Our Old Friend BH Kim

I will say that you’re discussing win probability, but win probability is extremely problematic when applied to an actual specific game. Who’s available in the bullpen? Who pitched last night? Who’s on the mound? Aren’t those numbers even the same when you assess a very high run producing team and an offensively toothless team? Because you can’t be generic there.

I hear this kind of objection all the time, but it seems badly misguided to me. It’s like saying that batting averages are useless because they are so extremely problematic when applied to an actual specific game — is it day or night, who is pitching, lefty or righty, runners on base?, and so on.
So if the point is that sometimes it would be helpful to know more: Yes! Of course! Call Tom Tango! But if the point is that the more generic info is useless, that’s just silly.

I’ll bet if you examined that team, what happens in an early inning matters a hell of a lot, because the number of times they’ve won when a team has scored three in the first two innings may be low, because they won’t beat that over nine.

That may be, but the leverage will still be very low in the second inning. I mean, you can get it a bit higher if the bases are loaded with two outs in a tie game.
But what is your point here? That you won’t pull your starter in the first inning even if it’s high leverage? Fine, there’s no problem with that. Nobody in history has ever claimed that you should do it.

With the situations where you ask if I’m giving you homework, I’m not giving *you* homework, just someone homework. I’m suggesting something I’d like to see run.

What else do you want me to do? Your stuff has already been run because you’re just in the stream of the new orthodoxy. I’m suggesting new ways of looking at things, which means that these things are much less likely to have been run already, and are certainly harder to find. What you put out there will have been said many, many times by people repeating the same things because they all sign on to the new orthodoxy. I’m a party of one. I have to suggest homework, or something I’d like to run if I had the time/resources.

If something is really interesting enough, I know one half of the freakonomics team – there is something I’m going to suggest they look into, but it’s obviously not as particularized as the things I’m asking about here.

The best I can do is suggest things that may not have been captured by an existing new metric/probability calculation. Which isn’t unreasonable. You have thousands of people who want to see things exactly your way, so that math’s been run. It doesn’t mean it’s the only math worth running. It’s just harder and less lucrative to be a skeptic. Show me? PAY ME. If someone were to pay me for it, I’d be delighted to do some of the statistical inquiries I’m suggesting may be worthwhile.

At this point, also, I’m not clear on which situations you tend to think are most often improperly managed, and I probably should have asked you to define those at the start, but now, like I said, I kind of need to “to be continued” this for the night.

Although you may have explained that one in your last post – is the boundary you want to see reached that teams start using their closers much more liberally in the 6th? Is that what any analysis should be cabined to?

I don’t think what you explain in [34] is all that different from what I describe in [32].

I’d really prefer to defer to a later time to discuss why that may or may not be true, but I don’t want to go through each step of that right now.

I will say that you’re discussing win probability, but win probability is extremely problematic when applied to an actual specific game. Who’s available in the bullpen? Who pitched last night? Who’s on the mound? Aren’t those numbers even the same when you assess a very high run producing team and an offensively toothless team? Because you can’t be generic there.

I don’t want to identify a certain team we may be playing in the near future for fear of bad karma, but let me tell you, what that batter does in that .5 will always be .5 situation you’ve described sure as shit matters if you’re a team that scores 1 or 2 runs in far too many games. And yet that low scoring team would be treated just the same. I’ll bet if you examined that team, what happens in an early inning matters a hell of a lot, because the number of times they’ve won when a team has scored three in the first two innings may be low, because they won’t beat that over nine.

And it’s fine if the 3rd is a red herring, so switch all those references to the 5th.

But back to win probabilities, I find those assessments all kinds of suspect when applied to singular games. I believe they would treat a team whose starter has thrown 100+ pitches by the 5th the same way they treat a team whose starter has thrown 60 by the fifth. Which are radically different situations.

I’d like to see what the probabilities are of winning if your starter has thrown 80 pitches after four. That’s the kind of number that would interest me, because I wonder if it’s as big a problem for a team as I suspect it is. But maybe it is, and maybe it isn’t. Maybe you could evaluate it by adjusting by a metric that does a broad evaluation of a team’s bullpen to see if it’s a bigger deal for some team than others. These are the kind of things I’d like to see studied.

If you regularly consult those tables, bully for you. Please point me to them, I’d love to check them out, and also examine their methodology.

Well, I sometimes consult them. I am not a manager, so I don’t have any particular cause to consult them constantly. Here’s a nice start:
http://www.insidethebook.com/li.shtml
I recommend this book highly.
If you want to find other leverage tables (there are other methods), here is the internets for you to search with.

So you would like to know how teams do in one-run games when their closer is not available? I don’t know the answer, but that is very easy to calculate, right? So if you are not, in fact, giving me a homework assignment, then I don’t understand your point.

And you are telling me that teams that lose games in the ninth inning tend to do worse in their next game? I’ve never seen evidence about that, but I am very skeptical. Show me.

Using Papelbon:
There are rarely high leverage situations in the fifth inning, and in the fifth inning a team usually has a good pitcher in the game, namely, their starter. So I do feel that I more or less answered that question. There are few but some high leverage situations in the sixth, and the starter may be spent. So I say, obviously you use your best pitcher in that situation, and I frankly think it’s just dumb not to. It’s hidebound, conservative, foolish.

This also all started with me ranting about how much I hate using Proctor where he was used, because of how it turns all resources previously deployed into worthless shit. So that’s a bit of a strange way to conclude your post.

Huh? Why am I constrained by what you said two hours ago? The example about Proctor makes my point nicely. I don’t think I’m debarred from using it because you were ranting about him earlier this evening!

I left the paaaaaaaaark
Off Frank Francisco….

If you regularly consult those tables, bully for you. Please point me to them, I’d love to check them out, and also examine their methodology.

But I explained that even managers who consult them may be consulting flawed tables, for all the reasons I point out above, in which case, they are hardly employing empirical fact.

Similarly, I’m not understanding what you’re missing about the question of teams that don’t have the closer available to close, but I’ll try to do better this time: how do teams do, in, say, one run games in the 9th, when the closer is unavailable? Now you might not have the deepest pool of statistics available because, as we’re discussing, teams *do* tend to wait to deploy their closers until the 9th. But you might be able to find some data for your analysis by examining games played after the closer’s been used three days in a row, or thrown 50 pitches over two days, or 60 over three days, etc.

Because situations exist like the one recently where Mo was not available because of use. You could pool together all those situations and see where they come out.

And run it up two runs in the ninth, no closer to use. And three runs. The three analyses may come out quite differently.

Again – not a close reading on the winning and losing and the next game. I said teams that lose a game *in the ninth* – NOT – teams that win versus teams that lose. I suspect that although you may have seen the latter, you have not seen the former.

I’ll get to the rest of your post in a bit, although to be honest, I’m running out of time here (have some stuff that I actually do need to do), and I will try to look up that season.

I’m with you, btw, on using Papelbon in that spot. One, because you shouldn’t be *that* inflexible (it’s the freaking eighth), and two, because Bard had lost it and needed to be rescued. Here’s reality again. Bard, having shit 7/8ths of the bed does not equal Bard coming in to start an inning. The Bard who’s done all that bed-shitting isn’t Daniel Bard. He’s the “oh my God, I’ve shat an awful lot of this bed so far, haven’t I” version of Daniel Bard. So pull him the heck out of the game.

The question is – do you use Papelbon in the 5th? How regularly are you willing to take that leap? How about the 6th? This is where managers leave the shore behind and take a full plunge into a new way of managing. Which may be the correct move, and nothing I said above states that it’s not. I suspect it may not be, for reasons that I’ve stated above. But you can’t half-ass it by acting like it’s the same thing to be more than on team “put Papelbon on in that spot in the 8th” and on team “put him in whatever these metrics declare to be the highest leverage situation, even if it’s the 4th!”

This also all started with me ranting about how much I hate using Proctor where he was used, because of how it turns all resources previously deployed into worthless shit. So that’s a bit of a strange way to conclude your post.

The answer cannot be that the Yanks need an actual long man (guy to go 3+ innings). That’s not Proctor at all.

The answer cannot be that the Yanks need a “long man” (ie, reliever who can pitch more than 1 but not much more than 2 innings), because that’s not really Proctor. They have Kontos for that.

That leaves the answer as the team needs another short reliever. Why waste any time on Proctor when you have Whelan in the minors? (BTW, he’s one of three guys on the 40-man roster NOT in the big leaguse.) he’s had a lot more success more recently than Proctor.

That Proctor is on the Yanks and pitching any kind of innings is pointless and stupid, unless the Yanks have decided that home field advantage in the ALDS and ALCS is without value.

Whoops, too late, they scored, ruining the example (but I’m not complaining!).

Now Francona turns to Albers, which is reasonable, because he’s not good. 🙂

[32] No, that’s not what leverage is. Leverage is in effect a matter of potential change in win probability. In some situations, like very early in the game, you’ll find that a team’s win probability is about .5 and will continue to be close to .5 no matter what the current batter does. But then late in a close game with a couple of men on base you’ll find that a double would raise the team’s winning chances from .2 to .8 and a double play would drop it to .1.

I don’t understand what you’re saying in the next paragraph. It sounds like you’re giving me a lot of homework, but I assume that’s not what you mean. For instance,

make sure you identify – by doing the math – how well teams do that do not have their closers available for the 9th inning!

There’s no general answer to that question, obviously. If they left Dan Bard in and then relieved him with Matt Albers instead of using their really good pitcher, then they lose the game whether their ace closer is still available or not. If they let Scott Proctor pitch because the game is tied and they only use Mariano when they have a lead, then it really doesn’t matter whether Mo is available or not, because they just lost the game in the bottom of the tenth. Because instead of using their best pitcher in the most important situation, they insist on using him in the special Closer Situation.
So, what is your statistical question about how well teams do when they don’t have their closer for the ninth? We need a more specific question than what you’ve asked.

Another example:

Perhaps momentum is only as good as the next day’s starting pitcher, or perhaps there is an ascribable value that is being left out.

What’s with the ‘perhaps’? This isn’t some unknowable abstractum. You can just check to see whether teams do better when they’ve just won, or not. And the answer is, they do not.

Red Sox interchangeable closer:
I doubt it. What season are you talking about? I am very skeptical about this. Show me.

And as for evidence about roles: if it’s true, there should be evidence for it. Relief pitchers didn’t have special roles in the 1970s or even the 1980s. So show me how badly relief performance suffered in those decades compared to later decades. Why is this impossible?

Finally, this thing that you said makes my point best:

There may be a particular leverage to a particular situation, but you’re still going to have to get at least 27 outs.

Right. And the outs you have to get in the low-leverage situations should be assigned to your worse pitchers, and the ones you have to get in the high-leverage situations should be assigned to your better pitchers. If I manage the Red Sox, you can bet your ass that Papelbon will be the one going after Bautista with two out, bases loaded and a one run lead in the eighth. Even though it’s the eighth. Someone who tells me, “Oh, but we might need him more in the ninth” is either being disingenuous or doesn’t know what he’s talking about. If I’m managing the Yankees, I’m not using Scott Proctor in a tie game in the ninth when I have the best pitcher in the history of baseball ready to go, just because there is no Save available. That’s just nuts.

But if you’re running those numbers on a massive mass of statistics compiled on people using the *old* methods, that’s a flaw.

Because those numbers might change, drastically, if the 9th inning guys were available much more infrequently because they were already used in the fifth. If you’re evaluating how likely it is that you lose a game in the ninth with an (x)run lead, and you’ve done that evaluation using games where 9th inning guys were deployed in those games, your stats, or, at the very least, your conclusions, are horrifically fucked.