.004016 0.839192
Buck Showalter 2434 0.012402 0.010271 0.828162
Clint Hurdle 1802 0.011576 0.009434 0.814962
Terry Collins 1476 0.012539 0.010163 0.810466
Chuck Dressen 1369 0.005540 0.004383 0.791101
Frankie Frisch 1537 0.004935 0.003904 0.791087
Red Schoendienst 1476 0.006013 0.004743 0.788754
Dave Bristol 983 0.007789 0.006104 0.783642
Jeff Torborg 939 0.006808 0.005325 0.782155
Bob Melvin 2301 0.014109 0.010865 0.770053
John McNamara 1624 0.006518 0.004926 0.755717
Fredi Gonzalez 1014 0.012004 0.008876 0.739401
Tony La Russa 4034 0.007210 0.005206 0.722003
Davey Johnson 1864 0.007608 0.005365 0.705122
Tom Kelly 1599 0.007124 0.005003 0.702249
Ozzie Guillen 1018 0.005659 0.003929 0.694336
Bruce Bochy 3199 0.011319 0.007815 0.690428
Sparky Anderson 2966 0.006443 0.004383 0.680293
Mayo Smith 922 0.005039 0.003254 0.645730
Cito Gaston 1279 0.006336 0.003909 0.616976
Mike Scioscia 2237 0.011668 0.007152 0.613009
Jimmy Dykes 2044 0.004881 0.002935 0.601384
Miller Huggins 1922 0.005274 0.003122 0.591925
Eric Wedge 1176 0.008049 0.004252 0.528201
Jack McKeon 1451 0.006588 0.003446 0.523082
Buck Rodgers 1126 0.006794 0.003552 0.522906
Ned Yost 1779 0.013108 0.006745 0.514601
Pat Moran 989 0.004405 0.002022 0.459119
Ron Gardenhire 1601 0.010689 0.004372 0.409054
Bobby Valentine 1747 0.006260 0.002290 0.365784
John Gibbons 1129 0.014564 0.005314 0.364898
Hughie Jennings 1298 0.004913 0.001541 0.313615
Art Howe 1601 0.007296 0.001874 0.256828

[Note: I improved the model dramatically since I wrote this yesterday, by including the number of chances the manager gave the pitcher to get out of trouble as a variable. I changed the table to reflect that and change some of the numbers below to reflect the changes. As far as the Braves went, Torre became a little less conservative, Snitker was roughly unchanged, while Cox and Gonzalez became slightly more conservative, which also fits my thesis better.]

Even though these data have been adjusted for pitching era, Craig Counsell is an outlier by a huge margin. He is almost twice as likely to remove a starting pitcher tied or with the lead on the cusp of the possibility of earning a “Win.” If we don’t adjust for era, he is 4.4 times as likely to pull a pitcher as the historic average. (Failure to adjust for era makes modern managers soar to the top of this list: the next four are Kevin Cash (3.6), Dave Roberts (3.0), A.J. Hinch (2.8) and Mike Matheny (2.5).)

On the other end, Art Howe stuck with his guy. He was almost 4 times more likely to let a guy pitch out of a situation, era-adjusted. (Hughie Jennings edges him out when we don’t adjust for era.)

So what about the Braves managers? Note that a ratio of 1.00 means you’re exactly average in your predisposition to pull pitchers. Descending the list, Snitker is 0.90 0.91, as is behind Joe Torre at 0.93 (this also includes the much longer time he spent managing the Mets, Yankees, Dodgers and Cardinals). Bobby Cox (including his time in Toronto) is just below that at 0.89 0.85. Fredi Gonzalez (including his time with the Marlins) is the most hesitant to pull starters at 0.80 0.74, about 20% 25% less likely to yank his starter. These are fairly modest proclivities (compare with similar guys on the other side of 1.00) but they all point the same direction.

Ignoring Weiss for the moment, the fact that Braves managers are all somewhat slow to remove guys who have a chance for the win is pretty strong evidence, especially, since all of these people have strong organizational ties. Everyone worshipped Cox and what he did, and Snit and Fredi sat at his side watching him manage for years. Torre played for the Braves at a somewhat earlier time, but his Braves managerial time was pretty minimal, so throwing him in as another example is a little strained, but it’s my essay, so I will. Note also that former Braves GM Paul Richards is also a 0.9 guy, and Cox acolyte Ned Yost is at 0.52 0.51, often regarded as the most likely successor to Bobby before Fredi arrived and he jumped ship to Milwaukee. Yost is half as likely to pull a starter.

What about Weiss? I haven’t put this year’s games in these samples, but we can look at Walt Weiss‘s managerial stint in Colorado. He had 449 opportunities, and he was predicted to pull 2.3% of pitchers, but he actually pulled only 1.3%, so at least in Colorado he was quite unlikely to do what he did last night. (Note the small sample problem… that’s only 5 hooks less than expected.) Of course, Coors Field being what it is, sticking with pitchers is probably a very different decision. With this data I could add Colorado home games to the model, but I didn’t. In any case, Weiss got fired from that job, and it is quite likely that he has learned something playing for Bobby and sitting beside Snit. What’s interesting is that the lesson he seems to have learned is to be more aggressive when pulling pitchers, when his influences were clearly less aggressive. In any case, nothing in the statistics can tell you anything about Weiss’s tendencies now — there just aren’t enough decisions to draw any conclusions. But he sure seems to be cut from a different mold in other decisions, so it isn’t a stretch to think that last night’s decision broke free of the Braves Way mold.

In principle, there are a ton of things one could do with this. First, one could add more variables to this model, like experience of the starting pitcher. Smoltz and Glavine often speak of this particular decision as an important waystation to pitcher development, implying that a quick hook hurts a young pitcher’s development. Second you could try interacting variables. One obvious example: pitchers are very likely to be replaced when they just gave up a home run, but unlikely to be replaced when the bases are empty, but you might want to subdivide empty bases by subtracting out the times the previous hit was a homer. If I were writing a PhD thesis, that’s the sort of stuff I’d have to do. More broadly, one could make a model that didn’t just evaluate this particular decision, but evaluated all situations. The nice thing about this situation, though, is that it is really clean. A model of every factor that would go into the decision to remove a starting pitcher in every situation would be hella complicated. (That’s a technical statistics term.)

Another thing you could do is use this data to evaluate strategy. Do hookers do better than leavers? Which group gives up more runs in the fifth? Which group wins more games?

That’s what I did between 8:15 am and 4:15 pm today. One day, I’ll get a life.

The Game

Martin Perez against Brady Singer, pictured above in what is probably the worst visual joke I’ve ever made. The Braves started the scoring in the second when Mike Yastrzemski knocked Ozzie Albies in. JJ Bleday hit his second home run in two days in the bottom of the 2nd to make it 2-1. Ronald Acuña Jr. then hit his third homer in three days to tie the game in the 3rd. (Statcast said it would have been out of 21/30 parks. I’m not buying it.) In the bottom of the third, RAJ misplayed an Elly De La Cruz double into a triple which could possibly have been an inside-the-park homer if De La Cruz had picked up the mishap. Perez got the next two batters, though, so the failure to risk it may have cost the Reds a run.

Jorge Mateo led off the fourth with the third homer of the game to give the Braves the lead, and bringing joy to Roger’s heart. Subsequently, Acuña walked and stole second and third, and Harris and Olson walked to bring Ozzie up with the bases loaded. He might have broken the game open if he hadn’t fouled a ball off his testicles first.

Mateo made an excellent play on De La Cruz to end the 5th. Roger is really making me look stupid now. Tejay Antone, appropriately named since he has had 3 Tommy John surgeries, entered in the 6th and wriggled out of some trouble. Tyler Kinley worked a nine-pitch inning in the bottom of the inning. (That reminds me that Pierce Johnson, who I always liked, went to Cincinnati and was having a good season before he just came up with the dreaded elbow inflammation. I wish him well.)

The fourth homer of the night went to left center from Matt Olson. 4-2. Robert Suarez outdid Kinley with an eight-pitch eighth. The fifth homer, RAJ’s second of the night and fourth in three days, made it 5-2 for Iggy. Could he beat Suarez’s eight pitches? He could not. 15 pitches. Ball game.

Multimorphism

I took this picture of my screen in the second inning. Are Ozzie and De La Cruz even of the same species? Ozzie is closer to the camera, so that makes him look larger. Ranging between Freddy Patek and Randy Johnson (and I’m not even counting Eddie Gaedel) the fact that baseball has a wider range of body types than any other sport (maybe golf beats it) is one of the great things about it.

40 Wins

This is the earliest Atlanta has ever gotten to 40 wins. The previous record was June 1, in 1998. The latest in a full season was August 12th in 1988. That team won 54 games. We’re going to beat that easily. This is better than my 162-0 prediction.

Chance for a sweep tomorrow. Strider against Lodolo at 1:40.