Skip to content

SABR42 Day Three

The last three research presentation slots were on Saturday, along with the player panel, the reprise of the Case Competition winners from the Analytics Conference, and a bunch of committee meetings, as well as the Trivia Contest finals. (By all accounts the Trivia Finals were a blast–I followed them on Twitter from my room.) Between trying to sleep off my cold and publications-related meetings, I managed to miss just about everything Saturday except for the research presentations themselves:

Andy Andres: The Effect of Temperature and Humidity on Pitching
Heroes at the Mike: Baseball’s Longest Serving Broadcasters
Michael Humphreys: We Have Underestimated Fielding Value. A Lot.

Andy Andres: The Effect of Temperature and Humidity on Pitching

Andy teaches a course at Tufts University on sabermetrics, and this research came out of a study begun by some of his students. They came up with this question: what is the effect of the weather on pitching? Is there a relationship?

The conventional wisdom is that pitchers have the advantage in cold weather. April is usually lowest scoring month and batters have more difficulty. Whereas warm and humid air is less dense, which is to the pitchers detriment. In hot, humid air fastballs are faster but other pitches have less drag and less movement. Humidity could cause the ball to expand and worsen the pitcher’s grip.

Past Research:
-fastball velocity studied previously: good correlation with high velocity and high temperature
-Hardball Times researched outcomes, showed more strikeouts in hotter weather, etc.

The velocity, movements, and effectiveness of pitches will be affected by weather.
The conventional wisdom will hold.

First Study: Spring 2011
The students had only one semester to do the project. They had to 1) select the pitchers, 2) examine the game logs, 3) gather PitchF/x data, 4) gather weather data, and 5) correlate it all.

To select the pitchers for the data set, they chose only pitchers who made 25+ starts for the same team every season 2008-2010. They had to have the following minimum percentages of the type of pitch they threw: 40% four-seam fastball, 8% curve, 8% changeup.

The result was 15 pitchers, including the likes of Paul Maholm and Wandy Rodgriguez. They used the BrooksBaseball PitchF/x tool, for the pitch type and break data, and Weather Underground for the weather data on temperature and humidity.

Andy then showed all the data plotted on graphs. (Linear weights.) Generally looked like blobs. No apparent trends, just one big cluster. On the polynomial trend line there might have been some bend to it.

Interestingly enough, there seems to be an optimal range for pitchers between 70-80 degrees F and 60% humidity levels. Given the conventional wisdom, they were expecting a linear relationship, not a bed in the middle, but the preliminary data suggested that pitchers are most effective in cooler, mild temps, not in the extremes of heat or cold.

Also totally against expectations: higher temperatures ran with LOWER velocity, what? And lower temps and humidity results in less movement.

Th students got an A and got to move on, while Andy was stuck with this unexpected result to look into himself.

He decided to look at all of 2011 PitchF/x data, time-stamped to the temperature that changes throughout the game. No domes allowed. Used the Tufts Weather Study from local airports for the weather data. This resulted in a huge data sample: 574,440 pitches.

He then sorted out the top 150 by pitch count, using a minimum of 12000 pitches per pitcher, and eliminatig all the “weird” pitches. The seven pitch types are: changeup, curve, cut FF, 4 seamer, 2 sueaher, sinker, slider.

Looking at each pitch separately, there is a different degree of change in pitch speed depending on the pitch. The linear change in speed per ten degrees was found to be:
changeup: 0.6 mph (per ten degrees)
cut fb 1.2 mph
4 seamer 1.6 mph
2 seam 1.2 mph
sinker : only .25 mph
slider .19

Also a positive 4 millimeters of break was added per 10% change in humidity was observed: also not what was expected. Andy admitted they don’t have an explanation of why that would be. Perhaps some physicists in the audience will look into that next.

* * * *

Heroes at the Mike: Baseball’s Longest Serving Broadcasters
presented by Gary Gillette

Stu Shea could not make it to the convention, so Gary presented in his place. The presentation was based on a database of broadcasters they have built of all the radio and TV broadcasters for individual teams (not national broadcasts). This same database is also the backbone of the upcoming book “Calling the Game.”

Since radio’s invention, baseball broadcasting has been an integral part of the American landscape. Gary asked the question: Why do some voices come to embody a region, a team, or the grand old game itself? Harry Caray came to represent the good times in baseball throughout the midwest. Caray’s unmitigated midwestern homerism might not have flown in New York, while Vin Scully’s refined tones might not have connected elsewhere.

He went on to show some interesting graphs of the longest-serving broadcasters, divided by NL, AL, radio, TV, #1 voice, color commentator, and longest serving broadcast teams. Of those who served their teams the longest, 20 of them were in one chair for over 40 years. Of course the longest serving NL radio voice is still going: Vin Scully. (Jack Buck is at #2.) In the AL teams, some might think Ernie Harwell would be at the top, but he’s all the way down at #4. Ahead of him: Herb Carneal, Denny Mathews, and Bob Elson. The two color men who worked longest for their teams? Phil Rizzuto and Ray Fosse. (I miss them both. I loved getting Fosse on streaming audio from the early days before when radio stations first had rudimentary websites.)

The longest serving duos: Brennaman and Nuxhall in the NL and Niehuas and Rizsz in the AL were the top.

Can you guess the TV announcer who has been doing it longest overall, not just for one club? It’s Ken Harrelson, the Hawk, who worked in Boston, New York, and Chicago (always for the AL team…).

The four men who worked in television who did color commentary without moving over to become the play by play guy:
Al Kaline 1976-2001 (DET)
Rick Manning, 1989 through present (CLE)
Tim McCarver (PHI-NYN-SF)
Jim Palmer, 1988 through present except for 1996 (BAL)

Gary then finished by showing the list of all broadcasters who were on air for over 40 years (Scully still tops the list, joined by Ernie Harwell and Harry Caray). Are these lists of the longest-tenured equal to the best? Some would say no. But all but three of the men on that list (and they are all men), have won the Frick award, which tells you something about longevity.

* * * *

Humphreys: We Have Underestimated Fielding Value. A Lot.

The final research presentation I went to created much buzzing, some because of the research itself and some because of the manner in which it was presented. I myself don’t feel qualified to judge whether what was presented represents a statistical breakthrough of not. I will leave that to the top statistical minds in the field, many of whom were present in the room, including Colin Wyers and Sean Lahman.

Humphreys began by saying that over the years, we’ve had a consensus view that the best players of all time can be determined based on combined batting and fielding (i.e. for hall of fame voting), but “I’m here to say we’ve been undervaluing players who are good fielders.”

-New runs “saved” by fielders (“Fielding Runs”) CAN be estimated, contrary to popular belief.
-Fielding Runs are “noisy” — but so are batting runs! People have glossed over the noise in batting runs. (At this point Humphreys asked, “How many here have heard of batting linear weights by Pete Palmer?” Nearly everyone in the room, probably 150-200 people, raised their hands. “Okay, I’m in the right place.”)
-Hardcopy and online encyclopedias have placed unfair emphasis on fielding. (I am doubting either my typing here or what he said… shouldn’t it be we’ve placed unfair emphasis on batting, not fielding? I think that was what was meant…)

Humphreys then recapped the known problems of batted-ball data (BBD). Among them, BBD for years prior to 1989 is unavailable, and from 2000 forward is proprietary. It’s inconsistent. And it’s systematically biased by human error such that it overestimates bad fielders and underestimates good ones. (People coding the BBD are biased toward thinking the ball was closer to the player’s original position. “BIS reports that it has recently solved this problem, but I’ll believe it when I see it.”)

Instead, Humphreys introduced the audience to DRA: Defensive Regression Analysis.
This was the point at which he waved his hands and said he wasn’t going to “bore” us by showing us the mathematical formulas. I feel this was a mistake: didn’t every teacher in the world tell you you have to SHOW YOUR WORK, YOUNG MAN? Also, this is a crowd that is not bored in the slightest by math or formulas. He did point out that the formulas could be found in his book, Wizardry. At best, he came off as insulting or misjudging the audience, at worst he came off as either lazy or just trying too hard to shill his book. At the time I didn’t really feel that one way or the other since I thought he’d get into more details about how DRA worked some other way, but I heard all these complaints from other people afterward. (I told you there was buzzing.)

Much of what he went on to say at that point was about how great DRA was, still without really saying what it was. He called it the first and only Fielding Runs system that is objective, open source in both data and methods, and which was applicable throughout MLB history. He also claimed it was the best Fielding Runs system in terms of minimizing bias and estimating the impact of fielding… which is sort of like saying it’s the best tasting tomato because it tastes the best. This is the point at which some grumbling in the audience really began, as impatience for something of substance grew.

Humphreys then gave us “Just a little about how DRA works.”
-Identify all non-duplicative, publicly available data (doesn’t double-count stuff)
-Measure the statistical significance of relationships between and among these variables, (eg “for each extra 100 fly outs, SS assists go up by 20”)
-make each variable independent of the others
-measure the value in runs of each variable
-Fielding Runs are just runs per play times net plays

Okay, all well and good. This was the point at which I expected he would go into detail on these points about how DRA works. But no, next we were treated to some slides about how DRA compares to Total Zone, not in terms of how the mechanism works… but just in terms of TZ’s shortcomings.

“DRA is significantly less biased than the non-batted ball system (TZ), used at and Fangraphs post-1950.”
-TZ has weaker ground ball estimator
-TZ misallocates plays on the left and right side of field
-TZ over weights errors about half a run
-TZ misallocates value between good and bad fielders on the same team more than DRA.
-Pre-1950 version of TZ is not disclosed.

According to Humphreys: Fielding Runs estimates are “noisy.” DRA identifies and adjusts for factors that systematically distort FR, but unsystematic “random” noise could nonetheless exceed a dozen runs per season. (Based on formula for standard deviation of coin flips.) But batting runs are noisy, too. George Lindsay, Pete Palmer, and Dick Cramer figured out Batting Runs circa 1960-1980.

He then introduced WOWY (With of Without You) as another metric, saying WOWY noise fades in time.
Count plays by fielder x and “his” balls in play over his entire career
Separately for each pitcher he fielded for

Test sample: all 46 fielders at 1st 2nd 3rd

Average Career Runs
TZ 96
DRA 133
WOWY 225

Average Fielding Runs per 162
TZ 11
DRA 15

Humphreys then showed a slide in which he had graphed the WAR, TZ, and DRA for many players, followed by a final column which he called “New WAR”–adjusting WAR using DRA. The slide was cut off on the projector, and I was typing fast, so hopefully the math below works out…

Using this metric, who was the greatest player of the Deadball era?

Cobb 145 0 5 150
Speaker 128 9 24 151
Wagner 126 9 10 135

WAR would have said Ty Cobb, but with New WAR, Tris Speaker moves up from #2 to #1.


Henderson 107 7 17 128
Schmidt 103 13 20 129

In this new system, Derek Jeter rates much lower than he had been, but still has a Hall of Fame number. Note, though, that Bert Campaneris rates just below Jeter. Cal Ripken is also too highly rated in TZ, Humphreys feels, as is Omar Vizquel who “just did not get to a lot of balls.” Some Hall of Fame cases are coming up before the Veterans Committee that this method rates highly: Keith Hernandez and Tony Oliva, for example.

All in all, it was a thought-provoking talk, but many left the room feeling like it was all a commercial for the book (which was mentioned no fewer than eight times in the 20 minute talk) or somehow a nearly personal attack on the work of Pete Palmer and others. I don’t think Pete would have felt that way: if new ideas don’t challenge our old ones, we’ll never progress, and I doubt Humphrey’s “bluster” was meant to come across as an attack. But since Pete wasn’t there as far as I could tell, that’s just speculation on my part. What I do know is that if the idea can survive the testing of many in the stats community, then it has merit, and if it fades from view in a couple of years, then we will know to pay no attention to the man behind the curtain.

Post a Comment

Your email is never published nor shared. Required fields are marked *