Walks, Strikeouts and Pitch Counts
Several pitchers like Max Scherzer, Brandon Morrow and Danny Duffy have struggled with control this season. Each has thrown a ton of pitches resulting in both walks and strikeouts. The number of pitches they have thrown has limited the amount of innings they are able to pitch in a game. By limiting the number of pitches thrown per batter, a pitcher will end up increasing the number of innings pitched over an entire season.
Many pitchers take a quite a few pitches to get through a game. The main causes for more pitches is a high number of strikeouts and walks. Each of these events take a certain number of pitches, 3 or 4 at minimum, for the event to happen. A pitcher that walks and strikes out 3 batters an inning will not allow any runs to score. Each of these innings will take 21 pitches to get through. The problem is that they will only throw 5 to 6 innings since the pitcher will be at 100 pitches near the beginning of the 5th inning. Depending on the pitcher’s pitch count limit, they will not even make to the 6th inning to qualify for a quality start.
Here are 4 pitchers from this season with 31 starts and have near the same number of pitches per game (data was taken last week so some of the number may have changed):
| Name | % K and BB | IP | Pitches per Batter | Pitches per Game |
| Ryan Dempster | 31.1% | 183.2 | 4.00 | 102.6 |
| Max Scherzer | 27.3% | 184.1 | 3.99 | 101.6 |
| Luke Hochevar | 22.8% | 198.0 | 3.73 | 100.6 |
| Randy Wolf | 22.2% | 200.2 | 3.74 | 102.3 |
Each pitcher averages just bit over 100 pitches per start. Hochevar and Wolf have less walks and strikes outs and average less pitches per batter than Dempster and Scherzer. The difference can further be seen in a 15 inning difference in IP this season from the 2 groups.
With these observations, I looked at the effects of strikeouts and walks on IP. I found a decent correlation between the number of walks and strikeouts when compared to pitches per batter (r-sqaured = 0.50) and pitches per IP (r-sqaured = 0.30). The final equation I ended up feeling comfortable with was:
Pitches/IP = 7.626 (K%) + 15.678 BB% + 13.518
Basically, pitchers averaged 13.5 pitches per inning and those numbers changed as the number of strikeouts and walks increased or decreased. Walks have twice the effect on pitches per batter than strikeouts. I expected to see the walk rate be a bit higher because of the extra pitch for a walk vice A strikeout. The doubling effect on pitches thrown shows how important it is to throw strikes.
Using the equation, here is the number of IP per season for a pitcher depending on the pitch limit they are on. The league average values (18% for the K% and 7.5% for the BB%) are used. The results of increasing or decreasing each value 2% is also shown.
| BB% | K% | 120 P/G | 110 P/G | 100 P/G |
| 7.5% | 18.0% | 239.0 | 219.1 | 199.2 |
| 5.5% | 18.0% | 243.8 | 223.4 | 203.1 |
| 9.5% | 18.0% | 234.4 | 214.9 | 195.4 |
| 7.5% | 16.0% | 241.3 | 221.2 | 201.1 |
| 7.5% | 20.0% | 236.8 | 217.0 | 197.3 |
The main item that sticks out is that small changes don’t make that much of a difference over the season. A change in BB% by 2% changes the total by only 4 IP. Not all changes are so small as seen in the 4 pitchers I looked at earlier. Here is the number of IP that they would have been predicted throw over 31 starts and 100 IP given their K% and BB%:
| Name | IP Season | IP Predicted |
| Ryan Dempster | 186.1 | 183.2 |
| Max Scherzer | 190.7 | 184.1 |
| Luke Hochevar | 194.9 | 198.0 |
| Randy Wolf | 193.3 | 200.2 |
The predicted difference is not as much as the actual difference, but it is a measurable difference.
Pitchers that have high pitch counts from walks and strikeouts will see an effect on the number innings they are able to throw. Over an entire season the difference could end up being the equivalent of two extra games worth of IP.
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.
quite*
All of this is pretty intuitive and expected. I’m curious about the guys who deviate the most from the expected results. That is, whether some players have an ability (or inability) to maximize the efficiency of their pitches, outside of merely allowing fewer K’s and walks.
If I was a little better at SQL, I might be able to look at a few theories, but right now I will have to live with just K and BB rates.
Jeff, I wasn’t challenging you to come up with explanations, just to point out the outliers if you graph your expected pitch counts against the actual pitch counts.
For instance, Health Bell this year has completely average K (19.0%) and BB (8.5%) rates, yet has been throwing a ton of pitches (4.3 per batter).
i’m actually surprised that the coefficient for K% you got wasn’t 0. i mean, K’s generally take more pitches than balls put in play, but they’re also guaranteed outs as opposed to ~2/3 outs.
in particular, I’m not sure what to think in light of this article from a few years ago: http://www.hardballtimes.com/main/article/the-kazmir-conundrum/
Without reading the THT article (I will), I think they get more strikes(pitchers) even when the get a ball in play. Just the nature of a strikeout pitcher.
So in theory a “pitch to contact” pitcher should pitch more innings than a strikeout pitcher does given equal H/9.
I was going to look at H/9 leaderboards but fangraphs doesn’t have it. I guess i’ll give more hits to baseball reference
Yes given the same number of pitches per start (see Mark Buerle or however he spells his name)
http://www.centives.net/S/2011/the-centives-line-up-as-inspired-by-moneyball/#more-1322 heres a story fangraphs fans might be interested in
Tango made the following formula a while back–
pitches = 3.3xPA+1.5xSO+2.2xBB. Looking at 2000-2010, min 180 IP, here are the guys that the two formulas disagree about the most [per 180 IP]:
2000 Pedro +394
2004 JSantana +274
2004 RJohnson +272
2002 Pedro +264
2001 RJohnson +247
2003 JSchmidt +233
2005 Pedro +231
2005 JSantana +213
2002 RJohnson +213
2002 Schilling +209
2005 Clemens +206
2009 Lincecum +200
…
2002 Sturtze -200
2004 Lowe -209
2000 Haynes -209
2006 Silva -212
2005 Francis -212
2003 JJennings -213
2002 Sparks -217
2004 Lohse -217
2000 Lima -223
2004 JJennings -242
So, at least relative to Tango’s stat, this metric is assigning a lot more pitches to guys with high SO+BB rates and alot fewer pitches to those with low SO+BB rates.
Using the sample of 715 pitchers, sorted by (SO+BB) rates and divided into five equal sized piles:
The top 20% in (SO+BB) rate averaged 32.1% and were estimated to have 2.5% more pitches than Tango’s formula.
The next group averaged 0.7% more estimated pitches.
The middle group, -0.1%
Next, -1.4%
Last, -2.0%
Granted– without a dB full of actual pitch count data [which I don’t have], no way to tell which is the more accurate estimator…
I have the pitch data, let me see how they compare. It may take me until tomorrow to find the time.
FYI, pitch counts are available on the “batted ball” tab on this website.
Thanks– exporting that data from fangraphs to import into my dB… will follow up with results.
My existing dataset varied a bit from what I got from fg; I didn’t bother summing up partial seasons, so a player with 180+ IP for one team and then some number for another, only counts the 180 + IP portion in my data and the full year data in fangraphs.
Also, sicne the fg data only went back to 2002, the sample shrunk from 715 to 592, plus looks like another 19 player seasons didn’t match up [likely players who have a Jr or something in their name; had to join baseball-databank ‘first’ and ‘last’ name fields to the fg name data.]
Anyways, on average the formula from this thread was off by 122 pitches and Tango’s 93.
Splitting up into three equal piles, by (SO+BB) rate:
The top group averaged being off by 126 pitches with this formula and 78 with Tango.
The middle group, 120 and 93. The bottom group, 120 and 108.
So, this method was about as good at estimating pitches regardless of SO+BB rates, while Tango’s improves as SO+BB rates increase.
I got a very similar formula for pitchers per batter:
P/batter=1.785(K%) + 3670(BB%)+3.187
The big difference is moving it from PA to IP
Using the above equation on 2011 data (I used 2008 to 2010 data to get my equation) here the results
Estimated Tango: 2889 pitches
Estimated Zimmerman: 2925 pitches
Actual: 2926 pitches
R-squared for the data and results
Tango (96.2%)
Zimmerman (96.5%)
Almost the same results. My equation for the IP will be off more than per batter, but I find it a little more useful for IP predictions.
I believe you have “IP Season” and “IP Predicted” reversed in your final table, or at least that’s what would make the inning totals match the first table.
“Basically, pitchers averaged 13.5 pitches per inning and those numbers changed as the number of strikeouts and walks increased or decreased”
You’re probably just wording things poorly here, but that doesn’t jibe at all with your formula.
Your formula says “Pitchers who don’t strike out or walk anyone average 13.5 pitches per inning and those numbers…”
Nice article Jeff. Although I’d interpret your equation as “pitchers average 13.5 pitches per inning plus 7.6*(K’s/inn) plus 15.7*(walks per inning).
Also, would it make sense to throw in HR’s/inn since we’re discussing things that raise pitch counts? Pitchers can control HR’s (theoretically) and they definitely increase pitch counts.