Follow BallHype on Twitter
The Hardball Times
found this 5/14/2008
-
MGL Nice work Josh.
A couple of things:
It would have been nice if you had a chart with the correlations, or did I overlook that in the article?
It is always important to compare pitchers' pitches in the context of how often they throw them. Let me explain. Let's say that pitcher A has exactly the same curve ball as pitcher B. But pitcher A throws it 15% of the time in fastball counts (probably typical) and pitcher B throws it 40% of the time in fastball counts (like a Bronson Arroyo). And let's say that at all other counts they throw it exactly the same percentage of time.
Well, it is going to look like pitcher B is more effective with his curve ball, which he is, but that is NOT because it is a better curveball (obviously, since we said that they have the same one). It is because it is less expected than the ones by pitcher A.
So we have to really careful about comparing characteristics of pitches among pitchers, unless we control for the percentage of time those pitches are thrown at the various counts.
The second thingI want to say is that just because we find that a pitcher's (like Issy) pitches are different, that does not necessarily mean that he is doing something "wrong" that may or may not be correctable. It may simply be that he has gotten "unlucky" in how his pitches are acting. In other words, we have to first look at the random fluctuation in how pitches act. Maybe ALL pitchers curveballs fluctuate randomly in how much they break, the release points, speed, etc. (to some extent they have to). Then when we determine that, we can make some kind of inference about whether a good or bad spate of pitches may be just sheeer luck or something fundamental that the pitcher is doing right or wrong. We also know that all the pitches do different things in different venues, based on altitude, temperature, etc. Again, maybe this year Issy's average venue thus far is different from his average venue last year, such that he is doing nothing differently this year than last. Again, we have to be careful in making conclusions about something a pitcher may or not be doing, even based on the characteristics of his pitches. Pitch characteristics no doubt have random fluctuations themsleves.
-
joshkalk I am glad you liked it MGL. I could provide you with the chart but I did leave that out because the article was getting long and I didn't think it would help the average reader. Obviously, you aren't the average reader though.
You point about normalizing the data is correct though. Right now I am correlating the different attributes to the best metric of success I have, runs100. As John himself pointed out in the ballhype chat after his article no park corrections or BABIP corrections have been made. My guess is an adjusted runs100 that corrects for the above and the pitch expectancy you mention above would produce better results. I am working on that right now and hope to have that completed by my next article. This is another reason I didn't add the correlations chart as hopefully things will become even sharper soon.
You are correct that the 2008 data isn't corrected yet. The 2007 corrections do take into account the venues by correcting for atmospheric conditions such as temperature and altitude. Once the 2008 data is corrected obviously we will have a better idea. That said, it is clear that Izzy's release point for his curve is different from his release point of his fastball in 2008. Because of that, it is very likely that the hump on his curve has increased and possible that his vertical/horizontal movements have changed as well. While I haven't published most of the side views I have created a 2007/2008 side view for all pitchers and I can safely say that, even without correction to the 2008 data, Izzy's curve stands out as a pitch that has changed since 2007. While possible, I don't believe that it is random fluctions at this point.
-
walshj58 Nice work, Josh.
You didn't talk about curveball speed much, but I would expect that a large hump is highly correlated with speed as well. If the ball is going slower, gravity has more time to act on it. So, the ball needs to be thrown higher if it's going to end up in the strike zone. Haren's CB is around 10 mph faster than Lilly's.
Of course, less speed will often mean more spin (and bigger break), since more of the available energy is going into rotation and less into speed.
BTW, I too would have liked to see the correlation numbers.
-
joshkalk John,
You are absolutely correct that the speed is going to correlate with the hump size, 0.68 correlation in fact. I should have mentioned that when I talked about Haren's curve. Going forward, I am probably going to need to uncorrelate these variables for best results. That said, here is the list of the correlations to runs100:
release point 0.15
max hump horizontal value 0.34
hump size 0.32
Adair's hump 0.08
speed 0.12
vertical movement 0.15
horizontal movement 0.04
speed diffrence 0.28
vertical movement difference 0.22
horizontal movement difference 0.05
I basically had worded out this table in one of my paragraphs but I didn't add it because I didn't want to scare any readers off with more length and more numbers. I kind of feel like I am walking a fine line with some of these articles. I want them to be rigorous enough for you guys but I also want them to be readable for people who care less about the numbers.
-
Ike Nice work Josh.
However, when running a correlation between runs100 and curveball properties that are measured against fastball properties (hump size, location), I think you would do better to compare to a fastball up in the zone or even a fastball up out of the zone. Most pitchers tend to work down in the zone with their fastballs, but come upstairs every so often. A good knee-buckler would give the appearance of a high fastball before having the bottom drop out of it. Comparing to an average fastball tends to bias yourself to comparing to fastballs low, and while the curve in the dirt is a useful pitch at times (which is the only way you can make a curve look like a low fastball for a bit), often it will break too soon.
You might even want to put in your correlation computation of what makes a good curveball the rate at which a pitcher goes upstairs with his fastball.
-
joshkalk Yeah I did play around with several other things besides just average fastball. I am aware that pitchers with a big breaking curve will go upstairs with their fastballs a lot to "hide" the curve but wasn't exactly sure how to go about measuring that. I will keep playing around though. Thanks for the suggestion.
josh
-
-
MGL BTW, I love those side views!
What I meant, which doesn't have that much to do with your article, was that when I see the run value of a particular pitch by a certain pitcher, assuming for a second that the sample size is large enough to ignore sample variation, what I am seeing is how good the pitch is in the context of how often he throws it in the various counts. So when I see the value a pitch from a pitcher I always want to know how often he throws it at the various counts in order to compare that pitch to similar piches by other pitchers who throw it about the same percentage of time in the various counts. That is the only way I can determine how "good" the pitch is. As I have mentioned before, even if I have a medicore curveball, if I throw it in fastball counts, it may have a low run value. That is not because of the quality of the pitch of course; it is because of the surprise factor. One of the important things in pitch f/x work (among hundreds) is to compare everyone's curves, sliders, etc. But the only way to do that is to somehow control for the frequencies of when they are thrown. If we don't, then we are not really comparing the quality of pitches, unless by "quality," we mean raw quality AND the context in which it is thrown. That definition is fine, but the distincion needs to be made when presenting the data. Once we do that, then we can go back and determine how the surprise factor changes the value of the various pitches! That way, we can begin to really see what is going on with each pitcher. In other words, we can say for pitcher A, "He has a B- curve ball, quality-wise, but he throws it 20% of the time in counts where the average pitcher throws one only 10% of the time, and by virtue of that he gets an extra .03 in run value."
Again, great work!
-
joshkalk MGL,
Thanks for the added explaination. I think I understand exactly what you mean now. I also think that I have a good way of normalizing for this effect but I would like to discuss this with you further. Could you email me at: joshkalk-at-gmail.com?
Thanks,
josh
-
-
mysportsrumors another amazing article, love your breakdowns!-
joshkalk Thanks for your kind words. Sliders will be up next week and hopefully there will be some really interesting things in that article.
-
-
Shandrax Hello,
I came to this article a little late, because I found the one on the slider first. While I definitely agree with the analysis on the slider - because it is essentially thrown "off a fastball", it seems to me that the curveball is successful for different reasons.
My prime example for this theory is Nolan Ryan. His curveball had a huge hump and still it was very effective. Why? Because he had the amazing fastball.
The reason for this successful combination of hard fastball and slow curve is that the curve requires a different way to hit the ball. You can't just swing at the projected plane like you do with a fastball, you have to somehow swing under the ball and catch it on the break.
The hitter has to expect fastball and adjust to the curve and he just can't do that, even if he recognizes the hump which essentially gives away the pitch.
There is also another point. Curveballs with big hump get thrown for a strike since the hitter thinks "too high" at first, while curveballs that start on a flat trajectory usually end up as a ball low, so the batter either takes it or misses it.
Throwing it for a strike or to get someone chasing is essentially a totally different approach.
Now I am not claiming to have the ultimate truth on that, but as a hitter and as a pitcher, this is how it appeared to me from my own experience.
-
joshkalk Thanks for this information. I need to clean up a few things befor eI come back to looking at these pitches but it certainly is possible that for some pitchers like Ryan could get away with "tipping" his curve with a big hump. These evaluations are just correlations not absolutes and will need to be improved a bit before they become more useful.
-
Curveball v Fastball
THE BOOK--Playing The Percentages In Baseball —
... Josh Kalk walks us through it. A link at the bottom (references) showing the results, would be a welcome addition. Otherwise, excellent article. ...
Smells Like Padres in Here - Wednesday Edition
Gaslamp Ball —
... is May after all.
Gerut likes playing every day.
Kouz finally got the night off and Edgar filled in nicely, thank you very much.
Edmonds appears a near certainty to be a Cub by tomorrow. This is really worth a quick read because there is a mention of a personality clash while he was with the Padres. Would sort of explain the willingness to eat his contract so quickly.
I love any sort of discussion about curveballs.
And another HT piece on Clayton Kershaw, a guy we should be seeing in LA ...
Friday links
Friar Forecast —
Linked to a couple articles when I came back on Wednesday, so just a few today ….
More Maddux analysis from Eric Seidman — Haven’t really read it yet, but I enjoy Eric’s stuff.
By the Numbers – Speaking of Seidman, he has an article in one of the recently posted “By the Numbers.” There’s also more from Victor Wang on the value of prospects, which I’ve linked to in the past.
Anatomy of a curveball – Josh Kalk with more PITCHf/x work with cool graphs showing a curve and ...
A Little Fun With Erik Bedard And PITCHf/x
Lookout Landing —
... To change course a little bit, now I want to show you something that I first saw used by Josh Kalk over at THT, just because I think it's super cool. The following is what Erik Bedard's average fastball and curveball from this afternoon look like from the side: ...
Brandon Backe in 2008
The Crawfish Boxes —
... So the hypothesis wasn't perfect, but I think the pattern is apparent. Backe is a much better pitcher when he throws a tight curveball that stays hidden. Just about all the games we looked at show that if he had an effective curveball, he had a strong outing and we know from pitch/fx that two things are important for the curveball: vertical drop and the horizontal plane. Thus, it's pretty clear that if Brandon Backe wants to avoid being optioned back to Round Rock in 2009 (or just released), he needs to figure out how to have a consistently ...
