THE BOOK--Playing The Percentages In Baseball

Hmmm… so if there is some persistent advantage to varying or not varying, would a simple raw count of the number of vary-ers (variers?) and non-vary-ers tell you which one was better, all else being equal?

Great explanation by the way.

Actually nevermind.  Because while in your example, the ratio of minor league vary-ers and non-vary-ers is equal, that might not be the case in reality, I suppose.

Also… the word/number confirmation for me is a contraction - “youre82”.  Weird!

The analogy with velocity is a good one. Isn’t it true, though, that harder-throwing pitchers are better overall?

Fangraphs keeps track of average fastball velocity. In 2008 there are 93 starting pitchers with at least 150 IP.

1. Ubaldo Jimenez, 94.9 MPH
2. Felix Hernandez, 94.6
3. Josh Beckett, 94.4
...
91. Livan Hernandez, 83.7
92. Jamie Moyer, 81.2
93. Tim Wakefield, 73.1

The correlation between fastball velocity and ERA is r = -0.31. The linear regression equation says that you lose 0.08 points of ERA for each extra MPH on the fastball. An average fastball velocity of 95 corresponds to a 3.65 ERA, and an average fastball velocity of 85 corresponds to a 4.42 ERA.

All this is in spite of the perfectly valid selective sampling argument.

We should expect a .25 or so change in ERA, not .08.  Indeed, you are making MGL’s point!

That is, let’s presume that the only difference between a pitcher who starts or relievers (Joba, etc) is that his fastball speed is faster as a reliever.  Everything else he does is the same.  Since we know that pitchers improve their ERA by 1.00, and if the only thing different is their fastball speed, and if that difference is 4 mph, then we’ll ascribe an impact of .25 ERA per mph difference.

(Feel free to substitute your own reasonable numbers.)

On the other hand, you have found a correlation of only .08 for mph difference.  The reason is that those with a lower fastball speed make up for the difference by being better at something else, be it knucklers, guile, or extra movement on pitches.

If on the other hand we allowed ALL pitchers who threw at least 75 mph in MLB, then we would indeed find that you would get a .25 ERA difference per mph, because now you have introduced a whole bunch of guys who can barely throw the ball who also have nothing else going for them.

Tango, the difference in fastball speed for pitchers between relieving and starting roles ranges from about 0-2 mph and averages 1 mph (2005-2008). 

The difference in runs allowed does not seem to correlate with the difference in fastball speed, however.  The mix of pitch types also stays very nearly the same, on average, for pitchers changing between the two roles (based on BIS data from Fangraphs).

What’s the difference then?  Movement?  Or Location?  Because that’s the only two things left.

One of the main differences between effectiveness of starters and relievers is the “times through the order.” If we want to compare them, we at least have to compare a starter’s first time through the order, say his first two innings, with his performance when he relieves (short relief).  That won’t work that well, though, since starters pitch differently in the first couple of innings, according to the pitch f/x data.  For example, I think one of the pitch f/x researchers has shown that starters use more fastballs in the first couple of innings.  They may also have a lower velocity (in the first couple of innings), I am not sure and I don’t remember.

Anyway my point still stands I think, despite some resistance and defensiveness from Josh.  I hope that all of the pitch f/s researchers, or at least the offending ones, stop analyzing one parameter of a pitcher or batter without controlling for the other parameters or using the “delta approach” with individual players (which ostensibly and ideally controls for everything else) and coming to the conclusion that the one parameter correlates or does not correlate (and by how much) with overall success.  It is somewhat a waste of an article I am afraid.

What’s the difference then?  Movement?  Or Location?  Because that’s the only two things left.

I don’t know.  If I knew that answer, you’d be seeing an article from me, but I don’t know yet.  I think fastball speed probably does have some impact, and I’m just not sampling or measuring quite carefully enough to see it.  Times through the order probably also has an effect.

I can think of other things besides movement and location (which I haven’t looked at, yet, btw).  Relievers pitch later in the game, so it’s probably cooler on average than when the starter is pitching.  Do relievers have the platoon advantage more often?  I forget if you guys looked at that in the Book--I’ll have to go check.  The strike zone might get bigger toward the end of the game, although I believe Dan Turkenkopf didn’t find that to be so when he investigated it.  Hitters might change their approach against relievers versus starters (more aggressive at the end of games, for example, although that doesn’t really fit with relievers allowing more walks).

What’s the difference then?  Movement?  Or Location?  Because that’s the only two things left.

What makes some pitchers successful, and other pitchers that seem similar less successful is one of the most interesting questions in baseball research, along with the related question of what makes a pitcher very successful on some occasions and much less successful on others.  It is a particularly difficult question to research, however, because the usual methods of aggregating many pitchers together will obscure the real differences that exist between pitchers who have been able to maximize their personal success by accentuating their own individual strengths and and minimizing their weaknesses.  The other complication is that one strategy may work against one batter but not another.  So even aggregating all the pitches of a single pitcher may obscure the different methods that a pitcher uses to be successful against individual batters.

Again, it might be perfectly appropriate to teach a young (or any age) pitcher to vary his release point as that may in fact contribute to more success.  Your data and analysis in no way shape or form tells us that any one pitcher is not better off varying his release point.  It only tells us that pitchers that do vary their release point are just as successful as pitchers that don’t - and again, the pitchers that don’t just might be better pitchers otherwise - in spite of their constant release point, not because of it!

This is a great point.

All of this discussion reminds me of studies from evolutionary biology on the beneficial nature of crypticity. I realize, tangent, but cut me some slack for a second.

Consider the case of a moth (prey) trying to hide from a blue jay (predator). One would think that it was best for the moth to be as cryptic or camouflaged as possible. There is a wing pattern in the environment that will allow it to be the most cryptic, so that it is hardest to see. But, blue jays are very good at finding moths, and their perceptual systems have also co-evolved to be very good at detecting them.

This problem was taken back into the lab, and computerized moth-like things were displayed on the screen and allowed to “evolve” while real blue jays tried to find them. What they found is that, like the real moths, computerized moths did NOT converge on the “optimal” cryptic solution. Instead, the most optimal solution for these computerized moths was to evolve to be not only very cryptic, but also very different from other “moths” in the experiment.

Why? Blue Jays form a search image while searching through the display. Their perceptual systems become tuned to finding these moths, even when they are *extremely* hard to find (think: where’s waldo times 10000). So, even though the task is very difficult, they adjust and will succeed.

On the other hand, when the moths are not only cryptic but also very different from one another, it becomes a much more difficult task for the blue jays. Although their perceptual systems could solve any one of these tasks alone, it becomes very hard for them to identify any one optimal search strategy and so all of the moths collectively are harder to find.

I think this is actually a fairly good analogy for what makes a successful pitcher. Sure, it’s good to have pinpoint location and accuracy and speed, and all of these things are important - just like overall crypticity for the moth trying to hide from the blue jay. But the natural (and extreme) variation in pitchers (and moths) suggests that being *different* from everyone else is also critically important.

So, it may be the case that release point variation (or pitch type variation or pitch velocity variation or whatever) does not help individual pitchers. But, it may be the case that variation in this regard helps pitchers as a whole.

This was a very long rambling post that I hope made some amount of sense.