Puig

JH,

what I'm saying is that hitting is not like a coin toss, in which the next flip is completely independent of the last flip.  The probability of a hit is a function of the hitter's performance in the last at-bat, or at-bats.  At least this is my assertion.  However small the sample size, I believe streaks and slumps exist.  Of course they end, but while the streak (or slump) is going the odds are with you if you take this into account in decision making.  They only go against you the time it ends.

If a hitter goes 4 for 4 in Wednesday's game and 4 for 4 in Thursday's game and 4 for 4 in Friday's game, there is no predictive value in that to say he will get a hit in his first at-bat on Saturday. He may feel more confident and in rhythm at the plate, but there is no way to quantify that he will get a hit strictly based on the sample size of the previous streak he had. As a decision-maker, it would be a bad idea to base in-game decisions on small sample sizes of results because of this. I'm not saying streaks don't exist, I'm saying they aren't usable predictive values.

Not for you JH, but maybe this leads to an interesting response. How do you know when the hot/cold streaks start and end and when you're in them.

If the type of streaks that you guys think exist do exist, you should be able to spot them and tell us when they started while in them and notice when they end as they are ending.

JH,

I'm not sure if you and I are disagreeing or getting crossed on terminology.  I am not suggesting that there's an ability to predict with certainty the outcome of an at-bat based on recent performance, but I do believe that the probability of the outcome changes based on recent performance.  Perhaps the math being used to analyze this today is too simple.  Algebra shouldn't be used to solve a calculus problem.

Serious question: what's the difference between predicting the certainty of an outcome and the probability of an outcome?

If you believe that being hot leads to continuing to be hot, why do hot streaks ever end?

If the results of the previous PA effect the next one, why can't I use those results predicitvely?  

Hint, if I could use those results predictively, I could retroactively look at the beginning of a series of ABs that had taken place and predict when a hitter would get hot or cold over the rest of my series of AB, but that doesn't actually work.

If it were possible to predict hot and cold streaks, the methodology for doing so would be worth millions of dollars. There are tons of really smart math people working in baseball and other lines of work who would kill for the chance to make money exploiting this power if it existed. Teams would routinely change their lineups substantially based on who's hot and who's not.  Yet, the Braves gave Dan Uggla 660 PAs of hitting .178 before pulling the plug.

Just for discussion purposes.  Barry Bonds was intentionally walked with his team behind by 2 runs with the bases loaded and two outs.

i would not have done that, but understand the logic. tying run is on 2B,You are facing the very best hitter and power hitter in baseball.  A much less hitter is on deck.  A hit by Bonds either ties or ends the game and a hit, walk, error, etc., ends the game with Brent Mayne hitting.

No matter what the numbers say, the manager's gut told him he thought the best chance to win the game is by facing Mayne with two outs and the tying run at 3B and winning run at 2B rather than pitch to Bonds with the tying run at 2B and the winning run at 1B. Both will end the game if they make an out.  Though I wouldn't have done that, it actually makes some sense In that situation.  Still, It sure took some courage to do it.  Obviously, this would not have happened if the hitter would have been anyone other than Bonds. The gamble did pay off for Showalter!

Joe Madden did pretty much the same thing with Josh Hamilton.  We are talking about two very good MLB managers that have been around the block a few times.

Tying run was on 2nd, it takes a 3B or HR to win the game (JT Snow is not scoring on very many doubles, heck he's probably turning some Bonds triples into doubles).

That you insist that going with your gut instead of logic is correct just because it worked out is whey Vegas exists. It's also why I made a pretty decent living playing poker for a while.

Really smart people make mistakes all the time. And you can get away with making some mistakes and still be very good at what you do, especially if your competitors make more mistakes than you do.  That doesn't mean you shouldn't try to learn from them, analyze them, and where appropriate correct them.

Why do you not allow for the Manager's gut to be wrong?

Because of this.

"Walking Barry Bonds with the bases loaded was never a good decision. Ever."

I assume,JH, you say that because you believe you know (illogically, I'd say) that there were absolutely no other variables, beyond baseball stats, that could have impacted the outcome?

But what if, hypothetically, the manager knew that the next guy up had a fractured wrist ... or a post-game appointment with the police who intended to put him in handcuffs ... or, God forbid: Hemorrhoids?

Would any of those situations make walking Bonds a good decision?

Not statistically, I suppose.

But aren't there an infinite number of variables that should influence managers' decisions? Up to and including, maybe -- streaks?

I think the cocoon you're weaving and getting trapped inside is putting complete faith in numerical data. It's a game played and coached by humans, right? And that just obliterates a whole lot of stats.

In my experience, great leaders use data but never rely on it.

(silly examples, but only to make a point)

Walking Bonds could be a good decision if Mayne owed money to LV bookies and was throwing the game, sure. I mean if you want to make up completely ridiculous hypotheticals I'm sure we could come up with some scenarios where walking Bonds had a positive expectation.

But in the absence of any evidence for any of those hypotheticals, there's no reason to give them any weight in analyzing the situation. And since we have plenty of evidence that "hot streaks" and "clutchness" don't actually exist in any way that's exploitable on the field, there's no reason to give them any weight either.

Here's something I'm sure will just make this argument worse. Results don't matter. Making the best possible decision given the available evidence (and I mean that word in the most concrete possible sense) is all that matters. If you throw Barry Bonds the perfect pitch for getting him out, and he hits a HR anyway, it doesn't matter, because you gave yourself the best possible chance to win.  If you walk him, you haven't given yourself the best chance to win, and if you happen to win anyway, that doesn't make your decision better, or right, after the fact.

If you play baseball (or manage it) long enough (or really do anything productive), and you aren't always making the best pitch, or taking the best swing, or making the best managerial decision you can, you're leaving money on the table, and someone else will benefit from that. If you aren't constantly reviewing the processes you're using to be the best you can be, you're going to miss the opportunities to correct your shortcomings, and someone will eventually come along and eat your lunch.

JH,

I was making a distinction between "predicting with certainty" versus understanding the probability of an occurrence happening.  I recognize that no amount of analysis will help us know with certainty if a batter will get a hit next time up, but we can recognize that a .300 hitter's probability of getting a hit varies wildly (at times) above/below 30%.

There are an infinite number of variables that slightly change a players true talent on any given pitch, in fact, players true talent changes with every single movement, but, the changes are so small and so quick that you can't worry about it. So in actuality, what you're left with is players hitting and pitching their historical averages.

This isn't something the 3 of us just made up. Streaks may exist, a player may be 'in the zone' but it winds up only being something you can point to after the fact and that feeling is transient and carries no predictive value and is in no way actionable on a Major League Baseball field. This has been proven over and over again. Managers cannot identify a player on a hot streak and those are the guys being paid to and theoretically have the best information!

"If it were possible to predict hot and cold streaks, the methodology for doing so would be worth millions of dollars. There are tons of really smart math people working in baseball and other lines of work who would kill for the chance to make money exploiting this power if it existed."

jacjakatk,

Yes, I agree with this.  Likewise with predicting earthquakes and tornadoes, we know these phenomena's exist even though we can't predict when they will happen and how they will behave.  If we don't have this capability today, the industry will get there, or at least get closer.  The data used today is far more sophisticated than it was 50 years ago, wouldn't you agree?  We're not standing still.  The analysis that will be used in 10 years will be far more sophisticated yet.  Maybe people like JH will be right in the middle of this advancement of the game.

I actually agree with this, jac -- to the degree that, for example, we teach hitters to have a productive at-bat ... and whatever happens happens.

But on the "other variables" front -- yes, mine were silly examples, but isn't it true that there are an infinite number of not-so-silly variables that a leader should consider when he's aware of them?

Yeah, because soon, no one is going to care about obp/slg, it'll all be velo off the bat and vectors, which may pick up on a bunch of this stuff.

For a given variable, there's a probability that it's actually going to have an effect. For instance, Showalter must implicitly account for the fact that Mayne will spontaneously combust during his PA.  The probability of that happening is so low, that its effect on the outcome of the game is negligible. This is going to be the case for almost every one of these infinite variables, so what a manager really has to do is account for the obvious ones, the ones whose chance of having an effect are meaningful.

Whether the hitter is on a streak or not isn't one of them, that is, it's not meaningful information. That a lot of managers/baseball people still treat them as if they were important is, essentially, a failure. One that other teams can exploit to their own benefit.

To be clear I very much believe in playing the percentages.  I used the bases loaded walk as an example.  Also mentioned I wouldn't have done it, but can understand the reasoning Buck Showalter did do it.

Forever in a day, I will insist that gut feeling is an important part of baseball.  What do you do when the percentages are 50-50?  Also there is the human element to baseball, these are not playing cards out there.  

I believe there is a right time to pitch around a hitter and a right time to intentionally walk a hitter.  There is a right time to bunt and a right time to do nothing but throw strikes.  Nearly every situation in baseball is different.  It's not just the outs, inning, score and numbers compiled by the players involved.  

The percentages are important for player decisions as well as coach or manager. 0 or 1 out, you are the runner at 3B with a runner at 1B. Ground ball is hit back to the pitcher. What do you do?  Should be an easy answer for a math expert.

Anyway, once again, I am a big believer in knowing the percentages and all the new metrics being used today.  I think they are very important!  However, the game is played by human beings.  Humans change from one month to the next.  One year to the next.  The numbers can be used to predict the future, but for the most part they tell us what has happened in the past.  Though other numbers give us true percentages that can be used in decision making.  When you combine the percentages with better than average gut feeling decisions/instincts, you have a big advantage.

BTW, I'm not saying anyone in this discussion is wrong about anything.  Just that I don't agree 100% with everything mentioned.  I understand JH when he says he didn't care who is on deck.  Makes sense for a pitcher to totally concentrate on the hitter.  I'm just saying part of the game is knowing who is on deck.  I might even set my defense based on the on deck hitter.  Might go into no double defense, if the on deck hitter lacks power.

Anyway, other than getting off topic and a few harmless insults. This has been a very interesting thread.  I do respect the intelligence you guys have.

jac -- if the only variables a manager must consider are the "obvious one," wouldn't it make more sense to just let the team be run by a computer?

Seriously. When it comes to probability, they're infallible.

In a way, that's what you seem to be arguing for.

I probably should have been more clear.  They really only need to consider the variables where the expected impact is non-trivial.

There's a small but non-zero chance Bonds gets tossed for arguing balls and strikes if you pitch to him. The chance is so close to zero, though, that the expected impact on the decision to pitch to him is effectively zero. There's also a small but non-zero chance he hits a walk off HR.  That will occur often enough, and have a big enough impact on the game outcome, to be part of the active decision making process.

There's a very real chance you'll be killed driving to work every day.  Do you consider and plan for the possibility every morning before you leave? No, because the expected value (no matter how catastrophic on the rare occasion that it occurs), isn't sufficiently large to warrant constantly worrying about it.

I guess I'm still unclear, jac. Sorry.

What would be an example of a variable that a manager should consider that a computer couldn't? Doesn't have to be about Bonds ... I'm just trying to understand what value you place on a human making decisions.

When the predictive odds are close to 50/50, a manager's experience (aka: gut) is likely a good method of decision making. I'll let jacjacatk cite some specific examples of this. 

That does make sense, JH -- and I'm interested in examples where the predictive odds are "close to 50-50." But of course that begs the question: How close ... where to draw the line ... and why?

Along those lines, what about research that demonstrates this:

"Snap decisions often yield better results than careful analysis because gut feelings are actually the result of unconscious mental processes—processes that apply rules of thumb that we've derived from our environment and prior experiences. The value of these rules lies precisely in their difference from rational analysis—they take into account only the most useful bits of information rather than attempting to evaluate all possible factors."

I'm on my phone - can answer more in depth in a bit. Some examples of "50/50" odds would be pertaining to pitch selection and, in some instances, choice between substitutes (I.e. pitching changes or pinch hitters). I will find some specific examples when I get to a computer.

The quote from Gigerenzer only applies if the mental process is derived from prior experiences that are rational. As indicated previously in the thread, many prior experiences within the game of baseball were largely irrational and therefore are not good bases of knowledge. For example, a manager may remember that Barry Bonds was intentionally walked with the bases loaded and the Diamondbacks won that game in 1998. Rationalizing that thought threw the selected memory would not be intelligent because evidence is overwhelmingly in the opposite direction. I am familiar with the quote from a sports psychology course I took in college and feel as though it applies in situations when the statistical predictive value is closer to even. Experience and prior knowledge is important in any managerial position, but discounting information that very obviously points in one direction or another would not be good decision making.