Baseball for the Thinking Fan

You are here > Home > Baseball Newsstand > Baseball Primer Newsblog > Discussion
Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand

Saturday, August 16, 2014

Retrosheet: Soper-Understanding The Value of the Next Run. Why Sabermetricians Are Wrong and Traditional Baseball is Right

By focusing solely on runs scored and not on winning baseball games sabermetricians miss the impact the context of scoring has on the value of a run.

“A hitter’s job is to create runs for his team” The Bill James Historical Baseball Abstract
Bill James is wrong. He is wrong on a fundamental concept of baseball.

For example, if the lead-off hitter in an inning reaches 1st the team is expected to score .941 runs that inning. If he steals 2nd the expected run total increases to 1.17. If he is thrown out, however, the expected runs drops to .291 (1out, bases empty). By stealing 2nd the runner is risking .65 runs to gain .229 runs. You can calculate the threshold of success by dividing the runs risked by the total of the runs risked and the runs possibly gained. In this case .65/(.65+.229) = .739, in other words the correct strategy is to steal 2nd if you will be successful more than 73.9% of the time. In this case you would gain .229 runs * 73.9% and lose .65 runs * 26.1% and would break even.
But if you only need 1 run, say it’s tied in the bottom of the 9th, you just want to focus on the first run.
We can use the right side of TABLE 3 to calculate threshold of a runner stealing 2nd in the bottom of the 9th after a lead-off walk. From 1st his team will score 44.1% of the time. From 2nd his team will score 63.7% of the time. If he is thrown out his team’s scoring chance drops to 17.2%. So by attempting to steal he is risking .269 to gain .196. We compute the threshold the same way: .269/(.269+.196) = 57.7%. “Holy Tony La Russa, Batman!” We knew that maximizing runs didn’t make sense in the bottom of the 9th, but still that is a big drop.

Correct me if I’m wrong but it doesn’t appear to me that Soper is accounting for the other team’s run scoring ability or that late innings have a different run environment than average.  Plus he’s using a strawman argument when declaring that sabermetrics would tell you to do one thing in a very specific situation based on a very general strategy.

McCoy Posted: August 16, 2014 at 08:59 AM | 61 comment(s) Login to Bookmark
Tags: sabermetrics

Go to end of page

Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.

1.  Posted: August 16, 2014 at 07:33 PM (#4772541)
I don't think many people would argue that a 3 run double that made the score 8-2 with 1 out in the bottom of 7th inning is as big a hit as a double that drives home a runner from first to break a tie with 2 outs in the top of the 8th. It turns out the 3 run double would increase your chances of winning by 2%, the tie-breaking double increases the hitting team's chances by a whopping 32%

Two problems here, at first glance. One is that players and teams that hit lots of three-run doubles and often win by six runs are better than those that eke out one-run wins. If the idea is to evaluate an offense, individual or team, you want to know more than the W column. If the idea is to know who won the game, why all the math.

Another is that unless it's the seventh game of the World Series, it does matter how much you win by. In the former situation, you can rest your top set-up man and closer, and caddy for your aching cleanup hitter and other such stuff, and the game has less chance of going to 14 innings and exhausting your team. Maybe minor stuff, who knows, but real.

I will leave it to our mathematicians to say if the rest of TFRP tells me something that WPA doesn't.
2. Walt Davis Posted: August 16, 2014 at 08:11 PM (#4772561)
For example, if the lead-off hitter in an inning reaches 1st the team is expected to score .941 runs that inning. If he steals 2nd the expected run total increases to 1.17. If he is thrown out, however, the expected runs drops to .291 (1out, bases empty). By stealing 2nd the runner is risking .65 runs to gain .229 runs. You can calculate the threshold of success by dividing the runs risked by the total of the runs risked and the runs possibly gained. In this case .65/(.65+.229) = .739, in other words the correct strategy is to steal 2nd if you will be successful more than 73.9% of the time. In this case you would gain .229 runs * 73.9% and lose .65 runs * 26.1% and would break even.
But if you only need 1 run, say it’s tied in the bottom of the 9th, you just want to focus on the first run.
We can use the right side of TABLE 3 to calculate threshold of a runner stealing 2nd in the bottom of the 9th after a lead-off walk. From 1st his team will score 44.1% of the time. From 2nd his team will score 63.7% of the time. If he is thrown out his team’s scoring chance drops to 17.2%. So by attempting to steal he is risking .269 to gain .196. We compute the threshold the same way: .269/(.269+.196) = 57.7%. “Holy Tony La Russa, Batman!” We knew that maximizing runs didn’t make sense in the bottom of the 9th, but still that is a big drop.

This is the oldest news in the world.

And if this was basketball where you can put the ball in Jordan's hands at the end of the game, this would matter a lot in determining an individual player's value. But given a manager generally doesn't even get to decide that Rickey Henderson comes to the plate to start the 9th, these opportunities are randomly distributed.

There is also the effect size issue. Yes, if you have a better chance than the break-even point, you "should" go for it. But say that even in his 9th inning situation, the guy has a 75% chance of making it. .75*.637 = .478. So, at a true 75% chance of a successful steal, a steal attempt moves the p(scoring) from 44.1 to 47.7 -- not exactly a massive effect.

Now, now many times a year are you tied in the bottom of the 9th and your lead-off hitter walks or singles? How often is that a 75% base-stealer (or you have one available on the bench)? I'd be surprised if that was more than once or twice a season. Heck, make it 10 times a season, try to steal all 10 times and you add .3 wins per year.

So even if we buy the scenario he's painting, this mythical Rickey Henderson is adding .3 wins per year that he's not getting credit for. Wow, better throw out all those formulas Tango's got cuz that's just a massive, massive bias that makes the whole enterprise useless.

In the excerpted example, the author is making the same "mistake" that the simpler models do. Your p(scoring) is also effected by who's coming up next, who's on the mound, who's warming up in the pen, the quality of the catcher, who's umpiring today, what are the weather conditions, how good is the pickoff move, how many times does he throw over, where is the 3B positioned and whatever else I'm not thinking of.

If the point is that big actuarial tables of average outcomes are not a very useful guide for making tactical decisions in context-specific situations then I agree. But that's exactly what the author is doing in the excerpt, his actuarial table just has more cells in it.
3. Walt Davis Posted: August 16, 2014 at 08:21 PM (#4772565)
I will leave it to our mathematicians to say if the rest of TFRP tells me something that WPA doesn't.

What, is there a new formula in the article? Ugh. Reinventing wheels.

Even if we accept WPA as a component of value, the simple fact of the matter is that for batters, these opportunities are pretty much randomly spread around. Yes, due to randomness (and some lineup position effect) some batters will end up with more such opportunities than others. These batters are arguably slightly under-/over-valued by the main formula. I can't imagine that, for batters, we could be talking about more than 1-2 wins over the course of a career -- maybe 5-6 for some extreme cases.

Managers do have control over which pitchers they use when so for pitchers the story could be different. But, for pitchers, WAR (at least bWAR) takes WPA into account. (Whether it should or not is another question.) For starters, WPA doesn't really matter but of course certain relievers are substantially more likely to be pitching high leverage innings than others. So the fancy models are already doing what he suggests.
4. Ziggy's screen name Posted: August 16, 2014 at 09:26 PM (#4772593)
Can we just burn and bury WPA already? The run you score in the first helps your team win just as much as the tie-breaker in the bottom of the 9th, you just didn't know it yet.
5.  Posted: August 16, 2014 at 09:40 PM (#4772605)
Can we just burn and bury WPA already? The run you score in the first helps your team win just as much as the tie-breaker in the bottom of the 9th, you just didn't know it yet.

I'm one of the most anti-WPA guys out there, but I would have to oppose this. It has value, it has story telling uses etc... it just doesn't mean as much as it's proponents think it does. On top of that, eventually people are going to be able to use it as a stat for evaluating managers decision, stolen base analysis or other evaluation stats.

As far as the example of first inning vs 9th inning, wpa/L1 does a good job of making up for that massive flaw in WPA.
6. Ziggy's screen name Posted: August 16, 2014 at 09:58 PM (#4772617)
Once you divide WPA by LI don't you just get the linear-weights value of the event? (And even if you don't, isn't that what you actually want?)

We don't need stats to tell a story, especially not when they're going to get mis-used. (As they do - and especially this one does - a lot.)

As for using it to help evaluate managers: I don't know, let me sleep on it.
7. Cargo Cultist Posted: August 16, 2014 at 10:01 PM (#4772619)
Does this guy not understand that batting opportunities are mainly random?

Or am I missing something?
8. Walt Davis Posted: August 16, 2014 at 10:15 PM (#4772631)
oops, I forgot to add in the .17*.25 bit -- the probability of scoring the run even if he's thrown out -- which adds another 4% to the steal attempt ledger.
9. Jeltzandini Posted: August 16, 2014 at 11:43 PM (#4772659)
Once you divide WPA by LI don't you just get the linear-weights value of the event?

Not in game winning situations. If a sacrifice fly is as good as a home run, WPA/LI gives them the same value even though they have different linear weights.
10. Yellow Tango Posted: August 17, 2014 at 07:55 AM (#4772716)
I'm not defending WPA, but can someone explain why it's so terrible? It seems to get bashed a lot here, but I think I missed the point where the arguments were made against it.
11. Misirlou doesn't live in the restaurant Posted: August 17, 2014 at 08:59 AM (#4772722)
I'm not defending WPA, but can someone explain why it's so terrible? It seems to get bashed a lot here, but I think I missed the point where the arguments were made against it.

One major flaw in straight WPA (and I'm sure there are others), is that it's possible for a player would have had a better WPA if he had a worse game. Take the game of June 27, 1964 for example. The Orioles beat the Senators 3-1. Boog Powell hit 3 solo HR for all the O's runs. He had a WPA for the game of .283. His HR in the top of the first was worth .11. His HR in the 9th to make the score 3-1 was worth .08. Had he struck out in the first, and the rest of the game went the same, his HR in the 9th to give them the lead would have been worth ~.4, and thus his WPA for the game higher than the 3 HR game. A guy who hits a HR in the top of the first of an eventual 1-0 game has a lower WPA than a guy who hits one in the second, who has a lower one than a guy who hits one in the 3rd...
12. Lassus Posted: August 17, 2014 at 09:36 AM (#4772730)
The first sentence of the excerpt was too annoying for me to read anything else.
13.  Posted: August 17, 2014 at 09:47 AM (#4772734)
This may be obvious, but WPA is not great for evaluating players, because of the way it's dependent on context (as noted variously above). Because of the random distribution, good hitters tend to have good WPA (Mike Trout is leading the AL this year, that's not very interesting).

But quirks in the distribution of opportunities can lead to discrepancies, which are equally uninteresting for evaluative purposes. My hero Ben Zobrist is having a better year at the plate by independent measures than Jose Reyes of Toronto. But Reyes is doing great on WPA and Zobrist lousy. There is zero reason to think that that situation will persist next year, because it's out of whack with the past. One really doubts that Reyes has become Clutch God and Zobrist a big choker as of 2014. Now, it's fun to know and may cause one to revisit some exciting Blue Jays box scores, but its value for analysis seems minor.
14. Yellow Tango Posted: August 17, 2014 at 11:12 AM (#4772764)
Thanks, guys.

To be honest, it doesn't seem worse than something like RBIs, though it doesn't seem better. It still seems like a good (highly) context-dependent stat, though I might be missing something important. It certainly seems lousy as a prediction.
15. depletion Posted: August 17, 2014 at 11:35 AM (#4772780)
I was informed (perhaps by Mike Emeigh) that the distribution of runs in a 9-inning game does not follow a Poisson distribution. This implies that the run totals by inning "communicate" with each other. My gut feel is that one reason for this is that managers "give up" in blowouts (that is they pull the star players and give AB's and IP's to weaker players).

It turns out that the distribution of time between flashes in a group of fireflies also does not follow a Poisson distribution. This implies that the fireflies communicate to each other with their flashes.
16.  Posted: August 17, 2014 at 12:24 PM (#4772808)
To be honest, it doesn't seem worse than something like RBIs, though it doesn't seem better. It still seems like a good (highly) context-dependent stat, though I might be missing something important. It certainly seems lousy as a prediction.

That is the problem with it, it was perceived to be a stat oriented way to replace rbi and ended up being just as flawed as RBI. I was at sabr St Louis and a guy was touting it as the only stat that matters any more. The way it rewards based upon timing is a great way to use for a narrative feeling, a solo homerun in the bottom of the ninth of a 1-0 game feels more 'clucthy' than a homerun in a 1-0 game in the third inning, but as far as helping to provide the win, they are exactly the same thing, wpa doesn't treat them that way.

I'm fine with a stat saying a two run homerun is better than a solo homerun, even though the batter wasn't really responsible for the guy being on base, but I don't accept that a 2 run homerun in a 5-3 game is more valuable simply because of the out situation and inning it was hit in.
17. Joyful Calculus Instructor Posted: August 17, 2014 at 12:42 PM (#4772818)
the bottom of the ninth of a 1-0 game feels more 'clucthy' than a homerun in a 1-0 game in the third inning, but as far as helping to provide the win, they are exactly the same thing

I disagree. when a player hits a solo home run in the 3rd inning of a 0-0 game, the result of the last six innings is undetermined, while hitting one in the 9th, the first 8 innings have already been determined. There is no way to know that the HR in the 3rd will end up being the game winner until it's over and unless you believe in a deterministic universe, it wasn't.
18. Jeltzandini Posted: August 17, 2014 at 12:52 PM (#4772824)
It's better than RBI, which is both noisy and biased. WPA is just noisy.
19.  Posted: August 17, 2014 at 12:55 PM (#4772827)
It's odd that ballplayer clichés meet up with sabermetrics on this issue. "Yes I got that hit in the ninth, but it's a team game and if Manny hadn't hit that home run in the third and Pedro hadn't kept the game close with those relief innings, I wouldn't have been in that situation, it's not about one guy on this club."
20.  Posted: August 17, 2014 at 12:57 PM (#4772830)
I disagree. when a player hits a solo home run in the 3rd inning of a 0-0 game, the result of the last six innings is undetermined, while hitting one in the 9th, the first 8 innings have already been determined. There is no way to know that the HR in the 3rd will end up being the game winner until it's over and unless you believe in a deterministic universe, it wasn't.

I disagree with your disagree. Value is value regardless of when it happens. Nobody can say how a homerun in the first affected the rest of the decisions from both sides for the rest game, nor having a shutout go to the ninth. The only thing you can measure is how much the player contributed to the victory. And a solo homerun in a 1-0 game is the same value towards that victory no matter what inning it is hit in. I can make reasonable arguments that a solo homerun in the 3rd inning is MUCH more valuable than one hit in the 9th inning of a 1-0 game, and I can make arguments of the other side of the equation, but those aren't really measurable fact based arguments.

Ultimately it's what was the final score and measuring the impact of each player in that result.
21. PreservedFish Posted: August 17, 2014 at 01:03 PM (#4772833)
I can make reasonable arguments that a solo homerun in the 3rd inning is MUCH more valuable than one hit in the 9th inning of a 1-0 game

I'd like to see those arguments.
22.  Posted: August 17, 2014 at 01:08 PM (#4772835)
I'd like to see those arguments.

If a pitcher is losing 1-0 he is more likely to be pulled for a relief pitcher, so you are increasing the chances that the team is going to bring in a lesser pitcher who may not be as sharp as the starting pitcher who has only made one mistake so far. If the shutout is still going on, then the other team will not be as willing to remove the pitcher.

This assumes a couple of things. One the starting pitcher is better than the relief pitching options and two that relief pitchers are erratic and there is such a thing as being on or off each day.
23. PreservedFish Posted: August 17, 2014 at 01:15 PM (#4772842)
Not convinced.
24.  Posted: August 17, 2014 at 01:28 PM (#4772849)
Not convinced.

I'm from the point of view, that if a guy is "on" it's best to get him out of the game, as relief pitchers are erratic. A relief pitcher might be better than a starting pitcher on average, but when you have a known commodity(starting pitcher "on") versus an unknown pitcher who you don't know is going to be on or not and by the time you find out, the game might be out of hand.

I'm not convinced that a homerun in the ninth inning of a 1-0 game is more valuable than one in the third inning. I'm not sure how anyone could even think such a thing. But yet WPA does exactly that.
25. PreservedFish Posted: August 17, 2014 at 01:30 PM (#4772851)
I'm not convinced that a homerun in the ninth inning of a 1-0 game is more valuable than one in the third inning. I'm not sure how anyone could even think such a thing.

This is an exaggeration, right? It's patently obvious why the late homerun might be considered more valuable.
26.  Posted: August 17, 2014 at 01:40 PM (#4772858)
This is an exaggeration, right? It's patently obvious why the late homerun might be considered more valuable.

It might be a slight exaggeration, but it's not obvious why the late homerun might be more valuable. The late homerun feels more important, but that is different than value.

I don't think it's accurate to assign value until all the known factors are known...meaning you don't assign value until after the game ends, tracking it as it happens live is not really tracking value, but instead tracking emotion.
27. PreservedFish Posted: August 17, 2014 at 01:50 PM (#4772863)
You need to define value. Most people on this website are willing to accept that if you have two otherwise equivalent players, the guy that hit absurdly well in the clutch was the more "valuable" of the two, even if it doesn't demonstrate any real ability that will continue to manifest itself in the future.

There's probably no flawless method here but waiting until the game ends, for example, threatens to render valueless any effort in a losing game.

Per WPA, if I hit a huge clutch homer in the 8th, I get credit for it even if my bullpen proceeds to blow the game. In your system? Does the clutch homer have any value at all?
28.  Posted: August 17, 2014 at 01:56 PM (#4772865)

Per WPA, if I hit a huge clutch homer in the 8th, I get credit for it even if my bullpen proceeds to blow the game. In your system? Does the clutch homer have any value at all?

yes of course everything has value whether it's a losing or winning cause, it's just that it's based upon the final score of the game, and not the emotional impact that the play has on the observer.
29. PreservedFish Posted: August 17, 2014 at 02:01 PM (#4772867)
But you see? If you refuse to assign value at the moment of the event - if you only award value after the game finishes - then you are hugely devaluing the clutch homerun. It is necessarily so.

Per WPA, a leadoff single is worth whatever it's worth no matter what follows. Per your accounting, the value of a leadoff single is contingent on what the following hitters do. In fact, it's contingent on the outcome of the entire game.

(Also, I think that "value" is nebulous enough that it might allow some consideration of the emotional impact. I think WPA is fascinating and fun, even if it's useless to GMs.)
30.  Posted: August 17, 2014 at 02:05 PM (#4772869)
Per WPA, a leadoff single is worth whatever it's worth no matter what follows. Per your accounting, the value of a leadoff single is contingent on what the following hitters do. In fact, it's contingent on the outcome of the entire game.

What is the difference between that and WPA where every value of the event is based upon what preceeded it?

I just do not understand or accept a system that will say a solo homerun in a 1-0 game, value is different just because of what inning it happened. That is utterly unacceptable. It makes absolutely zero sense to me.
31.  Posted: August 17, 2014 at 02:06 PM (#4772870)
(Also, I think that "value" is nebulous enough that it might allow some consideration of the emotional impact. I think WPA is fascinating and fun, even if it's useless to GMs.)

I think it's fascinating and fun, when used properly. As basically the stat version of RBI and equally as useful as RBI.
32. PreservedFish Posted: August 17, 2014 at 02:34 PM (#4772891)
What is the difference between that and WPA where every value of the event is based upon what preceeded it?

There is a rather stark difference between events that have already occurred and that we have knowledge of, and events that have not yet occurred. I shouldn't really need to explain that.

I just do not understand or accept a system that will say a solo homerun in a 1-0 game, value is different just because of what inning it happened. That is utterly unacceptable. It makes absolutely zero sense to me.

And yet you accept a system where the value of the homerun is different based on unknowable future events? Your system is vastly, vastly worse.
33.  Posted: August 17, 2014 at 02:42 PM (#4772901)
And yet you accept a system where the value of the homerun is different based on unknowable future events? Your system is vastly, vastly worse

Actually I don't accept any system that tries to sell narrative as a stat. I'm perfectly fine with raw numbers, ops+, war etc. When you get into the narrative stats like wpa or Rbi, I just don't care and am only arguing against their usage not for a better system.

WPA = RBI. They are both equally useful and equally useless. Neither tells you one thing about the value of a player. But if anything, I would prefer to see WPA/LI instead of straight WPA.
34. Jeltzandini Posted: August 17, 2014 at 03:51 PM (#4772938)
35. Joyful Calculus Instructor Posted: August 17, 2014 at 04:11 PM (#4772952)
I think it's pretty clear that the game context in which an event occurs matters. There is a reason that managers use their best relivers when up by 1-3 runs in the 9th and their worst when up or down by 10 in the 9th. When a batter hits a solo home run in the 3rd inning, it could be the only run of the game or maybe it ends up a blowout anyway or maybe his team will lose anyway so it's a home run that could be important, while in the 9th inning of such a game it is crucial. On the flip side, that home run in the 9th of a 12-3 game is extremely unlikely to be of consequence. So, HR late and close > HR early > HR late in blowout.
36. PreservedFish Posted: August 17, 2014 at 04:20 PM (#4772963)
Actually I don't accept any system that tries to sell narrative as a stat.

Basically you reject any attempt at finding context-based "value." If I were a GM, I would too. But I'm not, I'm a baseball fan, so I enjoy looking at and debating these things.
37.  Posted: August 17, 2014 at 04:24 PM (#4772967)
I think it's pretty clear that the game context in which an event occurs matters. There is a reason that managers use their best relivers when up by 1-3 runs in the 9th and their worst when up or down by 10 in the 9th. When a batter hits a solo home run in the 3rd inning, it could be the only run of the game or maybe it ends up a blowout anyway or maybe his team will lose anyway so it's a home run that could be important, while in the 9th inning of such a game it is crucial. On the flip side, that home run in the 9th of a 12-3 game is extremely unlikely to be of consequence. So, HR late and close > HR early > HR late in blowout.

But nobody on here has been talking about blowouts vs close games, it's been about 1-0 game and when the homerun hits and a system that rewards the homerun being hit later much more significantly than it does if it's hit earlier.

Hr late in a 1-0 game is not more valuable than a homerun early in a 1-0 game, I know I'm never going to convince you guys of that, but it isn't. It feels more valuable or clutch, but it actually isn't.

38.  Posted: August 17, 2014 at 04:25 PM (#4772968)
Basically you reject any attempt at finding context-based "value." If I were a GM, I would too. But I'm not, I'm a baseball fan, so I enjoy looking at and debating these things.

I enjoy debating everything(one of my many faults) but I just don't see any difference between the situation that we have been talking about(and to make it simple. Just assume we are talking about a solo homerun in a 1-0 game as this illustrates the point from both sides the simplest)

39. PreservedFish Posted: August 17, 2014 at 04:50 PM (#4772977)
Again, you need to define "value." I can't really engage you on this issue if you don't do that.
40. Fancy Crazy Town Banana Pants Handle Posted: August 17, 2014 at 05:03 PM (#4772984)
I enjoy debating everything(one of my many faults) but I just don't see any difference between the situation that we have been talking about(and to make it simple. Just assume we are talking about a solo homerun in a 1-0 game as this illustrates the point from both sides the simplest)

I think we can agree, if I started a new baseball league, with 6 inning games instead of 9, the value of runs go up. Teams will score fewer runs per game, and you would need to generate fewer runs above a replacement player, to be worth a win above said replacement player. The value of runs relative to wins increased.

So effectively, if you hit a HR in a scoreless game in the third, it is still a 7 inning game. Once you get to a tied game in the ninth though, it is a one inning game. The value of runs is magnified accordingly.
41.  Posted: August 17, 2014 at 05:05 PM (#4772987)
Again, you need to define "value." I can't really engage you on this issue if you don't do that.

Value is how much the player contributed to the victory. Or how much a player contributes to the chance of a team victory. Wpa doesn't do that, it measures the "feelings" of how a player contributes to a victory.
42.  Posted: August 17, 2014 at 05:08 PM (#4772991)
So effectively, if you hit a HR in a scoreless game in the third, it is still a 7 inning game. Once you get to a tied game in the ninth though, it is a one inning game. The value of runs is magnified accordingly

No, the value of how much the importance of that run "feels" is magnified, but it is not one bit of a difference in absolute value.

There is no difference between a solo homerun in a 1-0 game whether it happens in the first inning or the 12th inning.
43.  Posted: August 17, 2014 at 05:13 PM (#4772997)
Heck WPA treats the solo homerun in a 1-0 game differently depending on who is the home team.

If the visitor wins the game and their guy hits a homerun with two outs in the 9th inning, it's less valuable by WPA than if the home team hits the homerun in the ninth with two outs. That is absurd.
44. PreservedFish Posted: August 17, 2014 at 05:15 PM (#4773000)
I don't see how I'm supposed to make headway here. You're just wrong. Your definition of value means that the value of the homerun I hit in the third inning is totally unknown until the game has concluded. Two hours later, if we're still playing, the value of that homerun is still in question. That is what's truly absurd. You compare WPA to RBI - and that's fair, because they are both hugely dependent on context - but your value system is miles worse than either of those stats, because it's exactly as dependent on context but instead of considering real world context, facts that the players are actually aware of, it uses the context of future events.

Don't you see where the logical conclusion is? Explain to me how a homerun I hit in a loss is anything other than valueless, according to you.

I understand your objection to WPA but you throwing the baby our with the bathwater. To an extreme degree.
45.  Posted: August 17, 2014 at 05:25 PM (#4773010)
Don't you see where the logical conclusion is? Explain to me how a homerun I hit in a loss is anything other than valueless, according to you.

No it's incremental in your chances of winning the game. That is what most stats are trying to measure how much does your performance help your teams odds of winning.

WPA in theory is a FANTASTIC concept. Not kidding, before I ever heard of WPA, and just learned about win expectancy, this is something I thought would be a great way to measure value and replace clutch and rbi's. I never considered there would be flaws in something like this, as it seemed obvious and simple enough. Once I learned that a solo homerun in a 1-0 game has a different value based upon when it happens, then yes I realized there was a problem with the concept.

As I(and countless others) have mentioned, it is a great narrative tool for gauging the 'mood' of a game, it's just not nearly as good for rating value that a player provides to his team in victory. I just cannot consider taking a stat seriously that has such an obvious flaw as how it treats the point we are talking about(1-0 game, solo homerun)
46.  Posted: August 17, 2014 at 05:27 PM (#4773012)
I don't see how I'm supposed to make headway here.

You're not. There is absolutely no way you or anyone could ever convince me that a solo homerun in a 1-0 game, value is based upon which team you played for and what inning it happened in. Heck even hitting it in the bottom of the ninth inning has different values. If you hit it in the bottom of the ninth with 0 outs, it's less valuable than if you hit in the bottom of the ninth with two outs.
47. PreservedFish Posted: August 17, 2014 at 06:12 PM (#4773039)
No it's incremental in your chances of winning the game. That is what most stats are trying to measure how much does your performance help your teams odds of winning.

But if you are looking backward, then you know that it increased your chance of winning the game 0%.
48. Captain Supporter Posted: August 17, 2014 at 06:37 PM (#4773050)
Hr late in a 1-0 game is not more valuable than a homerun early in a 1-0 game, I know I'm never going to convince you guys of that, but it isn't. It feels more valuable or clutch, but it actually isn't.

Yes, it is.
49. Walt Davis Posted: August 17, 2014 at 07:19 PM (#4773085)
I was informed (perhaps by Mike Emeigh) that the distribution of runs in a 9-inning game does not follow a Poisson distribution. This implies that the run totals by inning "communicate" with each other.

Not necessarily true, it (generally) just means that the distribution is over-dispersed (negative binomial) or under-dispersed ("zero-inflated poisson"). The poisson requires the mean and the variance to be the same ... there's nothing particularly magical about the mean and the variance not being the same and you don't require non-randomness to get there.

And of course the poisson (or any similar distribution) can't possibly hold precisely because the baseline mean/probability changes from game-to-game and inning-to-inning based on the quality of the (tiring) pitcher and who's coming to the plate in any given inning. It can't possibly be anything but a good approximation. Any small lack of independence is likely due to that tiring pitcher -- assuming his team is silly enough to keep him in after having thrown 30 pitches the previous inning.

As to WPA, I give you this game: http://www.baseball-reference.com/boxes/CIN/CIN197106230.shtml

Wise threw a no-hitter and hit 2 HR driving in his teams 2nd, 3rd and 4th runs in a 4-0 victory. Total WPA for the day is just .548

In the top of the 2nd with nobody out and a runner on 1st, Willie Montanez hit a double that did not score the guy. That was worth .12 WPA ... Wise's first 3 perfect innings were worth .15 WPA.

By the way, the groundout that actually scored that first run was worth -.02 WPA.

After Montanez's double in the top 2nd, WPA says the p(win) was already 64%. Seems a bit high no, it's still a scoreless game. Anyway, 3 innings later just before Wise's first HR, the p has only risen to 68%. Wise's first HR moves it to 86% and then nothing any individual does makes a lot of difference.

Compare this to the Yanks-Rays game yesterday. Gardner singles and advances on an error to lead off the 9th and gets .17 WPA. Jeter singles and gets .2 WPA.

On May 3, 2013 ... Down 1-0 (to Kershaw but WPA doesn't know that) in the bottom of the 6th, Buster Posey hits a double that drove in the tying run. In the bottom of the 9th, he hits a walk-off HR. Total WPA for the day of .58, more than Wise. (How'd I find that game? Pure luck. It oddly was the first walk-off HR story brought up by Google.)

There's no way that Posey did anywhere near as much to win his game as Wise did to win his. Posey's WPA is an accident of timing, Wise's performance was a great performance.
50.  Posted: August 17, 2014 at 07:36 PM (#4773097)

Yes, it is.

No it isn't. It's one run either way in a one run game.
51. Jeltzandini Posted: August 17, 2014 at 07:56 PM (#4773108)
After Montanez's double in the top 2nd, WPA says the p(win) was already 64%. Seems a bit high no, it's still a scoreless game.

2nd and 3rd, no out? A big inning is quite likely. 64% doesn't seem high at all.

WPA has lots of problems as a value measure, obviously, but I don't think the expectation table is one.
52. Ziggy's screen name Posted: August 17, 2014 at 08:44 PM (#4773129)
WPA is useless, for the reasons CFB and I (and probably others) discussed up-thread. But CFB is steering us on the wrong path by measuring value by game result. That has the absurd consequence that a game-winning home run (in whatever inning, if we're tossing out WPA, as I sincerely hope we are, it doesn't matter the inning in which the HR was hit) is more valuable than a no-hit pitching performance that results in a loss because of a terrible error by a fielder. As far as I can tell, the only measures of value that don't have absurd consequences like this are those that are entirely divorced from game context. If forced to back one, I'd stump for linear-weights values as a measure of value simpliciter.

But all that is really a distraction from the main point, which is that WPA is useless, except for story telling (for which we don't need stats).

In addition, there's a confusion up-thread that we should sort out, between epistemic issues and metaphysical ones. There's a difference between not knowing what will happen, and there being no fact of the matter about what will happen. CFB, with his context-based account of value, is committed to saying that we don't know the value of an event until after the game is over. But that's not the same thing as saying that it doesn't have a value - it's just that we don't yet know what it is. This fact on it's own really shouldn't be objectionable (though, as I've said, there are other reasons to think that the view is objectionable).
53. DKDC Posted: August 17, 2014 at 09:17 PM (#4773150)
I like WPA.

There will be a high correlation between a game's WPA leader and a poll of fans who watched the game as to who was the "player of the game".

Like most baseball fans, I appreciate the ebbs and flows and dramatic turns in a game, and I think WPA (for all its flaws) measures that better than anything other stat.

Sure, the HR in the bottom of the 11th is no more valuable than the HR in the bottom of the first that allowed the game to go into extras, but guess which one I will remember?

Like a lot of stats, it's not the most predictive or stable stat over the long run, but it adds value and I'm glad that it is calculated and published by someone.
54. Walt Davis Posted: August 17, 2014 at 09:23 PM (#4773153)
2nd and 3rd, no out? A big inning is quite likely. 64% doesn't seem high at all.

I wouldn't expect the expectation table to be a problem either but that seems high to me. They put the run expectancy around 2 -- also seems a bit high. But regardless, down 2-0 entering the bottom of the second seems like a situation where you should have more than a 1/3 chance of winning.

Again, a double in the top of the 2nd that did not directly score a run had nearly as much impact on the final outcome of the game as the first THREE shutout innings of pitching. After the double, p(win) was 66%. After 4 innings, with a 1-0 lead, it was still just 66%.

I'm not saying it's wrong, I'm just surprised it "matters" so much.
55. Misirlou doesn't live in the restaurant Posted: August 17, 2014 at 09:40 PM (#4773165)
There will be a high correlation between a game's WPA leader and a poll of fans who watched the game as to who was the "player of the game".

I agree, but let's not read too much into it. In the Boog powell 3 HR game, he was second in WPA to the starter Milt pappas, and only slightly ahead of the guy who relieved him for 2 innings, Stu Miller. If Powell had struck out in the first, he would have been the WPA leader, and the player of the game.
56. Misirlou doesn't live in the restaurant Posted: August 17, 2014 at 09:46 PM (#4773169)
Javier Baez MLB debut game, he hit a game winning HR in the 12th inning. He was 6th on the team in WPA that game, behind, among others, a guy who walked in his only PA.
57. DKDC Posted: August 17, 2014 at 10:22 PM (#4773186)
#55,

Yeah, there are certainly holes to be poked in it, and silly examples aren't hard to find.

But it's pretty much the only stat that is multiplied by LI, just like all of our memories are multiplied by LI.
58. McCoy Posted: August 17, 2014 at 10:25 PM (#4773189)
Well, he was 0 for 5 with 3 strikeouts and a lineout with the bases loaded before that homer. He was a big part of the reason the game went to the 12th inning.
59. Ron J2 Posted: August 18, 2014 at 12:53 PM (#4773557)
#6 No. There's a standard errror of just under 4 wins/162 GP between pythags and actual winning percentages. And most of that is down to timing.

Or to restate, there's "only" a 92% correlation between RPA and runs created (WPA or WPA/LI is more complex, but is in the same general range)

Timing -- even if it's not an ability -- is of value in explaining the events. And I'm cool with using WPA as a tie breaker. I mean the standard error of even the best offensive metrics is on the order of 5 runs for a full time regular and I'm fine with minor adjustments based on any form of clutch state (like WPA)
60. Nasty Nate Posted: August 18, 2014 at 01:58 PM (#4773686)
Hr late in a 1-0 game is not more valuable than a homerun early in a 1-0 game

Things that happen later in the game are not the context in which an early homerun is hit.

Otherwise you end up with players supposedly adding value long after they have left the game.
61. McCoy Posted: August 18, 2014 at 02:02 PM (#4773693)
Timing -- even if it's not an ability -- is of value in explaining the events.

Timing is

You must be Registered and Logged In to post comments.

<< Back to main

News

Old-School Newsstand

Support BBTF

Thanks to
for his generous support.

Bookmarks

You must be logged in to view your Bookmarks.

Hot Topics

NewsblogRobinson Cano Reportedly Suspended for PED
(220 - 9:36am, May 20)
Last: Panik on the streets of London (Trout! Trout!)

NewsblogOT - Catch-All Pop Culture Extravaganza (April - June 2018)
(2027 - 9:27am, May 20)
Last: McCoy

NewsblogOTP 2018 May 14: Political leadership, fans needed to bring baseball to Portland: Guest opinion
(1251 - 8:56am, May 20)
Last: spycake

NewsblogOT - 2017-18 NBA thread (All-Star Weekend to End of Time edition)
(4530 - 8:02am, May 20)
Last: NJ in NY (Now with Baby!)

NewsblogWeekend OMNICHATTER for May 19-20, 2018
(133 - 1:00am, May 20)
Last: Randomly Fluctuating Defensive Metric

NewsblogHow gambling can help make baseball America's pastime again
(98 - 12:40am, May 20)
Last: Howie Menckel

NewsblogFortnite Addiction Is Becoming a Problem for Major League Basebal
(84 - 12:29am, May 20)
Last: GGIAS (aka Poster Nutbag)

NewsblogTaking Back the Ballparks - Colorado Rockies
(39 - 7:15pm, May 19)
Last: Omineca Greg

Gonfalon CubsHalf full or half empty?
(66 - 5:40pm, May 19)
Last: Moses Taylor, aka Hambone Fakenameington

(1976 - 4:43pm, May 19)
Last: AuntBea calls himself Sky Panther

NewsblogIllini rely on student who geeks out on baseball analytics
(2 - 2:51pm, May 19)
Last: What did Billy Ripken have against ElRoy Face?

NewsblogA face that could launch a thousand OMNICHATTER! for May 18, 2018
(105 - 11:47am, May 19)
Last: Perry

NewsblogPrimer Dugout (and link of the day) 5-18-2018
(119 - 7:43pm, May 18)
Last: RMc's Unenviable Situation

NewsblogThe Official Mets Fan Self-Immolation Thread
(829 - 7:25pm, May 18)
Last: Howie Menckel

NewsblogBraves fans flooded MLB's Twitter after controversial replay ruling erased a run
(104 - 5:06pm, May 18)
Last: Misirlou doesn't live in the restaurant

Page rendered in 0.4424 seconds
47 querie(s) executed