Question about WAR (Machado vs. Trout)

[Phantom]

Stats measure what has happened, not what is going to.

Except that WAR is attempting to measure how many more games a team filled with replacement players would win by adding said player, so it is attempting to project as much as measure.

[Tryptamine]

fangraphs is measuring a pitchers value, but they're isolating only what the pitcher controls Ks,BB,etc and removing things the pitcher does not control like defense in an attempt to isolate their individual value, not the value of them plus other factors.

[survivedc]

fangraphs is measuring a pitchers value, but they're isolating only what the pitcher controls Ks,BB,etc and removing things the pitcher does not control like defense in an attempt to isolate their individual value, not the value of them plus other factors.

That sounds pretty good, I guess the problem I have is it seems like it devalues pitchers that pitch to contact. Someone like Chris Sale, who this year has made a point of trying to go deeper in games by not striking out as many (lower pitch counts) and is beating his FIP by about a whole run. BBRef has him at 3.5 WAR and fangraphs has him at 2.7.

Last year there was almost a 3 win difference between his BBRef WAR and the one from Fangraphs.

I should say that this seems like a bit of an outlier, when talking about one of the best pitchers in the game, but still interesting, I think.

[Phantom]

That sounds pretty good, I guess the problem I have is it seems like it devalues pitchers that pitch to contact. Someone like Chris Sale, who this year has made a point of trying to go deeper in games by not striking out as many (lower pitch counts) and is beating his FIP by about a whole run. BBRef has him at 3.5 WAR and fangraphs has him at 2.7.

Last year there was almost a 3 win difference between his BBRef WAR and the one from Fangraphs.

I should say that this seems like a bit of an outlier, when talking about one of the best pitchers in the game, but still interesting, I think.

That's because "pitching to contact" is nearly always garbage. Generally the pitches you "save" by having fewer at bats you give right back by facing more batters because of the extra hits you give up.

Sale has a .289 career BABIP against, and a .256 this season (.323 last). Must be this change in approach, right? Except he has allowed harder contact this season than last. In fact, he has allowed his highest percentage of hard hit balls this year than his career. Seems like a disastrous strategy for a pitcher who is "pitching to contact". That is, unless you are getting lucky.

[Frobby]

Except that WAR is attempting to measure how many more games a team filled with replacement players would win by adding said player, so it is attempting to project as much as measure.

Not really. And by the way, do they do the same for the batters? It shouldn't be a measuring stat for hitters and a predictive stat for pitchers.

[Phantom]

Not really. And by the way, do they do the same for the batters? It shouldn't be a measuring stat for hitters and a predictive stat for pitchers.

It's a predictive stat for both. Hitters just have less to normalize for.

Regardless, I think you are misrepresenting what it does. Do you think WAR shouldn't adjust for park effects, for example, because that's the type of "predictive" stuff you are talking about. If batting average adjusts for errors, does that make it predictive because it is attempting to normalize for things that were not the batters doing?

[Frobby]

It's a predictive stat for both. Hitters just have less to normalize for.

Regardless, I think you are misrepresenting what it does. Do you think WAR shouldn't adjust for park effects, for example, because that's the type of "predictive" stuff you are talking about. If batting average adjusts for errors, does that make it predictive because it is attempting to normalize for things that were not the batters doing?

I don't consider adjustments that are made for park effects to be "predictive." They simply take into account the run-scoring environment in which the player was playing.

I will quickly get out of my depth on this topic, since I really don't know exactly how BB-ref and fangraphs calculate WAR. But for example, if an adjustment for BABIP is made for the pitchers but not the hitters, to me that would be an inconsistent approach. I can see arguments for making BABIP-related adjustments or not as an overall method; I just don't see treating the pitchers and hitters differently (if they are).

[weams]

I don't consider adjustments that are made for park effects to be "predictive." They simply take into account the run-scoring environment in which the player was playing.

I will quickly get out of my depth on this topic, since I really don't know exactly how BB-ref and fangraphs calculate WAR. But for example, if an adjustment for BABIP is made for the pitchers but not the hitters, to me that would be an inconsistent approach. I can see arguments for making BABIP-related adjustments or not as an overall method; I just don't see treating the pitchers and hitters differently (if they are).

I do not see them as particularly predictive. Either.

[Phantom]

I don't consider adjustments that are made for park effects to be "predictive." They simply take into account the run-scoring environment in which the player was playing.

Yes, but then you are discounting the idea that a pitcher could be "pitching to the ballpark", yes? So you are discounting his skill in an effort to try to "predict" how he would pitch in a neutral park.

Now, obviously I don't believe that, but my point is both WARs are being "predictive", they are just drawing the line in different places. BBref has decided to only gather the low hanging fruit, and let in the other variables even at the risk of letting in a bunch of noise (in this case noise is luck). Fangraphs has decided to be pretty hard line about it in an effort to cut out any pitchers who are coasting solely on luck, at the cost of pitchers that have truly mastered things that were are luck for most pitchers.

So the question isn't "do you want a stat to be predictive or not" so much as what level of balance do you want struck in the battle of luck and skill in your WAR.

[Tryptamine]

That sounds pretty good, I guess the problem I have is it seems like it devalues pitchers that pitch to contact. Someone like Chris Sale, who this year has made a point of trying to go deeper in games by not striking out as many (lower pitch counts) and is beating his FIP by about a whole run. BBRef has him at 3.5 WAR and fangraphs has him at 2.7.

Last year there was almost a 3 win difference between his BBRef WAR and the one from Fangraphs.

I should say that this seems like a bit of an outlier, when talking about one of the best pitchers in the game, but still interesting, I think.

FIPs biggest failing is that it doesn't not take batted ball profile or type of contact into account. There are those who consistently induce weaker contact than others so it is only natural their babip will be lower. The problem with identifying FIP beaters is that it usually takes a several year sample size. If I remember correctly, SIERA attempts to take batted ball profile into account.

[Tryptamine]

I need to start proofreading better, this lack of edit button is killing me

[Frobby]

Yes, but then you are discounting the idea that a pitcher could be "pitching to the ballpark", yes? So you are discounting his skill in an effort to try to "predict" how he would pitch in a neutral park.

Now, obviously I don't believe that, but my point is both WARs are being "predictive", they are just drawing the line in different places. BBref has decided to only gather the low hanging fruit, and let in the other variables even at the risk of letting in a bunch of noise (in this case noise is luck). Fangraphs has decided to be pretty hard line about it in an effort to cut out any pitchers who are coasting solely on luck, at the cost of pitchers that have truly mastered things that were are luck for most pitchers.

So the question isn't "do you want a stat to be predictive or not" so much as what level of balance do you want struck in the battle of luck and skill in your WAR.

I am drawing a distinction between (1) a stat that adjusts for luck vs. skill vs. (2) a stat that is "predictive." I don't think the purpose of WAR is to predict anything. Its purpose is to place a value on past performance. For example, a young player doesn't get extra WAR because he is likely to do better with experience.

FWIW, Dave Szymborski's ZiPS model generates something called zWAR, which is a projection of the upcoming year's WAR for a player. That is a predictive stat, needless to say.

[Phantom]

I am drawing a distinction between (1) a stat that adjusts for luck vs. skill vs. (2) a stat that is "predictive." I don't think the purpose of WAR is to predict anything. Its purpose is to place a value on past performance. For example, a young player doesn't get extra WAR because he is likely to do better with experience.

FWIW, Dave Szymborski's ZiPS model generates something called zWAR, which is a projection of the upcoming year's WAR for a player. That is a predictive stat, needless to say.

Well, neither WAR tries to predict future performance, so it seems like your worries are unfounded. They try to normalize past performance to evaluate value in a somewhat neutral environment.

[Phantom]

FIPs biggest failing is that it doesn't not take batted ball profile or type of contact into account. There are those who consistently induce weaker contact than others so it is only natural their babip will be lower. The problem with identifying FIP beaters is that it usually takes a several year sample size. If I remember correctly, SIERA attempts to take batted ball profile into account.

FiP does factor in weak contact a bit by giving a pitcher credit for his HR/FB rate. I think I read a study once where FiP still beats SIERA in predictive value even if you only sample pitchers that have outperformed (or under) their FiP for three years in a row or more.

FiP does screw over guys who legit can induce weak contact, there is just no good way to find them. And since there are way more fake outliers than real ones, tossing them out all together makes some sense.

Hell, I would have bet my life on Wei-Yen Chen being a FiP unicorn, but this year he's showing he may just be another horse with a cone strapped to his head.