FTFA

So, somehow Lee's Cy Young year versus this year, which is considered by many to be a “down-year” is almost a draw in terms of effectiveness.

Look, we're all on board with wins being less than useful but this statement is really wrong. Lee has an ERA+ of 119 this, 167 in 2008. Simply put he prevented runs at a much greater rate in 2008 than he did in 2012 and that is a pitchers job.

I really don't like using predictive stats for determining a pitchers effectiveness in a given year. I'm fine with looking at xFIP and using that as a barometer that Lee is fine and will continue to be fine but a retroactive look at his season should be based on what did occur. What happened is what happened.

Just skip the middle man and stop putting so much importance on the CY Young award.

as well as is in the top five of FRA Fair_QS

If your Cy Young case for someone relies in his being in the top five (!) of a stat nobody has ever heard of, you really don't have any case at all.

fra fair_qs??

that reads like c plus programming

good gravy

Cliff Lee's stat line this year is very good, but no one with 5 wins in 26 starts is going to win a Cy Young (or come close) unless their team is historically awful AND they're Steve Carlton in 1972.

I feel so behind. I've never even heard of Fair Run Average.

Lee has been a lot better than his W-L record this year - his run support is bottom 10 in MLB - but using xFIP and SIERA is just bad statistics.

(See Colin Wyers on xFIP and SIERA, they do not add value to FIP/DIPS.)

Further, using a FIP-based stat for player value ignores hit prevention (a real pitcher skill) and ignores clutch pitching (both a real skill and a real aspect of pitcher value). Lee has been relatively poor both at preventing hits and at pitching in the clutch. As a result, Lee is 11th in the NL in both ERA+ and WAR by Baseball-Reference's numbers. He's been a very good pitcher, but he's not a Cy Young candidate.

(See Colin Wyers on xFIP and SIERA, they do not add value to FIP/DIPS.)

What the piece actually says is that they add value, but that added value is a small enough amount that he doesn't consider it to be significant, given the difficulty involved in calculating them.

If you want the most accurate prediction possible, you'll probably disagree with him.

What the piece actually says is that they add value, but that added value is a small enough amount that he doesn't consider it to be significant, given the difficulty involved in calculating them.

Wyers' conclusion:

Simply producing a lower standard deviation doesn't make a measure better at predicting future performance in any real sense; it simply makes it less able to measure the distance between good pitching and bad pitching. And having a lower RMSE based upon that lower standard deviation doesn't provide evidence that skill is being measured. In short, the gains claimed for SIERA are about as imaginary as they can get, and we feel quite comfortable in moving on.

They do not offer any utility, they just offer a smaller deviation leading to a lower RMSE

In baseball terms, FB/Con is a worse predictor of HR rate than HR/Con. xFIP's supposed predictive utility is entirely a statistical mirage. It replaces a better predictor of HR rate with a worse predictor.

SIERA is more complicated, but there is no evidence that it's measuring something real about baseball. All it has going for it is RMSE, which is probably another effect of lower variance rather than a baseball issue.

FB/Con is a worse predictor of HR rate than HR/Con

If this is true, then I don't understand why anyone is still using xFIP.

If this is true, then I don't understand why anyone is still using xFIP.

Read Wyers. People like to test stats for predictive utility using root mean square error. RMSE is affected by the variance of the measures used, and FB/Con has a significantly smaller standard deviation than HR/Con, which leads to a lower RMSE for xFIP as a statistical artifact. It's not actually predictive, it just looks like it is if you use the wrong tools to test it.

To get rid of wins you need to replace it with something relatively simple to understand, and appreciably better. Quality starts probably should be that stat, it roughly measures what you want a starter to do every game, but it doesn't seem like it will ever have enough traction with the masses.

I mean, instead of Pitchers Wins, it's almost better to use Starts Won. You'd be giving credit to starters for games they pitched terribly in, but got bailed out by their bullpen/offense, but unlike Pitchers Wins you'd also give them credit for games where they kept their team close enough that it could rally after they were pulled.

People have been discussing this subject for years...

This all depends on how you define a pitcher's job.

There is the Welch/Morris level: "A pitcher's job is to win games."
There is the King Felix level: "A pitcher's job is to prevent runs. From there, it is up to the rest of the team and other variations to determine how many games are won."
And now the Cliff Lee level: "A pitcher's job is to prevent homers, other hard-hit balls and walks. From there, it is up to the defense and other variations to determine how many runs are allowed."

In my opinion, the reason pitcher wins continue to have resonance as a statistic is because they reflect something that actually happened - the team won a specific game - and, not only "something" that happened, but the main point of the enterprise. The goal is to win. One drawback, I think, to sabermetric "win" stats is that they don't tie to real, actual wins; they tie to some sort of context-neutral theoretical measure of wins. Detroit has more WAR on BB-Ref than Baltimore, for example (~32 to ~26-27), but Baltimore's won 5 more games. I get that this reflects "luck" and Baltimore's season has been fairly fluky, but, at the end of the day, Baltimore's won more games than Detroit, and I think folks like their win-statistics to match up with reality at the team level.

The problem with pitcher wins, in this regard, of course, is that batters and baserunners and fielders also contribute to team wins, not just pitchers. I created a set of player wins and losses (pWins, pLosses) based on play-by-play data that tie to real team wins and losses. I also constructed a set of context-neutral records (eWins, eLosses) that are more comparable to WAR and the like. So far, my data source for this is Retrosheet play-by-play data, so I only have results through 2011, but I think it's interesting stuff. While it's not simple to calculate, I think to some extent it is "relatively simple to understand" (at least conceptually) while still tying to actual team wins in a way that I think retains that link to reality while doing a better job of doling out credit and blame among major-league baseball players. (Sorry if this comes off as too self-serving/self-promotional)

It may sound crazy to consider (Cliff) Lee a Cy Young candidate, but consider this fact: He ranks second in xFIP and SIERA, as well as is in the top five of FRA Fair_QS and also ranks in the top-10 in FIP.

Ignoring wins isn't crazy, but giving an award based upon sierra and fip is beyond crazy, it's so far beyond stupid, that it's going to take a fantastic voyage type of vehicle to locate the brain of the person uttering this ####.

To get rid of wins you need to replace it with something relatively simple to understand, and appreciably better. Quality starts probably should be that stat, it roughly measures what you want a starter to do every game, but it doesn't seem like it will ever have enough traction with the masses.

Now that is a train I can get on. Quality start is one of the best "new stats that is easy to understand' created in the past 30 years. In a just world it would supplant the idiocy of the win stat for pitchers.

The problem with pitcher wins, in this regard, of course, is that batters and baserunners and fielders also contribute to team wins, not just pitchers. I created a set of player wins and losses (pWins, pLosses) based on play-by-play data that tie to real team wins and losses. I also constructed a set of context-neutral records (eWins, eLosses) that are more comparable to WAR and the like. So far, my data source for this is Retrosheet play-by-play data, so I only have results through 2011, but I think it's interesting stuff. While it's not simple to calculate, I think to some extent it is "relatively simple to understand" (at least conceptually) while still tying to actual team wins in a way that I think retains that link to reality while doing a better job of doling out credit and blame among major-league baseball players. (Sorry if this comes off as too self-serving/self-promotional)

For the record, I've had some fun playing around with your numbers and I think your system is super interesting conceptually, as well. So at the very least, you're not the only person who has fun with it :)

Lee has been a lot better than his W-L record this year - his run support is bottom 10 in MLB - but using xFIP and SIERA is just bad statistics.

(See Colin Wyers on xFIP and SIERA, they do not add value to FIP/DIPS.)

I remember discussion about that article here. What struck me at the time is the final chart, where he regresses each predictor to have the same standard deviation as xFIP:

_______ERA___FIP____xFIP___SIERA
RMSE___1.94__1.87___1.86___1.84

It seems to me that if we regard the 0.03 difference between SIERA and FIP as marginal, then we'd need to regard the 0.07 difference between FIP and regressed ERA as pretty marginal as well.

In other words, if his argument is right that the advantages of SIERA over FIP are illusory, it looks to me like he would also need to conclude that the advantages of FIP over regressed ERA is also illusory, which seems like a much bigger deal.

SIERA is more complicated, but there is no evidence that it's measuring something real about baseball.

Rereading the article, whats strike me now is that, if we're just worried about getting good predictions, SIERA really isn't that complicated. It's not really introducing any bold new assumptions, it's just fitting using up to quadratic terms instead of just linear ones. Granted, if you're trying to interpret the coefficients it makes things a bit more complicated, since there are more of them and they're multicollinear. But if we're just looking at what goes in and what the results are, it's really not that different, and the idea that it would only provide small improvement at best is not a terribly interesting claim.

Thanks, Arjun!

But it’s just stupid that baseball fans still look at Lee, as not so good, but Cueto as great.

Cueto 155 era+, 20 qs,(69%)
Lee 119 era+ 17qs(65%)..... (both have 4 unearned runs not a factor) that is enough. Their periphereals are about the same, while Cueto pitches in a significantly better hitters ball park.... where in the world does Lee enter the discussion? Yep, it's stupid to think the pitcher that is getting better results is having the better year.

It’s all a matter of perception

Right, so rather than trying to get rid of pitcher wins and losses completely, the best thing to do is to rebrand them. I've suggested "Yays" and "Boos" but for some reason it has failed to catch on.

Anybody consider using WPA for this task?

Anybody consider using WPA for this task?

WPA is useless as a stat for a starting pitcher, it's useless as a stat for the most part, but for a starting pitcher it's really doesn't bring anything to the table.

Yep, it's stupid to think the pitcher that is getting better results is having the better year.

I think you may have misread "not so good" to mean "not as good". Cliff Lee is not having as good a year as Cueto, but it is not like he has totally crashed and burned, as some people may think by just looking at the 5-7 record. 182 IP at 119 ERA+ is still pretty darned good, and he leads the league in fewest BB/9 and in K/BB.

FRA Fair_QS

An Italian monk who actually created sabermetrics at about the same time as DaVinci created the helicopter. Unfortunately the good friar was too far ahead of his time as baseball wouldn't become a big deal for another 400 years and his contributions were overlooked.

Quality start is one of the best "new stats that is easy to understand' created in the past 30 years

I kinda agree but it's a bit too vague itself. It would be nice to find an understandable way to distinguish the 6 IP, 3 ER QS (which aren't that common) from the 8 IP, 1 ER QS. And it's not clear why 6/3 should be a QS but 8/4 (and maybe 7/4) shouldn't.

But yeah, QS is a pretty good, simple (and completely arbitrary) proxy for "the pitcher did his job today."

it will always be difficult
to rationally combat
vagaries that
round ball
round bat

I kinda agree but it's a bit too vague itself. It would be nice to find an understandable way to distinguish the 6 IP, 3 ER QS (which aren't that common) from the 8 IP, 1 ER QS. And it's not clear why 6/3 should be a QS but 8/4 (and maybe 7/4) shouldn't.

I don't see a need for creating different versions of quality start to represent different levels of quality, but if someone wants to, then more power to them. My only complaint against quality start, is that it's possible to lose the quality start after earning it. That is something I wish could be changed. Once you have reached the minimum to get the stat, you should be awarded it, regardless of what happens afterwards.

It's just a listing of a pitchers performance and whether or not an average team is still in the game with that type of performance. Along with the pitcher being able to last long enough that the long relievers aren't necessary.

It's a horrible stat for pitchers prior to 1920 or so when complete games was over 50% of the starts, but once teams started reducing complete games below 50% of the starts, it supplants wins as the better stat.

SO, Brandon McCarthy should've won the CYA in the AL last year?

Oh, and a young man named Gio has something to say before Old Man Lee, I don't give a #### what stat you're using...

What's wrong with using Game Score as a "simple to understand" metric? No decimals, addition and subtraction only, gives you a good overall sense of value a pitcher provides in a game independent of his team's offense, provides a finer grain than simple QS counting, it's called "Game" "Score" so it still sounds like something the pitcher is doing to contribute to his team's chance at winning ...

And since it's additive, a cumulative game score also gives peak-y pitchers a chance to catch up on the workhorses.

Cueto's 2012 cumulative game score is 1,689.
Cliff Lee has 1,503.
Gio Gonzalez, 1,731.

RA Dickey has 1,809 with all those mid-season gems.
Clayton Kershaw is the NL leader with 1,869.

What's wrong with using Game Score as a "simple to understand" metric? N

Same as the other nerdy stats, it doesn't really tell you much other than a little bit about dominance(at the individual game level) and it shows up a lot of noise, add in the large numbers you are pointing out and it's not going to catch on. That is akin to using raw runs allowed without modifying it by "per 9 innings pitched" etc.

Wins and quality starts aren't hurt by negative games. I mean once you get the stat you have it. It doesn't average out over the course of the season with good and bad starts.

The win stat is fine. It's a document of what happened. Read too much into it at your own peril.

Wins and quality starts aren't hurt by negative games.

Of course Wins are balanced out by Losses. Almost everyone thinks a 20-5 year is better than a 22-12 year. And most people will compare the # of Quality Starts with the # of Games Started.