pERA Update From SABR Analytics Presentation

by
March 13, 2017

This past Thursday, I spoke at the SABR Analytics conference on my per pitch valuations (pERA).  I originally created them to form an understandable framework for comparing prospect pitching grades and major league results. Some byproducts of the work became useful like the effects of dropping a pitch. Today, I will make available new information I provided at the conference.

For the readers who aren’t familiar with the original work, it can be read in its 2500 word entirety in this previous article. Here is a summary.

  • The key is to give each pitch an ERA value (pERA) based on the pitch’s swinging strike and groundball rates. All the values are based on the average values for starting pitcher. Closers will have higher grades because their stuff plays better coming out of the bullpen.
  • The pitcher’s control is determined from their walk rate which is separate from the pitch grades.
  • Each pitch is placed on the 20-80 scale with 50 being average, 80 great, and 20 horrible.

After creating the framework, I attempted to line up pitching prospect grades with the per-pitch-results. I wasn’t able to get the prospect and pitch data to match. I assumed the prospect grades would be bullish. People creating prospect lists are going to insert the best reports they collect and therefore be optimistic. Even adjusting for the expected bullish values, the grades didn’t meld.

The biggest cause for the disconnect is that pitchers quickly change. Most of the scouting publications collect their data after the season ends until they are published just after January 1st. Pitchers aren’t stable. They add a new pitch. Lose some velocity. Find control. These changes can be known as soon as their first spring training game but fantasy owners following prospects are stuck with dated data.

The other main issue is fastball grading. Grades are about 100% based on velocity readings, not their movement. I’ve found context needs to be added to the velocity measurements. For reference, this table contains the standard fastball grading scale.

Pitching Prospect Fastball Grades
Grade Tool Is Called Fastball Velo
80 80 97
75 96
70 Plus Plus 95
65 94
60 Plus 93
55 Above Avg 92
50 Avg 90-91
45 Below Avg 89
40 88
35 87
30 86

To show the disconnect, here are two examples of the scale failing. Kyle Hendricks fastball averages 88 mph. This puts his fastball at standard 40 grade and should barely get any MLB consideration. Last season, his fastball results were amazing with a 57% GB% and an 12% swinging strike rate. These values put his pitch grade at 67 or in the 94-95 mph fastball bucket.

On the other hand, Wily Peralta averaged 95 mph on his fastball. His groundball rate was 53% and he only generated a 6% SwStr%. These results are equivalent to a 46 grade or at 89 mph on the standard scale. He’s supposed to have above average results but the pitch is barely playable.

For these reasons, I needed to find out the characteristics of a good pitch. For this analysis, I looked at pitch break, regular spin, effective spin, velocity, and velocity difference from fastball for change. Manipulating various inputs, I came up with some characteristics for individual pitch quality. Here are the results so far.

Four-seam fastball

Most of this pitch’s results could be determined by spin and velocity.

4-Seam Fastball Grades Based on Velocity and Spin
RPM/MPH 87 88 89 90 91 92 93 94 95 96 97 98 99
1500 31 33 36 38 40 43 45 47 49 50 52 54 56
1600 33 35 37 40 42 44 46 48 50 52 54 55 57
1700 34 37 39 41 43 45 47 49 51 53 55 56 58
1800 36 38 41 43 45 47 49 51 52 54 56 57 59
1900 38 40 42 44 46 48 50 52 54 55 57 59 60
2000 39 42 44 46 48 50 51 53 55 57 58 60 61
2100 41 43 45 47 49 51 53 54 56 58 59 61 62
2200 42 45 47 48 50 52 54 56 57 59 60 62 63
2300 44 46 48 50 52 53 55 57 58 60 61 63 64
2400 45 47 49 51 53 55 56 58 59 61 62 64 65
2500 47 49 51 52 54 56 57 59 60 62 63 65 66
2600 48 50 52 54 55 57 58 60 61 63 64 66 67
2700 49 51 53 55 56 58 60 61 62 64 65 66 68

Pitchers can get average results (50 grade) with a 88-mph fastball up to a 96-mph fastball.

Two-seam fastball

I found a small correlation with velocity but that was all.

Cutter

I found no consistent traits.

Curveball

Along with the four-seamer, the curve was the only pitch in which the results could be accurately measured.

Curveball Grades Based on Velocity and Spin
RPM/MPH 66 68 70 72 74 76 78 80 82 84 86
2000 30 33 36 39 42 44 47 50 53 56 58
2100 31 34 36 39 42 45 48 50 53 56 59
2200 31 34 37 40 42 45 48 51 54 56 59
2300 32 35 37 40 43 46 49 51 54 57 60
2400 32 35 38 41 43 46 49 52 55 57 60
2500 33 35 38 41 44 47 49 52 55 58 61
2600 33 36 39 41 44 47 50 53 55 58 61
2700 34 36 39 42 45 47 50 53 56 59 61
2800 34 37 40 42 45 48 51 53 56 59 62
2900 34 37 40 43 46 48 51 54 57 60 62
3000 35 38 40 43 46 49 52 54 57 60 63

The harder the curve is thrown, the better the results.

Slider

A higher velocity gives the pitch some improved results.

Change

I found the bigger the difference in velocity from the pitcher’s fastball, the more effective the change.
—
While informative, the pitch characteristics are only useful for four-seam/curveball pitchers. I need to investigate more on what makes the other pitches work. I acquired a few ideas from people after the speech. I was informed I need to add the difference in pitch movement, break, and velocity from the starter’s fastball. Before, I only included the velocity difference with change ups. At some point in the near future, I will reexamine the data to see if I can improve picking out good pitches.

Another area I can improve is adding in more outside factors besides pitch control like deception, repeatability, and tunneling.

I am not giving up hope using preexisting prospect pitch grades to project a pitcher. For now, I am still working on the disconnect between what scouting publications report and major league results. It has taken me a few steps more than I would have liked but I feel I am getting closer. Hopefully soon, I can find a way to use published scouting reports or assume they are completely useless in their current format.





 

Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.

5 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
MattabattacolaMember since 2020
8 years ago

I was playing around on Baseball Savant and was seeing some of the same results. It seems weird just based on the eye-test. Is this the beginning of the argument for sequencing?

1
Jeff ZimmermanFanGraphs Staff
8 years ago
Reply to  Mattabattacola

I have thought about sequencing and it just makes my brain hurt.

1
kbn
8 years ago
Reply to  Jeff Zimmerman

Markov Chains could simplify the sequencing problem considerably, allowing you to tractably calculate the stochastic value of the sequence distributions based on prior knowledge of the pitch values. The edge weights would presumably be derived from a linear weighting of league-wide results (e.g. a high fastball followed by a 12-6 curve is probably good for almost every pitcher) paired with pitcher-specific results as they deviate from the predicted model.

This would be straight machine learning, but I think it would produce a relatively useful model. It’s also possibly more general than it needs to be. Over time, it may become apparent that certain pitch sequences are simply always good, strengthening the league weights and reducing the reliance on individual factors (and thus, sample size problems).

0