Jack Cust Watch

[skmsw]

for catchers it's generally 68-72; for ss/2b it's generally 72-75.

I get your point about "under $1mm being essentially the minimum," but I don't completely agree. Sometimes when you're eying a mid-season trade, or when you're selecting the last few relievers and bench players to keep, that extra $700,000 makes a difference, and in my opinion when you spend it you should get more than Miles.

But then, I also disagree that he's an average defensive 2b.

I do agree that theoretically we're arguing over a pretty trivial point; production from your third-string infielder and 25th man on your roster is kind've a pick-'em for the coach. But unfortunately, in this case we're paying more for, and giving more playing time to, the guy in question. And when you've settled into being a team that struggles to score runs, giving this many AB to a guy like Miles is indeed an issue, whether it is his fault or not.

[haltz]

Or more precisely, I don't think the numbers say what they're supposed to say.

Yeah, going back to my Cabrera and Miles comparison upthread, is it just a case of not making sense, or is there a rational explanation? I thought I had a pretty decent handle on the stat before I started comparing players. Is the .08 pLI discrepancy possibly that meaningful? I might be looking at this all wrong I guess.

I'm looking all over the internet and I'm not coming up with much.

[Popeye_Card]

To put it at an extreme, would anybody say Bobby Thomson's 1951 season was one of the most valuable seasons in history?

What do you consider valuable? It's one of the most storied moments in baseball history. Isn't that partially the point of sports as entertainment?

--P--

[haltz]

OK, how about the idea that each previous event effects the ensuing leverage state positively or negatively? So in hindsight we can say that run created in the 1st to make it 1-0 is just as valuable to the W as the 5th run in the bottom of the tenth to walk off? I read that posited, more or less, in the comments on a post at Dave Studeman's blog. I think that makes a lot of sense and answers my question.

So can we say that while interesting and useful in a lot of ways, that WPA probably isn't the greatest way to compare win contributions in a player-to-player sense?

[Colbert Rasmols]

jag,

I'm arriving late to this party, but I just wanted to say I've had some of the same thoughts regarding Miles. And I think a similar discussion could be had regarding John Rodriguez vs., say, Skip Schumaker. I never felt like Rodriguez's defense was so bad that it would probably cost us games. In fact, I can remember him making some very good plays. His defense and his baserunning may leave something to be desired, but how can a player as productive as he has been offensively be left off of the current roster? And why has he gotten so few big league plate appearances since we acquired him? Relative to much weaker hitters, that is.

[haltz]

Here's a scenario I found that some people used that might illustrate the point a little better -- if I actually have a point, I'm still not positive that I do (the actual numbers depend on the base-states, but it's the same idea either way):

Player A hits a 2-run shot in the top of the first for a WPA of +.166.
Opposing team scores a run in the middle innings.
Player A hits hits a 2-run homer in the top of the ninth for a WPA of +.142.

So Player A was stud that day as his two 2-run jacks powered the team to a 4-1 win and he earns +.308 WPA for just that game. A really great outing for player A.

What would have made it a better day for Player A is if he had grounded out in that AB in the 1st. That would have made his 2-run salami in the top of the ninth worth .637 WPA! He would then be twice as valuable (relative to the other version in WPA) as the 2-homer version of himself.

Unfortunately, if this player is Preston Wilson circa September, 2006 then his team will lose when Jose Vidro hits a 2-run walk-off single in the bottom of the inning. Probably should have hit that first homer in the lower leverage situation. Obviously the more runs the better, but then how does that play into using WPA to compare players?

[edited back since it was quoted]

[skmsw]

Here's a scenario I found that some people used that might illustrate the point a little better -- if I actually have a point, I'm still not positive that I do (the actual numbers depend on the base-states, but it's the same idea either way):

Player A hits a 2-run shot in the top of the first for a WPA of +.166.
Opposing team scores a run in the middle innings.
Player A hits hits a 2-run homer in the top of the ninth for a WPA of +.142.

So Player A was stud that day as his two 2-run jacks powered the team to a 4-1 win and he earns +.308 WPA for just that game. A really great outing for player A.

What would have made it a better day for Player A is if he had grounded out in that AB in the 1st. That would have made his 2-run salami in the top of the ninth worth .637 WPA! He would then be twice as valuable (relative to the other version in WPA) as the 2-homer version of himself.

Unfortunately, if this player is Preston Wilson circa September, 2006 then his team will lose when Jose Vidro hits a 2-run walk-off single in the bottom of the inning. Probably should have hit that first homer in the lower leverage situation. Obviously the more runs the better, but then how does that play into using WPA to compare players?

This is the sort of situation that reminds us -- WPA is not perfect.

But while I recognize its MANY limitations, I also recognize that it captures something important that nothing else quite captures.

It's true that in some ways it might, in the view of some, under-estimate the butterfly flapping its wings in China (the impact of events earlier in the game that lead up to the moment in question). But the reality is, while those events are important, the concept of leverage is that if P-Dub does NOT hit that first homer, there are still 8-plus more innings to go in which to amass runs and limit runs against. The existence of continued opportunity does indeed reduce the value of the first-inning blow somewhat. Not that it is not important, because you've made the case very effectively that overall, those runs ARE "just as important to winning." HOWEVER, if you fail in the first, many things can still happen to contribute to a win or a loss. If you fail in the 9th, it's a loss.

So, how does it "play into using WPA to compare players?" I think that as a global metric measuring runs produced or saved, it is poor. But as a measure of TIMING, it is very good. NOT to determine who has an ABILITY to get "key" hits, but as a post-hoc measure of who really did.

I would say, to compare a player's ability to generate runs, use other metrics. To compare whose performance had the biggest impact in CHANGING A TEAM'S PROBABILITY TO WIN, wpa is on the right track. Even in the face of examples like the one above. It's essentially a vastly superior replacement to using "close and late" or "post season split" data -- as long as you keep in mind that it is not a significant separate skill, it does a much better job than those things of getting at the question, "who made the biggest difference when the game was on the line?"

[haltz]

But while I recognize its MANY limitations, I also recognize that it captures something important that nothing else quite captures.

I think I'm just trying to get a feel for those limitations, really.

While concept of leverage is obviously very important and WPA (or WPA-like metrics) demonstrate(s) this uniquely, I've never really thought about it as a useful stat for comparing the offensive value of a couple of second baseman. I rolled with it earlier in the thread until it started to make less sense to me.

I've also surprisingly not come across much meaningful commentary other than a couple of interesting examples of where it's flawed and people saying it's either, A) a junk stat, or B) the ultimate stat. Surprisingly little reasonable middle ground. Someone in a BTF thread even went so far as to say it's been "debunked" when I don't know how that's even possible. It's obviously sound, I just think the utility is in question for some people.

Can we give the guy that hits the RBI single, down by two, in the second inning a WPA "Assist" when his teammate hits the 2-run game-winner in the ninth? OK, that's more of a joke.

Anyway, relievers? Absolutely. Position players? I don't know, a lot of the low-leverage run-scoring can turn out to be pretty important (although I can't argue with the math), unless the clutch added shows a significant difference in production that we can pile on top. Even then, since the clutch part doesn't have a whole lot of predictive value it might only useful for bar/message board arguments or awards voting.

So, how does it "play into using WPA to compare players?" I think that as a global metric measuring runs produced or saved, it is poor. But as a measure of TIMING, it is very good. NOT to determine who has an ABILITY to get "key" hits, but as a post-hoc measure of who really did.

I would say, to compare a player's ability to generate runs, use other metrics. To compare whose performance had the biggest impact in CHANGING A TEAM'S PROBABILITY TO WIN, wpa is on the right track.

I went and did something else before finishing my reply and then posting so I missed your edit. I'm all over the basic definition of the stat, and I think it's really cool in that regard. I just don't know how useful that is (for relative position player value, production, wins) if it undervalues previously scored runs somewhat, and if it doesn't then I kind of want to know how I should change someone's RC in relation to winning ballgames because of a certain WPA. At this point, I don't really think I should unless that player demonstrates a high "clutch."

It's essentially a vastly superior replacement to using "close and late" or "post season split" data -- as long as you keep in mind that it is not a significant separate skill, it does a much better job than those things of getting at the question, "who made the biggest difference when the game was on the line?"

Absolutely.

[skmsw]

I think I'm just trying to get a feel for those limitations, really.

I think the limitations are pretty clear --

the nature of what it measures is such that it has very limited predictive modeling power.

the nature of what it measures is such that it is heavily influenced by random variations in performance, chance events, and other things outside the player's normal control.

the nature of what it measures is such that it weighs performance in ways that PUT BACK the randomness that many metrics try to filter out (e.g. models and metrics based on linear weights try to remove team and luck dependent factors, whereas WPA relies on them).

I think keeping those limitations in mind, thus not mis-using wpa, it still is an interesting and useful measure. I believe that the two-run homer in the first inning, where there are still unlimited opportunities to influence the outcome of the game, really is a little less valuable than a two-run homer in the 9th. From the standpoint of probability, and from the standpoint of opportunity cost, the two events really do have different value and really do influence the probability of winning differently.

When it comes time to have an MVP conversation about an almost or already compelted season, or when discussing what has REALLY HAPPENED SO FAR, WPA informs it quite nicely with a sound way of examining who really improved his team's chances of winning the most.

When it comes time to evaluate a player's global contributions, or who is most likely to improve his team's chances of winning TOMORROW, put wpa away.

This gets interesting when using WPA on someone like Miles.

It's probably fair to say, "from a WPA standpoint, he has not hurt the team." Whether by luck or design, some of his meager hits have come at opportune times, so in hindsight, as weak a player as he probably is for the role he has been given, in theory he's contributed more to winning than to losing.

On the other hand, there is absolutely no reason to expect that to continue (WPA by its very nature makes no claims otherwise). By using a metric that weights timing, team-dependent variables, and random variation so heavily, we're crediting a replacement level player for the luck with which things have fallen his way, and counting on it to continue as the main argument for why he is an acceptable instead of bad choice.

His overall performance has been sub-replacement level. His successes as captured by WPA can, to some extent, be attributed to timing and random variations -- the sort that ANY player, repalcement level, above, or below, could be expected to experience.

[haltz]

This gets interesting when using WPA on someone like Miles.

This is back to where my questions originally were. And I don't know, I like WPA as a stat for what I'm convinced it's useful for and I really do have a pretty good grasp on most of the stuff you are saying here.

It gives a good measure of a players performance in the clutch, because it knows right then that is a high leverage situation (i.e., the hitter can't see into the future any better than WPA). But it literally cannot account for what happens after that situation so I don't see how it expresses a players real run/win contribution in a meaningful manner.

It's a great narrative, and it's cool to put a player "back in context" and evaluate him that way, but for comparing players I think whatever Slwts style metric of choice is going to be better. If you could reinsert the player with the actual win expectancy numbers being correct for the finality of that game, I would trust it as a value judgement.