Somebody explain WAR already...

[CBL-Commish]

Originally Posted by Curtis

Sample size DOES come into play here, but it's still not a bad example. One more relevant might be David Wright's performance. Statistics tell me that Dave is a lifetime .309 hitter. Watching him play tells me that he's really two people — a .359 hitter and a .259 hitter. He's the streakiest hitter I'm aware of, and I don't know of any stat that would reveal that. (Which isn't to say that there isn't one.)

I think analyzing David Wright's performances would be greatly helped by stats, and probably aren't unusual at all.

Here's Wrights career BAs grouped by month and 10-points bins:
.230: 1
.240: 2
.250: 0
.260: 1
.270: 2
.280: 5
.290: 2
.300: 6
.310: 2
.320: 2
.330: 2
.340: 1
.350: 1
.360: 1
.370: 3
.380: 0
.390: 1
.400: 0

Looks like his monthly performances are grouped around .280-.310, with some outliers on each side.

I also took a look at Nick Markakis, since he seemed like a pretty consistent guy with a similar batting average, who I'm familiar with:

.180: 1
.220: 1
.230: 2
.240: 2
.250: 1
.260: 0
.270: 0
.280: 2
.290: 3
.300: 2
.310: 1
.320: 2
.330: 2
.340: 0
.350: 3
.360: 0
.370: 0
.380: 1
.390: 0
.400: 1

Markakis seems to be at least as streaky as Wright, he's had wider variances in his performances, and less grouping in the middle.

I believe that consistency isn't nearly as consistent as most folks believe.

[fhomess]

Not sure that's the best way to analyze it (months are rather arbitrary cutoff points), but I'm not a statistician so I've got nothing better to offer. That said, I tend to think that the impression of streakiness in a player (or clutchness) is tied to that player's performance in memorable situations. Doesn't have to be a postseason run, it could just be a difficult month for the team. For example, if Wright struggles during a Mets losing streak, and performs great during a winning streak, people might get the impression that he's streaky and cast blame or credit on him. Whereas if he performs well during a losing streak and poorly during a winning streak, people might not notice because he's not the one to blame for the poor team performance or because the team is doing well in spite of him.

You often hear about players who tend to perform better in the first half of the season or better in the second half. It's often hard to tell from the statistics if it's just sample issues or if it's a real trend for the player. I do believe catchers tend to fall off offensively as the year goes on because of the wear on their body.

[Curtis]

Originally Posted by CBL-Commish

I think analyzing David Wright's performances would be greatly helped by stats, and probably aren't unusual at all.

Here's Wrights career BAs grouped by month and 10-points bins:

Markakis seems to be at least as streaky as Wright, he's had wider variances in his performances, and less grouping in the middle.

I believe that consistency isn't nearly as consistent as most folks believe.

Originally Posted by fhomess

Not sure that's the best way to analyze it (months are rather arbitrary cutoff points), but I'm not a statistician so I've got nothing better to offer. That said, I tend to think that the impression of streakiness in a player (or clutchness) is tied to that player's performance in memorable situations.

First of all, my hat is off to CBL-Commish for looking that up. I tend to agree with fhomess that months are rather arbitrary cutoff points, but your analysis does show that any hot or cold streaks were mostly less than a month in duration, and therefore not terribly impressive.

I'm wondering a little what 'grouped into ten point bins' means. If it means 'rounded to the nearest ten', then I have to wonder how his career mark of .309 happened, given that nineteen of his months were under that and only eleven over it. I was also plainly amazed that he had only one month above .370. If it means 'we truncated the final digit' then it becomes more understandable, since that would yield thirteen months above and thirteen months below his mean (and six just on it, which seems to prove your point).

At any rate, your statistical analysis does contradict my memory/impression, so obviously it is useful… if for nothing else than prolonging an argument.

[CBL-Commish]

Originally Posted by Curtis

I'm wondering a little what 'grouped into ten point bins' means. If it means 'rounded to the nearest ten', then I have to wonder how his career mark of .309 happened, given that nineteen of his months were under that and only eleven over it. I was also plainly amazed that he had only one month above .370. If it means 'we truncated the final digit' then it becomes more understandable, since that would yield thirteen months above and thirteen months below his mean (and six just on it, which seems to prove your point).

Truncated. If he hit .283 that got thrown into the .280s bin.

And I just picked months because it was easy. Weeks or fortnights or something would have required some more serious spreadsheet, database, SQL kind of stuff.

[Syd Thrift]

Late to the party but I wanted to add that I don't believe anyone has "proven" that clutch hitting doesn't exist. People have failed to quantify it anywhere, and it seems to be at the point now where if it does exist, it exists on the level of the individual at-bat (i.e. under the criterion "late-close games in a pennant race with two outs and men on base"), but any statistician worth his salt will tell you that it is darn near impossible to prove a negative.

To the larger discussion, I get what's being said. The map is not the field. Players do not really have the ability to hit homeruns. They have the ability to generate good bat speed, to pick out fastballs and breaking pitches, to avoid swinging at bad pitches so pitchers have to throw them something out over the plate to get them to swing... but they don't actually have the ability to hit homeruns.

The thing is, while we can't really quantify all the things that go into a player hitting HRs, we can quantify the results, and the results are very, very stable from one year to the next, one league to the next, and so on. Maybe one day we'll be able to quantify those nuts and bolts themselves - pitch f/x is beginning to get to that point with pitchers, which we've long done a pretty bad job at quantifying in the first place - but as of right now this is the best we've got.

One thing that does really get my goat WRT stats though is the people who start whingeing about statheads and then start mentioning the Triple Crown categories or, say, wins for pitchers. Batting average, RBI, and homeruns are statistics, people, just the same as VORP or WAR. When you're saying "you statheads stay in your mom's basement blah blah blah", you're really not saying that personal observation trumps stats, by and large, you're saying that your stats are better than my stats.

[Killing Time]

Yeah. How do you prove a negative?

Statistics hasn't helped me any with my understanding of clutch hitting.

[Curtis]

Originally Posted by Killing Time

Yeah. How do you prove a negative?

That would depend upon how persnickety you are. When statistics show me that the odds of something happening are the same as random chance, that proves to my satisfaction that there is no cause and effect.

The poster who used to exhaust himself explaining and re-explaining to Styx how various Sabremetrics worked and what they meant would carefully qualify that just because the chances that something happened are the same as if they happened by chance, that doesn't mean that they didn't happen on purpose… just that you can't prove they did happen on purpose.

[fhomess]

Originally Posted by Curtis

That would depend upon how persnickety you are.

We need a stat for this!

[Goody]

Originally Posted by fhomess

We need a stat for this!

PEACE

P.ersnickity E.valuation A.glorithm/C.hances E.ndevoured

[Russ]

I think WAR for pitchers is flawed. The reason for me is that it relies heavily on FIP. While FIP definitely has its uses, it's more of a predictive stat. And, while there may be a strong correlation between the amount of HR, BB, and K a pitcher had last year, it's not telling me what I want to know.

Take these 2 pitchers:

A 200 IP, 75 K 65 BB 20 HR 60 ER 50 R
B 200 IP, 125 K, 45 BB 10 HR 90 ER 100 R

Now, I would say pitcher A was the better pitcher last year. I understand that might not be a popular view, but the way the really smart guys are looking at things like BABIP is changing. Take this excerpt from Tango's blog just today:

NOTHING is random. Nothing, nothing, nothing. There’s no such thing as 100% luck. None. Anyone who says that, feel free to remove his sabermetric card from his wallet.

The #1 job of a saberist is to tell you, the reader, HOW MUCH the observed metric is associated to the player’s skill. A pitcher’s BABIP has ALOT of noise. It’s not 100% noise. But, it’s alot. A hitter’s BABIP has alot of noise. Not ALOT, but alot.

I posted somewhere how much noise. For a hitter, it’s something like after 500 balls in play, half of it is luck. For a pitcher, it’s something like after 2000 or 3000 balls in play (I forget exactly), half of it is luck.

On the other hand, after less than 200 plate appearances, half of a hitter’s walk rate is luck. And after less than 100 balls in play, a pitcher’s tendency to give up groundballs is half-luck. So, you can see that we can tell VERY QUICKLY how much a pitcher is a groundball pitcher. It takes us a very very long time to figure out how good the pitcher is at preventing hits on balls in play.

So... I don't know. But I've never bought the argument that a pitcher has no control outside of the TTO. I've never bought that he has 7-10% control over it. It's unquantifiable right now. With the hit f/x stuff coming out, hopefully the smart guys can figure out a lot more stuff about defense and maybe about how much credit they should receive. There's just so much that goes into whether a BIP lands safely or not.

Also, the WAR formula for position players needs UZR which I expect would be difficult to replicate in the game.

[Syd Thrift]

I think there's a big difference between calling something straight-out luck and calling it, I don't know, second or third degree luck. Jarrod Washburn was insanely lucky last year with the M's in that he gave up an awful lot of warning track fly balls but not a lot of HRs with the team. He was second order lucky in that he played half his games in a ballpark particularly suited to him in Safeco Field and a fast, rangey outfield. Are the last two things really "luck" in the sense that we mean it? Not really. A pitcher who plays to the people behind him is exhibiting a skill. Even though that team could change fielders or he could get traded, that has to count for *something*... and I am not sure that a single metric like FIP really works so well for that.

That being said, something not feeling right is not in and of itself a good enough a good enough reason to scrap it. Tom Tango may be right that about half of luck rights itself for pitchers in about 2-3 seasons of starts, but if you look at the numbers themselves there is not a great deal of spread. There *is* a great deal of spread in a single season's worth of BABIP totals, but once you get to 2-3000 PAs you see a lot of guys with BABIPs around .290 and a few select folks - Mariano Rivera, knuckleball pitchers - with some lower totals (and extreme groundballers IIRC have slightly higher numbers but more than make up for it by and large with lower extra-base hit and HR numbers). Tango's findings aren't invalidating what Voros McCracken came up with in the late 90s, they enhance it.

[Russ]

Originally Posted by Syd Thrift

I think there's a big difference between calling something straight-out luck and calling it, I don't know, second or third degree luck. Jarrod Washburn was insanely lucky last year with the M's in that he gave up an awful lot of warning track fly balls but not a lot of HRs with the team. He was second order lucky in that he played half his games in a ballpark particularly suited to him in Safeco Field and a fast, rangey outfield. Are the last two things really "luck" in the sense that we mean it? Not really. A pitcher who plays to the people behind him is exhibiting a skill. Even though that team could change fielders or he could get traded, that has to count for *something*... and I am not sure that a single metric like FIP really works so well for that.

That being said, something not feeling right is not in and of itself a good enough a good enough reason to scrap it. Tom Tango may be right that about half of luck rights itself for pitchers in about 2-3 seasons of starts, but if you look at the numbers themselves there is not a great deal of spread. There *is* a great deal of spread in a single season's worth of BABIP totals, but once you get to 2-3000 PAs you see a lot of guys with BABIPs around .290 and a few select folks - Mariano Rivera, knuckleball pitchers - with some lower totals (and extreme groundballers IIRC have slightly higher numbers but more than make up for it by and large with lower extra-base hit and HR numbers). Tango's findings aren't invalidating what Voros McCracken came up with in the late 90s, they enhance it.

I don't disagree with anything you said here. The Westbrook example is a good choice to illustrate the shortcomings of FIP. I believe there are good things to come out of FIP and the DIPS theories. I just don't believe they tell an accurate story about past value.

[KBLover]

Originally Posted by Syd Thrift

That being said, something not feeling right is not in and of itself a good enough a good enough reason to scrap it. Tom Tango may be right that about half of luck rights itself for pitchers in about 2-3 seasons of starts, but if you look at the numbers themselves there is not a great deal of spread. There *is* a great deal of spread in a single season's worth of BABIP totals, but once you get to 2-3000 PAs you see a lot of guys with BABIPs around .290 and a few select folks - Mariano Rivera, knuckleball pitchers - with some lower totals (and extreme groundballers IIRC have slightly higher numbers but more than make up for it by and large with lower extra-base hit and HR numbers). Tango's findings aren't invalidating what Voros McCracken came up with in the late 90s, they enhance it.

Thing is, though, why are we saying "luck" when it's mostly physics at work, meaning we should be able to measure this stuff. For example, if a pitcher allows harder hit grounders more often than softer hit grounders - that could be an explanation. Instead its "well pitcher A is just "luckier" than pitcher B".

That doesn't make sense to me.

Why can't we measure these things in this age of high technology that's produced stuff like pitch f/x? Why can't we measure the angle of trajectory off the batter's bat? The force put behind a swing? The batted ball velocity?

Instead, why do we insist on using "luck" as the explanation instead of truly discovering what's at work?

Heck, why is it that .300 is the average BABIP for pretty much the existence of baseball? Why that number? What's at work to make that happen so steadily and what are the traits of those that defy it (either for higher or lower)?

[Russ]

Originally Posted by KBLover

Thing is, though, why are we saying "luck" when it's mostly physics at work, meaning we should be able to measure this stuff. For example, if a pitcher allows harder hit grounders more often than softer hit grounders - that could be an explanation. Instead its "well pitcher A is just "luckier" than pitcher B".

That doesn't make sense to me.

Why can't we measure these things in this age of high technology that's produced stuff like pitch f/x? Why can't we measure the angle of trajectory off the batter's bat? The force put behind a swing? The batted ball velocity?

Hit f/x is coming out this year and does everything you're talking about here.

[RchW]

Originally Posted by KBLover

Thing is, though, why are we saying "luck" when it's mostly physics at work, meaning we should be able to measure this stuff. For example, if a pitcher allows harder hit grounders more often than softer hit grounders - that could be an explanation. Instead its "well pitcher A is just "luckier" than pitcher B".

That doesn't make sense to me.

Why can't we measure these things in this age of high technology that's produced stuff like pitch f/x? Why can't we measure the angle of trajectory off the batter's bat? The force put behind a swing? The batted ball velocity?

Instead, why do we insist on using "luck" as the explanation instead of truly discovering what's at work?

Heck, why is it that .300 is the average BABIP for pretty much the existence of baseball? Why that number? What's at work to make that happen so steadily and what are the traits of those that defy it (either for higher or lower)?

Every game you watch has evidence of luck. The insane curveball off the plate that the batter cue-balls down the line for a run scoring double vs the absolute square hit line drive that the 3B or 1B or pitcher catches in order to save their face. Even the best ML outfielders get fooled on a massive swing that goes for a bloop hit over 2B. Every great fielding play could be termed a hit taken away, yet the physics on the batted ball was = to a hit 98%? of the time. Many fielders will admit that they just stuck the glove out there and the ball hit it, ie pure luck. You may say, "well he got there", and that would be true but how many times do they get there and not make the play? In most cases a hit is given if a fielder goes a long way but doesn't catch the ball. That result is bad (what?) for the pitcher. Is one result luck and the other physical?

I'm not denying the physics and the importance of them. I'm of the opinion that baseball as observed has a perverse ratio of good results from bad quality plays (poor physics) and vice-versa. More so than any other sport I think.

[andymac]

Originally Posted by KBLover

Thing is, though, why are we saying "luck" when it's mostly physics at work, meaning we should be able to measure this stuff. For example, if a pitcher allows harder hit grounders more often than softer hit grounders - that could be an explanation. Instead its "well pitcher A is just "luckier" than pitcher B".

The reason we use "luck" is because the data shows us that over time it is extremely likely that pitcher A and pitcher B will pretty much end up in the same place when it comes to batted balls. Sure, there are physics at work, but the data suggests (and batter f/x will likely show) that the variance in that physics will be correlated to a much larger extent with the batter than the pitcher. What will be curious to see is whether there is any significant correlation with the pitcher at all.

[sansterre]

Originally Posted by Killing Time

Yeah. How do you prove a negative?

Statistics hasn't helped me any with my understanding of clutch hitting.

A very interesting study on this subject was conducted by Nate Silver and was published by Baseball Prospectus in the book 'Baseball Between the Numbers.' Their methodology was fairly intuitive. They have a system for evaluating actual run value of batters by using a run expectancy matrix on each individual hit.* They then compare the total run value of the player's contextual performance with the run value of an evenly distributed performance with the same total stats. (Ie, a batter that hits a HR every 20 at bats is assumed to have a 1/20 chance of hitting a HR with men on, where the batter in reality might have hit a HR 1/15 times with men on.) They then took the difference of these two and labeled it 'clutch hitting'. This makes intuitive sense - players who hit better in more important situations will come up with a positive value here, and reasonably be said to be more 'clutch'.

The results? They are somewhat difficult to summarize, but I will do my best. Basically, Silver concluded that clutch hitting does exist, and for some players more than others. But, that it plays less of a role than most people think - the difference is not very large. (His estimate from the mathematical model he constructed was that scoring was 70% hitting, 28% dumb luck, and about 2% clutch hitting.) In 2005 David Ortiz was worth an extra 3.5 wins in pure clutch hitting, but the season before -.3, before that +.2, before that -.3, and so on. In other words, he was very clutch that one season, but besides that, there was no evidence of clutch behavior.

Since 1972, Silver found about eight batters that averaged a full win's worth of clutch hitting per 650 PAs (if you're curious, Matt Lawton is #1.) And correspondingly, there were about four that were worth a loss of clutch hitting. (Royce Clayton was worst.) So the effect is not irrelevant, but those are a dozen batters of all those that played in the last three and a half decades. Silver took the study farther, and found a correlation in certain player traits. Players that struck out little and made good contact (Jason Kendall, etc) tended to be better clutch hitters generally than the opposite, and this is intuitive. Players with good plate discipline and good contact are less likely to end their at bat having achieved nothing.

In summary of the summary: clutch hitting exists. There are a very few hitters that are respectably clutch (and anti-clutch). But generally, you're better off with a good hitter than a clutch average hitter - the increased value in the latter's clutch performance will likely be insufficient to compensate for the general superiority of the former. So it does exist, but is not a huge deal (but still worth noting), and that players with good plate discipline who make good contact tend to be slightly more clutch than most.

Hardly crazy conclusive, but it seemed quite rigorous to me, Nate Silver is a sharp guy, and the results are very intuitive, to me anyways. I hope I've added to the discussion.


* A run expectancy matrix is a chart with the expected run value for an average team in each of the men on base / out situations. To figure out how valuable a steal with a man on first and one out would be, you would take the expected run value of a man on second with one out (the result of a successful steal) and subtract from it the expected run value of a man on first with one out (the original state.) The difference is the estimated run value of a steal in that situation. (The cost of getting caught can of course be determined by subtracting the original state from the expected run value of 'no men on, two outs', which is the resultant state.) In this context, the matrix was used for every hit. Man on first with one out, and you ground out? Take 'Man on first, one out' and subtract it from 'man on second, two outs', and viola! And so on. As you can see, this makes every batter's value incredibly context specific. Normal stats, Runs created etc, assume neutral contexts, that everyone is hitting exactly the same in every situation.

[KBLover]

Originally Posted by RchW

Many fielders will admit that they just stuck the glove out there and the ball hit it, ie pure luck. You may say, "well he got there", and that would be true but how many times do they get there and not make the play? In most cases a hit is given if a fielder goes a long way but doesn't catch the ball. That result is bad (what?) for the pitcher. Is one result luck and the other physical?

None of it is luck.

There's no mystical force or such that made the ball hit the glove. There's no mystical force of randomness that makes a fielder fail to get a ball after running 100 feet or that makes a "seeing-eye single" get through. If a fielder is incapable of making that play 98% of the time, that doesn't make the 2% "luck". It just means a better fielder (say Ozzie Smith vs Joe Fielder) would make the play maybe 10% of the time since he has a better skill set.

There's a reason why BABIPs of groundballs hit to different angles is different than others, but I don't think that reason is luck but physical ability (timing, reaction time, reading the ball, able to manipulate his body in ways to get in position) as well as location of the ball (which also is physics - ball hits bat, physics dictates what happens). Some plays will be harder to make than others (harder to field balls, fielder reactions, etc) - that's not luck, imo.

To me, all of that happens based on physics.

Stat models might not can capture this, players might not know "why" they were able to get to the ball, but that doesn't mean the reason it happens is pure luck.

[KBLover]

Originally Posted by andymac

The reason we use "luck" is because the data shows us that over time it is extremely likely that pitcher A and pitcher B will pretty much end up in the same place when it comes to batted balls. Sure, there are physics at work, but the data suggests (and batter f/x will likely show) that the variance in that physics will be correlated to a much larger extent with the batter than the pitcher. What will be curious to see is whether there is any significant correlation with the pitcher at all.

If it correlates very strongly to hitters - wouldn't that make finding out how it works on the pitching side even MORE important? What skills the pitchers that DO have significant impact have over the resulting batted ball physics - and how do we find them? Not just say "unlucky pitcher" or the like?

The rarer the skill, the more desirable and sought after, no?

I would think if hit f/x does show, say, 85% of it is impacted by the hitter (just an illustration) then finding pitchers who can maximize that 15% would be of high order importance instead of just tossing it aside as luck for pitchers.

Maybe I'm off base here, but that seems to make sense to me - find the rarer skill and put more time in understanding it so you can seek it out for your team and gain an advantage.

[fhomess]

Originally Posted by KBLover

None of it is luck.

There's no mystical force or such that made the ball hit the glove. There's no mystical force of randomness that makes a fielder fail to get a ball after running 100 feet or that makes a "seeing-eye single" get through. If a fielder is incapable of making that play 98% of the time, that doesn't make the 2% "luck". It just means a better fielder (say Ozzie Smith vs Joe Fielder) would make the play maybe 10% of the time since he has a better skill set.

There's a reason why BABIPs of groundballs hit to different angles is different than others, but I don't think that reason is luck but physical ability (timing, reaction time, reading the ball, able to manipulate his body in ways to get in position) as well as location of the ball (which also is physics - ball hits bat, physics dictates what happens). Some plays will be harder to make than others (harder to field balls, fielder reactions, etc) - that's not luck, imo.

To me, all of that happens based on physics.

Stat models might not can capture this, players might not know "why" they were able to get to the ball, but that doesn't mean the reason it happens is pure luck.

This conversation is rather pointless unless the parties involved agree as to a definition of "luck". It seems to me that by your definition, luck doesn't exist anywhere in anything, because everything can always be explained by natural laws (this is an area of much philosophical debate that isn't worth going into here). By this definition, a sudden change in wind in the stadium that affects a high fly ball to either stay in the park or leave it is within the realm of physics, so it's not either lucky or unlucky.

Generally, most people consider luck to be the realm of things that are both out of an individual's control and unable to be reasonably predicted. By this definition, luck can even change depending on the perspective you approach it from. For example, the pitcher can control what pitch he throws and how he throws it. He can predict with some degree of accuracy, based on his knowledge of the batter and location of the pitch, whether or not the batter will swing. However, he can't reasonably predict (for most batter-pitcher combos), when and where the batter will swing. The result is that the location of a batted ball is, from the pitcher's perspective, partially controlled by luck under this definition. The batter may not feel it's luck at all in the sense that he controls when and where to swing, as well as how to pull his hands through the hitting zone (to either influence a pull or a slice on the ball). He's prearmed with the pitch information because he observes the ball coming out of the pitcher's hand. So while he can't control what the pitcher throws, he has enough time to react to the pitch and predict where it's going to go. So by this definition, the batter doesn't have much luck in this interaction. Now some would argue that the batter's ability to react how he wants to the pitch is variable, and he has little control over that variance, so there is still some luck involved from the batter's perspective. At any rate, I think we can agree that by this definition of luck, the aforementioned fly ball being buffetted by the winds meets the criteria of luck for both parties, since neither the batter or pitcher has control over the wind, and neither could have predicted the sudden change.