Filter posts by...
Talent_Distribution
Thursday, May 03, 2012
I looked at how all the free agents did in the 2011 season. A team comprised of ONLY free agents, and randomly chosen, would have cost a team 160 million dollars, and in return, would have won around 70 games. (You also would have had to sign 40-45 players.)
In comparison, your typical team that ignores the free agent market, your Royals, Pirates, Rays, Marlins (except today), Twins, A’s, etc, those teams are a bit better than 70-win teams, and they do it at a below average payroll. And if you add up all the player development costs (minor leagues, scouting, baseball ops, signing bonuses, etc), probably has a baseball budget of say 100MM$. They have to sign some 200-300 players.
If 2011 is representative of what happens, then the free agent market is about 60% overpriced overall (with all the overpricing happening at the mid-to-high end). That’s if 2011 is representative.
My guess is that 2011 was a bad year for free agents, but that the free agent market is probably 25-50% overpriced, if we expand the number of years.
Wednesday, April 25, 2012
If you have (at least) a 12-game losing streak, can we predict what the rest-of-season record will be? Let me show you a really easy way to do that. First off, an (at least) 12-game losing streak means that in a 13-game period, you have one win and 12 losses. The losing streak starts with your last win, and then is followed by (at least) 12 losses. Everything after that 12th loss is undetermined. To forceast the rest-of-season record, all you have to do is apply regression toward the mean, and that means adding 35 wins and 35 losses.
So, our best guess for a team that is 1-12 is that they are a true 36-47 team. A true 36-47 team will win 43.4% of the remaining games. If we figure a team will play about 161.7 games in a season, then of the remaining 148.7 games (161.7 minus 13), we’d expect them to win 64.5 games (148.7 times .434) and lose 84.2 games.
A 1-12 team is expected to play 64.5-84.2 in their other games.
Poz was nice enough to compile this list of teams that had (at least) as 12 game losing streak at some point in their season. Their end-of-season record was 64.2 wins and 97.5 losses. If we remove their 1 win and 12 losses in their streak, that leaves us with 63.2 wins and 85.5 losses.
19 actual 1-12 teams ended up playing 63.2 - 85.5 in their other games. Or, what happened in those 13 games was almost exactly predictable of their other 149 games.
Also note that three of those teams played at close to, or just above, .500 above in their other 149 games. So, there’s a 16% chance that the Royals will play .500 ball in their other 149 games.
There are several ways to construct a team-based WAR metric. Let me go through a few.
1. Make it directly proportional to actual team wins and losses. Win Shares for example does that. WPA by definition does that. In every game, the winning team gets +.500 WPA and the losing team gets -.500 WPA. So, for example, if a team is 3-13, if you add up the WPA of each player, you will end up with exactly -5.0000 wins. It’s built into the system. HOW you actually assign wins to the players is really not germane to the issue. You are really limited to the play-by-play description, so if the event doesn’t show that Gutierrez stole a HR over the fence, the system won’t know to credit him for a huge play. You can’t say “Hey, Guti stole that ball! WPA sucks!” That’s because WPA wasn’t TOLD that Guti stole that ball.
2. Make it proportional to actual runs scored and runs allowed. Sometimes, runs and wins are not directly related, when you aggregate over a season. You could end up scoring more than you allow, but end up losing more games than you won. Some people take the view that run scoring is more indicative of the players than wins, and so, their system is only concerned to adding up at the run level. If a team scores 800 runs, then the system will try to figure out how the team scored 800 runs. An RE24-based approach is one such way to make sure it all adds up. You would of course apply this approach to both the offense and defense. This is important, and it’ll come up shortly. Again, like WPA, if you are not told that Guti stole a home run, then the RE24 approach may misassign the runs. But, that’s a shortfall of not telling the system all the information. The system is quite capable of handling the Guti-steals-HR, if only the system is told of that event.
3. Make it proportional to actual bases and outs. If you think about it, this is identical to step 2. Four bases = 1 run. Any runners left on base is absorbed as a minus to the guys who made the outs, etc. I only have this here, so I can talk about…
4. Make it proportional to SOME actual bases and outs. We give the HR +1.4 runs, regardless to how many runs the HR actually produced for that particular play, whether there was a disproportionate number of HR with runners on base or bases empty. Basically, this is Linear Weights (for both the offense and defense), and you can include SB, CS, and basic baserunning (taking the extra base, also given fixed weights, regardless of when they actually happened). Again, you do all this for both offense and defense.
5. Take step 4, and then regress each component a certain amount. A strikeout is mostly skill, so that counts almost as much as it should. A single has a huge random variation dimension to it, so we count those less than we normally would. Interestingly, since results on offense tells us more about the hitters than the results on defense, this would mean that we would regress each component differently for hitters and pitchers.
***
So, we come to the point of some twitterings yesterday, regarding the Royals. In Fangraphs’ version of WAR, the offense and the fielders follows the step 4 scenario, the pitchers follow the step 5 scenario. That means there’s going to be a gap.
Under scenario 1 (wins-losses), WPA will mostly satisfy you, and the Royals will look horrible. Under scenario 2, RE24 will mostly satisfy you, and the Royals will look bad.
Under the scenario 4, we see that the big difference in the Royals offense and defense outcomes is that the Royals defense has given up 20 more uintentional walks plus hit batters than the offense has gotten. (The Royals defense also has gotten 17 more strikeouts, which either means it almost cancels out those 20 extra free passes under scenario 5 or it means nothing at all under scenario 4.)
The Fangraphs version of Team WAR is a combination of scenario 4 for offense and fielding and (mostly) scenario 5 for pitching. So, strikeouts minus free passes is not considered under scenario 4, but is important under scenario 5. Fielding (UZR) is considered under scenario 4 for the defense, but the “opponent UZR” is not considered under scenario 4. Balls in play is considered under scenario 4 for the offense, but is ignored for the defense in scenario 5.
Is this necessarily a bad approach? No. Is it a good approach? I don’t know. Would a scenario 1-only approach be better? Probably not. A scenario 2-only approach (which basically includes sequencing of events)? Better than scenario 1, but still probably not.
What we REALLY care about is the TRUE TALENT LEVEL of the players involved. We’re not going to get that from 15 games played. We need to look at each player’s performance in context with his career to figure out his talent level. Indeed, after 15 games, you are better off just going with the pre-season forecasts than anything else, if you had to choose just one thing.
Therefore, a strict adherence to Scenario 5 is what you ultimately want. And that means looking not only at 2012 performance of the players involved, but also their performance in seasons past and their age.
So, I see why people on Twitter are bothered, and why some readers at Fangraphs are skeptical of the results. The good thing about the Fangraphs approach is that they are transparent about the process. It’s all there if you look hard enough. This is far better than the other approach to “power rankings”, which really could just as well be a whim more than anything, like coaches’ poll in NCAA. Better as in easier to understand and ultimately decide if you can move forward on it or not. With the other “power rankings”, I just ignore them altogether (which I have termed infernally useless), because no one has shown the evidence that they are any better than just going with the pre-season forecasts, after 15 games.
Monday, April 23, 2012
Bill suggests these characteristics, among others:
The real issue is that 19th century has no genuine characteristics of major league baseball. If you think of the things that make “major league” baseball major league--large crowds, media attention, high-salaried stars, professional umpiring, a weeding-out and selection process that goes on for decades and ensures that only the very highest level of players are playing in these games--19th century baseball doesn’t have ANY of them.
...
But if you go back to 1900-1909--not to the 1870s, but to the first decade of the 20th century--there were two instances of this happening in the MAJOR leagues; paying customers wound up playing in the games. If you go back a decade before that, it is likely that there were MANY examples of that happening. One can’t imagine that happening now in a good college program.
So then I would ask you: is that what you think of as major league baseball? But that’s just one tiny example; there are hundreds of things like that that make it clear, to me, that 19th century baseball does not have ANY real or meaningful characteristic that would identify it as major league baseball.
Well-put.
You can make the case that “Major League” in baseball didn’t start until the early 1950s. Something like three out of every eight superstar player is non-White, since Willie Mays was born. With such a huge amount of talent missing pre-Jackie Robinson, can we really think of that version of baseball as “Major League”?
I don’t follow basketball, but given that there is a far greater percentage of superstar non-white basketball players, then the Major League in NBA really only started whenever race was no longer an issue. (Can someone educate us?)
Consider even hockey, where something like 50% of 1st round picks are from outside North America. You can even make the case that Major Leagues in the NHL only started sometime in the 1970s, or later.
(And feel free to talk about NFL.)
And of course, Major League in soccer doesn’t exist in North America. I’d like to know more about the history of Brazil, Africa, and European Leagues in this regard, if someone knows it.
Thursday, April 05, 2012
Lots of science, and plenty of art. Focus on the part that you like, and accept the part that you don’t.
Read More
Friday, March 16, 2012
I’ll use a hockey example first, and then I’ll switch to baseball.
In hockey, goal scoring follows a Poisson distribution. If we have two distributions, one with a mean of 3.0, and another with a mean of 2.7, we can figure out how often the one with a mean of 3.0 will have a random value higher than the one with the mean of 2.7. Ties are broken down in sudden-death OT fashion. In this case, in a 60-minute game of a 3 goals per game team facing a 2.7 goals per game team, the better team will win 55.3% of the time.
Now, what if a game of hockey was only one period? Setting aside any “change of pace” argument, we can model this as simply a 1.0 goals per game (that is, a game is 20 minutes, or one-third as long as the standard game) team facing a 0.9 goals per game team. In that case, Poisson says that the better team will win 53.5% of the time.
As you can see, changing nothing about the sport other than the number of periods, we can drastically alter the home-site advantage. It’s all based on the number of confrontations. The longer the game, the more the confrontations, then, the more the gap in talent will override the effect of random variation.
If we look at how often teams are tied heading into the third period, which is the same thing as I’m talking about here with the one-period game, I’m sure we’d see this kind of result, that home-site advantage will drop proportionately as I’m showing here.
We can see that with baseball as well. Now, baseball doesn’t follow a Poisson distribution, but we can model it as well. A 9-inning game gives us a .540 win%, while a 1-inning game would give us a .520 win% (which is close to the empirical result).
We can go through this with any sport, and the same thing will happen.
This is most clear in tennis, where the chance of Federer, Nadal, or Djokovic losing a 5-set match to someone other than the other two guys is much smaller than losing a 3-set match. For example, say that the big 3 is up 2 sets to 1 against the #20 seeded player. What is the chance that they would end up winning one of their next two sets? It’s going to be pretty high. Now, suppose the #20 seeded player is up 2 sets to 1 against one of the big 3. What is the chance that this #20 seeded player is going to win one of his next two sets? I don’t know what it is, but it’s DEFINITELY less than when the roles are reversed.
If Tiger in his prime had a 33% chance of winning a four-round tournament, then what’s the odds of him winning a single-round tournament? It’s definitely less than 33%. Probably something like 10%. That is, if you looked at each day’s results, I’d bet that Tiger in his prime won something like 10% of his rounds (if he won 33% of his tournaments). Something like that.
So, when people compare the home-site advantage of various sports, and trying to explain why one sport has a “higher” advantage than another, it’s meaningless. It’s entirely dependent on the number of confrontations.
Thursday, March 15, 2012
I wish I would have thought of this.
Tom Tango’s Marcel projections, the smartest “dumb” projection system (or dumbest “smart” system), can be used to show this more specifically for Cleveland. In addition to forecasting the usual rate and counting metrics, Tom’s projections (found here) also include a measure of reliability. The reliability measure is a product of how long a track record a player has, the volatility of that track record, and their forecasted playing time for 2012. The less of a track record, the less forecasted playing time, and the more volatility, the less reliability in the forecast.
The strike part is untrue, which is why I took it out. Anyway, so the writer looks at the Indians to see how much uncertainty they’ve got with their expected roster. Which is a great idea. I’d love for someone to do this in opening day, using the community forecasted playing time, and then coming up with the average reliability figure for each team. That’ll tell us how uncertain the forecasts are for each team, and how much teams are relying on veterans compared to youngsters or bench players.
Wednesday, March 07, 2012
By , 10:39 PM
I’ve been doing some interesting research on all of the above - actually how they relate. The data I am using is this:
Average seasonal fastball pitch speeds from FG, which uses BIS data, not pitch f/x, I think. DL data, which for this research, was just number of days spend on the DL in a season. For most of this research, I only noted “on DL” or “not on DL” for a particular season. Finally, I used my own NERC (normalized component ERA) data which are ERA looking numbers produced from a BaseRuns formula on the underlying components, including WP I think. They are all adjusted for context - defense (as best as I can, using UZR data), park, opponents, and league. 4.00 is defined as a league average pitcher in both AL and NL and a 4.00 in the NL is equivalent to a 4.00 in the AL since I do league adjustments. Anyway, those details are not that important.
Here is some really interesting data and discussion:
Read More
Monday, February 27, 2012
Ah, the old Dave Berri conclusion just won’t die, will it?
As you’d expect, the early draft choices got a lot more playing time than the later ones. Even disregarding seasons where they didn’t play at all, and even *games* where they didn’t play at all, the late choices were only involved in 1/4 as many plays as the early choices. Berri and Simmons don’t think that’s a problem. They argue—as does Gladwell—that we should just assume the guys who played less, or didn’t play at all, are just as good as the guys who did play. We should just disregard the opinions of the coaches, who decided they weren’t good enough.
That’s silly, isn’ t it? I mean, it’s not logically impossible, but it defies common sense. At least you should need some evidence for it, instead of just blithely accepting it as a given.
That’s the difference between a subject matter expert and someone who is given data and doesn’t understand the selection bias of the data. (And so, shockingly, presumes there is no selection bias in the data.)
Here’s the punch in the face to Gladwell, Berri, and anyone who keeps spouting their nonsense:
As best as I can tell, the most valuable draft pick in MLB is whoever drafts the #1390th pick as a position player. That’s because the AVERAGE number of HR hit by ALL position players that made it to MLB is 427. 427 HR as the average? Yes, it’s true. Did I forget to mention that only ONE position player who was drafted #1390th has made it to MLB? That’s irrelevant. Because, we must assume that the player who DID make it MUST be representative of those that did NOT make it. So, the difference between Mike Piazza playing and Tim Casper, Damon Lembi, and Brad Smith NOT playing in MLB is… I dunno… luck? Maybe it is. But, the chances that Casper, Lembi and Smith were equals of Piazza? The chance of that is close to zero. Maybe at the time they were drafted, with the imperfect information available, they were equals. But, either there was a gross misvaluation, or, shockingly, Piazza, as a human being, developed. He learned. He improved. These guys are not automatons.
Piazza is not representative. He’s an exception, not only in terms of opportunity to play, but then in terms of performance of his play.
To argue that the average QB in the 1st round has the same performance, per play, as the QB in the nth round is ridiculous, if we limit ourselves only to those who ended up playing in the NFL. It presumes that the QB who did not play in the NFL is equal to those who did play in the round they were selected in.
It’s an embarrassing argument to make, and if it takes one million bloggers to get it through the heads of those who don’t want to listen, then so be it.
Thursday, February 23, 2012
Poz says it can’t just be beating 5:1 odds. We’d have major upset dozens of times each day.
Seems to me then, we’re talking 100:1 odds or something. If that’s the case, would the Americans beating the Russians in hockey actually qualify? I don’t know what the odds were then, but 10:1 at worst? If someone wants to run it through Poisson, get an expected goals scored and allowed, and report back the results. I gotta run…
So, Roy Oswalt may only sign mid-season, ala Peter Forsberg? Sure, why not.
Say for example that Oswalt is worth 7MM$ to 10MM$ before the first game, depending on which team would sign him, for 25-30 starts, or about 250K to 400K per start. As the season progresses, some teams have greater need for a top players, and others don’t. So, now, he’s worth say 100K to 500K per start. At that point, say he’s got 15-20 starts to go (which means, he’s now worth 1.5MM$ to 10MM$). And as the season is drawing to a close, he might be worth 0 to 1MM$ per start. With 8-10 starts, that’s 8MM$ to 10MM$ of value. (All numbers would have to include potential post-season.)
So, regardless of when Roy Oswalt signs, he should be worth the same salary to whatever team needs him the most!
Welcome to Peter Forsberg’s world.
Thursday, January 26, 2012
By , 04:36 AM
It is generally accepted in the sabermetric community that the AL is a better league than the NL, at least for the last several years. This is evidenced by the fact that the AL has a large advantage in IL games, although at least some of that edge could be something other than overall “talent”, although this is not likely and several people, including myself, have found little or no inherent advantage to the AL in IL games (e.g., the NL teams do not have any DH’s, so they have to juggle their lineup in AL parks, on the other hand, in NL parks, AL teams have to sit their DH’s or juggle their lineup, perhaps putting a bad defender - their DH - in the field, the AL pitchers typically are poorer hitters than the NL pitchers, etc.).
Read More
Monday, November 21, 2011
Clay notes:
In seven of the 24 leagues between 2000-2011, the second wild card actually had the fourth-best record in the league, beating out at least one of the division winners. In two more seasons the WC2 was tied with one of the division winners – so that quite often the “fifth” team has as good a claim to the playoffs as one of the division winners.
So, 9 times out of 24, the play-in team was as good or better than one of the division winners. Remember, a division is simply some artificial construct.
He notes however:
In eight of the last 12 years [in the AL], the new card winner (assuming things played out as before) would have trailed the real winner by 5 or more games, with the 2001 Twins finishing a whopping 17 games behind the “102 win but still just a wild card” A’s.
In order to give that statement some value, I’d like to know what the corresponding number is between the best and worst division winners in each year. Presumably, we’ve had plenty of 99-win teams and 87-win teams being the division winners. Somebody want to do that work?
Friday, November 18, 2011
Nolasco is one of the handful of pitchers that requires sabermetric study (Matt Cain, and a few others as well). As Derek points out, Nolasco’s K rate drops substantially (from 23% to 18%), his BB rate jumps substantially (4% to 6%), and his BABIP jumps a bit (.304 to .315). All of that conspires to make the number of runs allowed by Nolasco to diverge significantly from his peripherals (peripherals that we presume are not disproportionately influenced by the base/out state, any more than the average pitcher).
Andy in The Book noted that there is definitely a skill with men on base (and it could simply be for the reason of going from the full windup to not). Regardless. It’s a real skill that could impact by about 5 wOBA points. Not a whole lot of course. But given that a pitcher sees men on base 45% of the time, it’ll come into play often enough.
And of course, if we know more about the pitcher’s change in mechanics, then we’d be able to come up with a much better estimate. Unlike the clutch skill, this is more like a handedness skill: a real physical change in the environment that the player participates in.
Tuesday, November 15, 2011
Evidence.
The point of a uber-stat is to make sure it’s not biased, as well as being reasonable. Any teenager out there (me included at the time) has attempted to combine various outcome numbers to create a superstat. It’s literally impossible to create a stat where Pujols is not near the top, as long as you are being anything close to reasonable. But, it’s extreme cases, guys who walk a ton, or steal a ton, or are fantastic fielders, or hit a ton of HR, disproportionate from the rest of their outcome lines, that are the issue. Since we don’t get that many kind of extreme players, we end up with significant overlap between Elias rankings and WAR-based rankings.
It’s clear that one source of bias is relief pitchers.
Another way to verify the rankings (after all, why is WAR necessarily better than Elias?), is to create the A and B types based on how much they signed. (Of course, the compensation aspect affects the signing, and that can be adjusted based on Victor’s numbers.)
By the way, I think Elias is just the trustee, and not the creator. So, scorn should be reserved, if deserved, to the creators of the system, and not the messenger.
Monday, November 14, 2011
A reader asked about using plate appearances (PA) instead of innings played (IP) as the playing time component in replacement level.
Notably, leadoff hitters earn more value than a 9th place hitter, even if both are equal as hitters. Since replacement level is roughly .003 wins per plate appearance, and the gap between the number of PA for a 1st and 9th place hitter is 8/9 x 162 = 144 PA, that’s roughly 0.45 wins extra the leadoff hitter gets.
I agree, and PA is being used only because of lack of IP data being available. And general laziness. I think Rally accounts for this rWAR.(*)
(*) By the way, who started calling it bWAR? It’s rWAR.
The only issue is what to do with PH and DH. What you can do for them (and for all of them) is to look at fractions of PA for the lineup slot they are in. So, if they have 4 PA for the day, and their lineup slot came to the plate 5 times, then you give them 80% of a game.
Of course, we should do the same for innings played, so if they played 8 innings out of 9, that’s 89% of a game.
Generally speaking, a non-pitcher’s value is generally about 62.5% offense and 37.5% fielding. So in the above case, I’d credit the player with 83% of a game.
So, if you care about every 0.1 or 0.2 wins to give out, use this method instead.
Thursday, November 03, 2011
A few years ago, I asked what is the breakeven point if you don’t consider two 3-WAR players to be equal to one 6-WAR player.
Rally responded with:
From that formula, a 5.3 WAR player is equal to:
2 3.0 players
3 2.25 players
4 1.9 players
5 1.7 players
Which, I think, is a reasonable viewpoint among those people who give value to scarcity. I responded as follows:
Good stuff Rally, just the basis for discussion I was looking for.
Ok, if for trade purposes:
5.3 = 3.0 + 3.0
Then what would be the equivalent here:
x + 0.7 = 3.0 + 3.0
And once you do that, then answer these as well:
y + 1.7 = 3.0 + 3.0
z + 2.7 = 3.0 + 3.0
Give me your answers for x, y, z.
This trade proposal from Dave had spurred some interest in the comments and elsewhere.
***
Now, I do NOT want to talk about Dave’s plan specifically, not the least of which is because it involves the Mariners. But, we can talk about something along those lines. Let’s consider a 1-1 trade proposal (so we don’t waste time talking about the value of roster spots). And we’ll talk about Joey Votto, one of the best players in baseball, and Matt Moore, perhaps the best pitching prospect in baseball.
Here are my questions: you have Joey Votto for two years, and you are going to get two compensation picks in the draft once you let him go. What you are willing to pay for two years of Votto and access to those two picks? 70MM$? Give me a range too. Imagine that in the second half of 2011, he tanked, or that in the second half of 2011, he Bondsed-out. Now what are you willing to pay for him? 40MM$ to 80MM$? Something like that?
Now, what of Matt Moore? Highly rated entering 2011, Minor League Pitcher of the Year for 2011, tantalizing end-of-season start to his career (including playoffs). You get him for six years (and let’s say you get one comp pick for him). What are you willing to pay for him? If the stars align, maybe you give him what Sabathia and Lee and Jered Weaver get, so 130MM$. Of course, there’s always a certain level of uncertainty with guys with limited to no experience, not to mention he’s a pitcher. What’s the downside for him? 30MM$? So, maybe you’d realistically pay 70MM$ for him?
Let’s say this is where we are at: you are willing to pay Joey Votto 40MM$ - 80MM$ (average of 70MM$), and you are willing to pay Matt Moore 30MM$ - 130MM$ (average of 70MM$).
Now, let’s say that Joey Votto WANTS 26.5MM$ for those two years, and Matt Moore WANTS 26.5MM for those six years. Take as an assumption of fact everything I have said above.
Which of the two do you sign for 26.5MM$? Votto for 2 years, or Matt Moore for 6 years?
Monday, October 31, 2011
Dave Cameron, using Fangraphs’ latest feature, gives us his list.
Wednesday, September 14, 2011
Rob Neyer points out that you don’t even have to worry about wins and losses when it comes to divisional standings: just look at runs scored and allowed differential over the entire season. Because you get the exact same order for 28 of the teams (within the division anyway), with one little swap for the other 2 teams. Indeed, even for the Wild Card, you get a very strong ordinal ranking match. In the AL, you swap two teams (Cleveland and some other team). In the NL, it’s not so clean, notably because of the Padres (only -29 runs, yet 2nd to last in the league).
I’m sure you can look at other leagues, and get the same result. This is the big thing that is sold to the public: look at each game as if you are starting from scratch. The reality is that you don’t even have to start the game from scratch, because just keeping a running total of runs scored and allowed will give you the same answer.
The same applies for tennis, I am sure (and NHL, NBA, NFL, etc, though NFL has the advantage of having only 16 games). Basically, sell the public that there’s a winner or loser every game, when in reality, what we have are running totals of runs, points, or goals for the entire season.
from Amazon
Recent comments
Older comments
Page 320 of 344 pages « First < 318 319 320 321 322 > Last »Complete Archive – By Category
Complete Archive – By Date