Using Plate Discipline to Estimate Walk and Strikeout Rate - Corrected

by Jeff Zimmerman on Mar 14, 2010 1:02 PM EDT

For a while, I have wondered if pitch data could be used to estimate a player's walk and strikeout rates. At Fangraphs.com, they display the percentage rate for pitches swung at and hit inside and outside the strike zone for each player (O-Swing%,Z-SWing%,O-Contact%,Z-Contact%). Using multiple variate regression, I took the 4 variables (outside swing and miss, outside contact, strike zone swing and miss, strike zone contact) and compared them to strikeout and walk percentages.

For this first run, I looked all the qualified hitters (500 PAs) from 2009. For the strikeout percentage, I got a r-squared of 0.89 and a standard deviation of 2.0% on the difference from the projected and final values. For the walk percentage, I ended up with an r-squared of 0.63 and a standard deviation of 2.0 on the difference from the projected and final values.

I went to look through this dataset and saw that some players had an actual much higher actual walk rate vice projected from a 6% to 8%. These players were all great hitters (Fielder, Pujols, A, Gonzalez) and it dawned that IBB was included in the walk rate and I needed to factor it in. I included a fifth variable in the walk calculations, IBB/PA and re-ran the regression. The results were much better. With an r-squared of 0.79 and the standard deviation of 0.15%. The highest percentage difference was 4% vice 8%. Here are the equations for estimating walk and strikeout rate:

SO% = ((-0.0407*O-Swing%)+(-0.2417 * Z-SWing%)+(-0.2429*O-Contact%)+(-0.8765*Z-Contact%) + 1.2885)*100%

BB% = ((-0.4134*O-Swing%)+(-0.0328*Z-SWing%)+(0.0216*O-Contact%)+(-0.2595*Z-Contact%)+ (1.7203*IBB per PA)+0.4217)*100%

Using these values, here are the players that I looked at the most deviate from the estimate and could be due for a correction in 2010:

Name	2010 Team	2009 Walk Rate	2009 Estimated Walk Rate	Estimated – Actual
Ichiro Suzuki	Mariners	4.7%	8.4%	3.7%
B.J. Upton	Rays	9.1%	11.9%	2.8%
Franklin Gutierrez	Mariners	7.3%	9.9%	2.6%
Jason Kubel	Twins	9.7%	12.0%	2.3%

Ben Zobrist	Rays	15.2%	11.8%	-3.4%
Nick Swisher	Yankees	16.0%	12.5%	-3.5%
Kosuke Fukudome	Cubs	15.4%	11.6%	-3.8%
Nick Johnson	Yankees	17.2%	13.3%	-3.9%

Name	2010 Team	2009 Strikeout Rate	2009 Estimated Strikeout Rate	Estimated – Actual
Brian Roberts	Orioles	17.7%	12.2%	-5.5%
David Wright	Mets	26.2%	20.7%	-5.5%
Alfonso Soriano	Cubs	24.7%	20.3%	-4.4%
Kevin Youkilis	Red Sox	25.5%	21.2%	-4.3%

Yadier Molina	Cardinals	8.1%	12.2%	4.1%
Hunter Pence	Astros	18.6%	23.0%	4.4%
Brandon Phillips	Reds	12.8%	17.2%	4.4%
Yunel Escobar	Braves	11.7%	16.5%	4.8%

I like the initial results and I am planning to add a few more years worth of data to get a better equation. I can see this formula being used to see if changes in walk and strike out rates is because of changes in plate discipline or just noise in the data.

Tweet Comment 21 comments | 1 recs |

Do you like this story?

Comments

Does this include intentional BB’s?

Bettman's Nightmare: A Blog Where Hockey Aficionados Dismantle That Mighty Empire, One Balsillie at a Time

http://bettmansnightmare.blogspot.com/

by Bettman's Nightmare on Mar 14, 2010 2:12 PM EDT reply actions

Nm, I breezed over that part on accident.

Bettman's Nightmare: A Blog Where Hockey Aficionados Dismantle That Mighty Empire, One Balsillie at a Time

http://bettmansnightmare.blogspot.com/

by Bettman's Nightmare on Mar 14, 2010 2:13 PM EDT up reply actions

This is pure, unadulterated awesome

Also might add bad calls to the list, or categorize it under noise.

I’d like to use your approach on Rotobase, if that’s OK Jeff. I’ll make a graph with career BB%, BB% and fxBB%.

I think it will rock. Let me know.

by Josh Hermsmeyer on Mar 14, 2010 2:19 PM EDT reply actions

that is, of course, assuming I can generate them from pitchf/x data. I believe it’s possible. Has anyone ever tried, or is BIS the only place for it?

by Josh Hermsmeyer on Mar 14, 2010 2:51 PM EDT up reply actions

It's quite easy to generate the FanGraphs stats with Pitch f/x

If you need help Josh, just email me.

by vivaelpujols on Mar 15, 2010 1:59 AM EDT up reply actions

Thanks Nick! I’ll take you up on it if I run into a wall :-)

by Josh Hermsmeyer on Mar 15, 2010 2:20 PM EDT up reply actions

It can be done on pitch fx and you can do it.

you may need to set the zone like a did in this article:

http://www.beyondtheboxscore.com/2009/11/5/1107712/umpire-strikezone-analysis-the

- .-. ..- … – / – …. . / .—. .-. - .. . … …

by Jeff Zimmerman on Mar 14, 2010 9:39 PM EDT up reply actions

Perfect. Just what I needed.

by Josh Hermsmeyer on Mar 15, 2010 2:20 PM EDT up reply actions

Awesome stuff

Jeff, this is fantastic.

Did you run this with 2008 numbers and see how the results looked for 2009?

Come check out Bullpen Banter
Follow Bullpen Banter on Twitter
Follow me on Twitter

Remember: baseball guys... baseball...

by JD Sussman on Mar 14, 2010 3:42 PM EDT reply actions

lol..

sorry it was a long weekend.

Come check out Bullpen Banter
Follow Bullpen Banter on Twitter
Follow me on Twitter

Remember: baseball guys... baseball...

by JD Sussman on Mar 15, 2010 2:44 AM EDT up reply actions

Published formula is not working out for me, does anyone else have this problem?

I tried numbers both in % and decimal form; the BB rate is almost always negative, since the coefficient on the O-swing and z-swing are nearly identical, and Z-Swing % is a lot higher.

Why does O-swing rate have a positive coefficient w.r.t BB%? Shouldn’t they be inversely related?

by Telegraph on Mar 14, 2010 6:33 PM EDT reply actions

I will look at them later tonight.

- .-. ..- … – / – …. . / .—. .-. - .. . … …

by Jeff Zimmerman on Mar 14, 2010 9:07 PM EDT up reply actions

Equations corrected

- .-. ..- … – / – …. . / .—. .-. - .. . … …

by Jeff Zimmerman on Mar 14, 2010 9:33 PM EDT up reply actions

Don't hitters change their approach with 1 strike and 2 strikes?

Or is this effect negligible?

by benderbrodriguez on Mar 14, 2010 8:38 PM EDT reply actions

It's not negligable or un-neglibible

It just isn’t in the scope of what Jeff is presenting in this article.

by vivaelpujols on Mar 15, 2010 1:59 AM EDT up reply actions

What nick said

- .-. ..- … – / – …. . / .—. .-. - .. . … …

by Jeff Zimmerman on Mar 15, 2010 9:52 AM EDT up reply actions

Cool stuff Jeff

I think Mike Silver did something similar back at StatSpeak, but of course we can’t access the pages now, so let’s just say you’re the first one to do this ;)

The real test is to take the guys with the biggest discrepancies in 2008 and see whether or not their 2009 numbers we’re closer to the actual walk rate or the xwalk rate. You should also look at a year to year correlation of walk rate – x walk rate. If the difference is just luck, we should see a near 0 correlation.

by vivaelpujols on Mar 15, 2010 2:02 AM EDT reply actions

I am going to get 5 years worth of data first and run the regression again

I used one year’s worth just to see if anything was actually there. Once I get the 5 years of data, it should then be easy to look at trends – year to year rates

- .-. ..- … – / – …. . / .—. .-. - .. . … …

by Jeff Zimmerman on Mar 15, 2010 9:55 AM EDT up reply actions