17 March 2016

Two Problems With Fangraphs' Projected Standings

A few months ago, Fangraphs published its projected win totals for 2016. They weren’t very complimentary of the Orioles as they projected them to win just 80 games and be the worst team in the AL East. The Royals, the 2015 World Series Champions, are projected to win only 77 games this season despite minimal changes. I thought it would be worth looking into why Fangraphs thought that the Orioles might struggle.

Fangraphs projects the Orioles to score 4.64 runs per game (752 total) while allowing 4.69 (760 total). If the Orioles did score 752 runs, then that would be the most they’ve scored since 2008. It's clear that the projections predict that the Orioles’ pitching is their weak point.

Fangraphs WAR depth charts tell a similar story.  They project the Orioles’ to have the eighth highest offensive WAR totals, but tied for the sixth lowest pitching WAR totals. However, what’s interesting about the Fangraphs WAR depth charts is that they project a total of 1085.8 WAR. Given that there can only be 1000 WAR in a season, they have an error somewhere. Neil Weinberg suggested “the reason for this is because no one has gotten hurt yet, so the depth charts are oversampling PA/IP from good players”. If this is the case, then Fangraphs is being overly optimistic projecting health from established players. The Orioles are depending on a number of star players and have limited quality depth available, so they are more vulnerable to injuries than other clubs. Such a flaw would overrate the Orioles’ ability.

Fangraphs doesn’t think much of the Orioles starting pitching. It is ranked third worst in WAR and are projected to have the second lowest ERA. They also project the Orioles’ starters to throw the second lowest amount of innings in the majors, even fewer than teams like the Reds and Phillies. The Orioles rotation performed poorly in 2015, and lost Wei-Yin Chen in free agency so such an event would be unfortunate but also plausible.

Fangraphs is higher on the Orioles’ bullpen. It is ranked sixth in WAR, and 20th in ERA. The Orioles’ rotation isn’t expected to throw many innings and therefore the bullpen will have to take over the slack. As a result, the Orioles top relievers of Britton, O’Day, Brach and Givens are projected to throw only 230 of the 523 innings thrown by the bullpen. The more innings thrown by long relievers and non-elite arms, the more runs that the Orioles’ bullpen will allow to opposing teams. It’s very possible for the Orioles to have elite relievers in the bullpen and still have a poor bullpen ERA.

It makes sense to presume that the Orioles pitching will struggle due to the weakness of the Orioles rotation, but not that they would be the second worst in the majors especially given their decent bullpen. I decided to create a table showing the total runs allowed by each teams’ pitching, and the amount of earned runs allowed by starters and relievers. In addition, by subtracting total runs from earned runs, I was able to derive the total amount of unearned runs allowed by each team.

I found that the Orioles rotation was projected to allow roughly 455 earned runs or the fourth most in the majors. Their bullpen was projected to allow 218 earned runs or the seventh most in the majors and all told the Orioles were expected to allow 674 earned runs good for the fifth most in the majors. It wouldn’t be particularly surprising if the Orioles allowed that many earned runs. The Orioles allowed 642 in 2012, 678 in 2013, 558 in 2014 and 646 in 2015 so allowing 674 in 2016 would be within reason even if I’d expect 650. The chart looks like this.


The reason why the Orioles do so poorly is that they are projected to allow 86 unearned runs or tie for worst in the majors with the Blue Jays. This is surprising as the Orioles allowed 31 in 2013, 35 in 2014 and 47 in 2015. 86 unearned runs is more then the Orioles allowed in 2014 and 2015 combined. No team has allowed 86 unearned runs or more in the past three seasons and these projections project that each team will allow an average of 75 unearned runs this year versus roughly 50 from 2013-2015.

The numbers look even more bizarre when looking at the five teams projected to allow the most unearned runs in 2016. The Royals are actually tied for third with 84 despite having an elite defense and the other two clubs are the Diamondbacks and Reds. All five of these clubs are projected to have top eight defenses in 2016 as measured by Fangraphs Field Metric and would probably be expected to have some of the lowest unearned run totals in the majors.


Neil did argue that fielding isn’t equal to unearned runs. This is completely true, but the field metric does have a correlation with unearned runs. The correlation between the fielding metric and unearned runs was -.436 in 2013, -.491 in 2014, -.489 in 2015. This is the expected result as it suggests there's a moderate correlation between having good fielders and limiting unearned runs. One shouldn't expect a strong correlation because factors such as sequencing impact the amount of unearned runs a team allows and is independent of defense. Likewise, teams with good range will have good fielding scores but may have many errors.

For 2016, the projected correlation is +.487. This suggests that there’s the same amount of certainty between 2013-2015 and 2016, but Fangraphs is projecting that good fielding teams will allow more unearned runs. This is counter to historical data and basic common sense. Given the relationship between the field metric and unearned runs, it would seem likely that there is a faulty addition or subtraction operation in their projection model.

Using historical data and 2016 projected defense, I built two models that project how many unearned runs teams should be expected to allow in 2016. I found that teams like the Orioles and Royals should be expected to give up forty fewer unearned runs than currently projected while teams like the Pirates and Padres should probably allow roughly 60 unearned runs. If so, when taking into account the lower run environment, teams like the Royals should be expected to win an extra two games, while the Padres should be another two games worse. Two wins may appear to have only a minor impact, but the difference between the third-best team and third-worst team in the majors is only 17 wins and another six wins would change the Royals from being in the cellar to being in first place. Small changes have a larger impact then one may think.

The Fangraphs model has 1080 WAR instead of 1000 and projects each team will allow 75 unearned runs rather than only 50. They may want to consider looking into these problems and fixing them.

46 comments:

Roger said...

MAtt, you mentioned how many more wins the Royals would get. How about the O's? If it's also 2 wins, that's the difference between last and third for the O's and maybe there wouldn't be as much griping around here...... :-)

Matt Perez said...

+2 Wins for the Os (82 wins)
+1 Win for the Rays (82 wins)
+1 Win for the Yankees (83 wins)
+1 Win for the Blue Jays (85 wins)
0 Wins for the Red Sox (88 wins)

Roger said...

Somehow the difference between 82 and 85 doesn't sound very big overall. As this is the beginning of the season injuries, recoveries, and deadline deals are going to have a lot of impact on these small differences. It seems hopeful to me.

And I agree that Fangraphs seemed to have some issues in their rankings even based just upon their WAR estimates for each team.

Jon Shepherd said...

One way to look at it is that the Jays are roughly 18% better.

Matt Perez said...

It doesn't seem that big of a deal because projections are so imprecise and baseball win totals are inflated. I mean, sure a team has a chance to win 162 games or win 10 games, but realistically a team is going to win between 50 and 110. Even then, most teams will win somewhere between 70 and 92.

A few wins look to be largely irrelevant because of the number of games that teams play. But when you look at results, it's larger then it seems.

Anonymous said...

That stat crap is usually a bunch of BS, leave it to the geeks who can't play the game. Woody Hayes once said about statistics: "That reminds me of the fella who drowned in the river with an average depth of three feet!"

Jon Shepherd said...

I don't know man. Woody has not won anything in almost 50 years.

Roger said...

Matt, of course you're right, but the difference between a projection of 78 (where we were before Gallardo and Alvarez) and 82 seems to me the difference between competitive and non-competitive with regards to the playoffs. And no one is projected higher than 88 and we know someone will win more than that. It becomes about breaks and in season moves. A breakout rookie (Wright? Bundy? Walker? Kim?) or a smart deal.

By the way, just looking at some spring stats. Anyone think that Marte, Walker, or Avery could beat out Reimold for the last roster spot? Assuming Rickard is the late inning defense specialist, that leaves the 25th spot open to the best hitter/pinch hitter that can play the OF. I'm not a big Avery fan but he is a LH......

Matt Perez said...

I don't think much of Spring Training in general. I accept that it has some minimal impact on predicting the future season and otherwise ignore it.

Roger said...

Matt, with regards to the regular season, yes. But with regards to who gets selected for the roster, spring means everything. Why in the heck do they have 40 man rosters and another 15 or so non-roster invitees if spring means nothing???? Spring is the time where we learn where the surprises might be coming from - good and bad. I know that pre-summer callups often have more impact but even those are sometimes influenced by spring results with strong early season minors followup. Or an injury at the majors level. That kind of comment is why the anonymous troll can quote Woody Hayes.

Jon Shepherd said...

40 man roster are a contract issue thing and STIs are often to ensure there are enough arms, gloves, and bats to go around.

Basically, 20-23 spots on this roster were decided going into spring training. Injuries and complete disaster decide about 2-4. The final spot or two is just to see if someone has something.

For the most part, Spring Training has more to do with getting in shape and figuring out who the next person or two coming out of Norfolk is.

Roger said...

Jon, I agree with you and that's basically what I was saying. There's a lot of good those last 2-3 roster guys can do if you get someone who has some near term upside. And your point about the next guy up from Norfolk is good too. As much as we talk about everything else here and there have been two posts about guys that were added this year at Norfolk (including Tolleson who isn't going to make the opening day roster), it seems like some discussion of who might get those last two roster slots and why might be fun to have. I don't think Matt's being so dismissive adds to the enjoyment of the discussions being had on these pages.

Pip said...

Jon, I wrote you about Rickard. He's definitely going to make the team, but have he and Kim revealed enough in the SSS that is Spring Training to consider that Rickars might earn the starting job?
Kim is owed ~7 million, which isn't nothing, so he probably has a spot, but Kim isn't giving us anything but reasons to regret.
Meanwhile, Hoes or possibly Avery might well supplant Reimold, Matusz opens on the DL, and McFarland in Norfolk, and who gets that BP spot??

Roger said...

Hoes or Avery will not take Reimold's place. Maybe Marte or Walker have a better chance. If Matusz starts on DL then McFarland will be in the bullpen (if healthy, if not, not sure who gets a crack at the LOOGY job - maybe Chris Jones? Cesar Cabral?). That's kinda why I was asking if anyone had a chance of beating out Reimold. I guarantee Rickard will not start at least until late spring/early summer only if things are going bad for Kim or Trumbo or Reimold.

Pip said...

I was wondering about Walker as well, but he's already confirmed as heading for the minors to work on outfield stuff.

Roger said...

Who confirmed that? The last I read is that Walker was being considered for LF.

Pip said...

I read on Roch's blog or at Camden Chat that Walker was being sent down to work on his outfield skills.
He might come up during the season as an OF, but he won't break camp on the 25-man.

Roger said...

I found it along with DD's remarks. I also found where they said Miguel has an option left. Sure would like to see Wilson/Wright break camp instead of Gonzalez. Seems like Matusz plans on being ready by OD but I wonder when Beliveau might be available. Forgot he was around. All the talk seems to be about Rickard but there is still a 25th man to consider. Without a bunch of other chatter, the default has to be Reimold, but I wonder what they're going to do with Marte.

Pip said...

Gonzo has to go down. It would really be bad if he made the team, sucked for a few starts and THEN got sent down.
And I think Worley has a chance too. My Money is on Wilson and Worley making the team, Roe and Paredes getting tossed, Matusz on the DL, and Gonzo being in Norfolk.

Matt Perez said...

"It seems like some discussion of who might get those last two roster slots and why might be fun to have. I don't think Matt's being so dismissive adds to the enjoyment of the discussions being had on these pages."

Eh, we all have areas that interest us. I'm not so interested in who the 4th guy off the bench will be or who will be the first guy up from Norfolk or who the seventh reliever in the bullpen will be. I understand that Spring Training performance will play a huge role in who gets those spots, but I don't care particularly much about the Jake Fox's of the world.

I tend not to react to much in Spring Training. Or to about the first quarter of the season. It takes about forty games to get a glimpse of how a team is going to perform in a given year.

Roger said...

The problem is that Worley doesn't look too good either. Wright is one day on, one day off. Miguel had one good start and a bunch of bad ones. Wilson is the most consistent of the bunch. If options mean anything then you're probably right, Wilson and Worley will get picked. Maybe it will give Miguel a kick in the butt to give his best. Miguel is much better in warm weather anyway. I hope Bundy turns into a serviceable middle reliever when these guys start flaming out.

Matt, I understand your position misguided though it may be (especially as to early season regular games). I actually thought that one of the reasons for statistics and analytics is to turn fans into bigger fans not to keep geeks (note: I'm a geek) from being bored with the whole thing. Plus on this particular Orioles team, the seventh reliever is going to be a big freaking deal and will have to pitch more than than the norm.

Jon Shepherd said...

The point of analytics is to make or better understand decisions.

Loud music between innings is supposed to reduce fan boredom.

Roger said...

Jon, I disagree. I love analytics as a better way to understand the game and the players in it. I, personally, will never have to make a baseball decision (unless I start playing fantasy again). Although it is fun to discuss and consider what the big boys should do but I want to understand the game that I love. Unless you guys are on the O's payroll then you're not having much say about their decisions either. I love this blog and I love the O's and baseball in general, but I don't pretend it's going to make any difference to the decision makers. At the risk of sounding like a Kevin Costner movie, baseball is equal measure romance/mystery and statistics/analysis. Statistics matter but so does Bucky Dent and the Pesky pole. We could still have this blog even if Bill James never existed.

Roger said...

2 pitches, 2 hits for Kim today. He's making a move. Hardy just got a hit, too, and having fantastic spring. I hope we've underestimated Hardy. This year is going to be a lot like 2012. Win a lot of 8-7 games in the 7th-9th innings or extras.

Jon Shepherd said...

Roger...first of all, uncertainty and the wondrous nature that delivers that uncertainty is part of the process. Second...even though I exist on the fringe of a couple front offices, it would not matter if I did or did not. Our interest here as fans is second to our interest on how life works. The part of life that we choose to tease apart here just happens to be baseball. We do not strive to make any difference. We strive to understand and if someone finds that interesting to read then great.

Roger said...

Jon, that's cool. It's why I'm here. That, and you guys occasionally talk about the O's.

:-)

Jon Shepherd said...

Programming notice...we will soon be focused on improving efficiency of producing paper towels for the next three months.

Matt Perez said...

Spoiler Alert: Putting Orioles sluggers on paper towels makes them 10% more likely to miss spills.

Put it this way. You can't study everything.

Pip said...

Jon/Matt:
I suppose I should probably be posting on a different article by now, but this is where everybody seems to be posting at the moment.
Rickard had another hit today, and another stolen base. I realize it's spring training, small sample, blah blah blah, and I take to heart the reservations you have each expressed so far.
But Rickard is young, and has an upward trajectory that neither enough of a spike to worry about luck, nor flat enough to think he lacks genuine potential.
You had given him a 40/40/45 line, I think.
Have you revised that projection based on Rickard's spring?
If not, what would it take?
Thanks!

Jon Shepherd said...

I think that line is still accurate. I think he provides more value for club right now.

Bulbasaurus Rex said...

While you're correct that something is screwy with Fangraph's projected unearned run totals, don't forget that unearned runs are correlated strongly with both defense AND pitching, as an unearned run is more likely to score when the pitching allows lots of earned baserunners in the first place to take advantage of the error. Very rarely does an unearned run score without any other baserunners allowed in the inning.

Roger said...

There are lots of innings with multiple hots where no runs are scored. Sometimes an unearned run is just an unearned run. There is no statistical doubt that the Orioles defense has limited unearned runs the last three years. To project the O's to have twice as many unearned runs as they have given up in any of the last three years is unreasonable. This year's pitching will not be that much worse than past years. It just won't. Even if someone bombs, they can't be worse than Bud was last year and we'll have time to find a fix.

Roger said...

As a side note, anyone see the play Machado made at third today? I hope he saves a few for the regular season. Not sure why anyone wants him to play SS; he is a natural 3B and hits like one too.

Roger said...

An interesting quote from Steve Clevenger on the Orioles' use of analytics.

Steve Clevenger changed uniforms this past winter. After two-and-a-half seasons with the Orioles, the 29-year-old catcher is now in camp with the Mariners. On Friday, I asked him if the teams utilize data differently.

“Definitely,” responded Clevenger. “It was more old-school in Baltimore, with Buck (Showalter) over there, and Dave Wallace as the pitching coach. Over here, Scott (Servais) played 14 years, but later in the game (1988-2001). He’s kind of old school, but at the same time, he’s bring new-school over here, with a new philosophy.”

Jon Shepherd said...

Orioles compartmentalize their statistics usage. Clevenger did not get to hear what the front office was actually doing. He heard it as it was distilled lower. I mean, the Orioles were on the front end of the shift movement as well as identifying second level run production. Stats usage is not the issue with the club. A hollowed out minor league system is where things are troublesome.

Pip said...

Jon, your reply to my query about Rickard was vague.
Are you saying that no, your estimate of his value hasn't changed, but he's giving more value to the team, meaning he's closer to 45 than to 40?
Your meaning was unclear.

Regarding our front office.
1) The problem is that Dan doesn't seem to have any kind of long-term plan or even a basic philosophy.
It's possible to disagree with a move or non-move while understanding the logic behind it, but Dan seems to use a coin flip for most of his moves. There's literally no logic to them.

As Fangraphs said," Dan needed an outfielder and went out and signed a DH."

Jon Shepherd said...

What I mean is that he is a fringe outfielder among a bunch of other fringe outfielders. Rickard does not exactly look over matched and he seems to be more versatile defensively, so he would be one to roll the dice. I think it is arguable who to go with. I do think some of the discussion that Rickard is a 1.5 win or better player is looking at the 7th percentile and not what is likely.

Jon Shepherd said...

75th percentile

Matt Perez said...

"While you're correct that something is screwy with Fangraph's projected unearned run totals, don't forget that unearned runs are correlated strongly with both defense AND pitching."

Huh. That's fascinating. So, your statement is probably incorrect.

The correlation between earned and unearned runs is only .23 which is not a strong correlation. I wouldn't expect a strong correlation for either factor. A regression model including both earned runs and field data has an r^2 of .193. It also suggests that the earned runs stat has extremely minimal predictive value when combined with fielding.

The problem with my analysis is that fielding has an impact on earned runs and therefore they're not independent of each other. I'd have to use FIP instead of ERA to be sure I'm correct. But I'm probably right.

But what's fascinating is that it has a huge impact on Fangraphs projected model. Earned runs has a large impact on unearned runs for their 2016 model --- even more so then fielding. Those two factors explain 42% of the variance. It appears that they decided to take your statement at face value and implement it in their model because it's logical. The problem is that that it's wrong and therefore further screws their model.

Roger said...

Matt, the obvious variable for correlation to unearned runs is errors not fielding per se. Bad fielding can result in more EARNED runs as well as unearned runs because a reduced range factor would turn some batted balls into hits (e.g. Orioles pitching would really stink if not for some good defense). Sometimes errors increase when the range factors increase (fielders getting to balls others would not touch). That tells me why fielding is not highly correlated to unearned runs. Can you correlate quantity of errors to quantity of unearned runs? How strong is it when quantities get involved? What's the proportion? Is it consistent from team to team (does 1 error = .75 unearned runs for the O's mean the same for the Astros)?

Jon Shepherd said...

Errors would correlate better to unearned runs because that is effectively the definitional connection. Fielding is considering the who shebang and not getting caught up in only the error aspect of it. Focusing on errors is kind of missing the point here.

Matt Perez said...

Steamer and Zips project fielders fielding score and not errors. Fangraphs and myself can't use data that doesn't exist to attempt to project unearned runs. Unless Fangraphs is receiving error data from Steamer and Zips and not sharing it with the public. Seems unlikely.

Roger said...

Jon, I disagree with that in some measure. The definitional aspect is why I called it "obvious" but an unearned run implies there was an error or passed ball but an error does not imply there will be an unearned run or how many unearned runs.

Matt, the data exists on errors and passed balls and unearned runs in basic reported data - does not require Steamer or ZIPS. Fielding data has too many conflicting variables related to unearned runs.

I guess part of the point is to determine if unearned runs are pitching independent which they are likely not to be. But if you can relate some variable (like errors) to the production of unearned runs then you can more reliably use them in team projections. But if you can tie fielding data directly to the production of errors then you get the same thing, What I'm getting here is that fielding may not be tied directly to errors. The issue with the Fangraphs projections is that they ascribe an improbably high number of unearned runs to the Orioles based upon what we know the likelihood really is. Why is that? Is it the way analytics uses fielding data or is it because no one is actually predicting errors specifically. Shouldn't fielding data predict actual errors?

Jon Shepherd said...

Roger, I think you are completely missing the point. Errors are part of fielding. The only reason why errors have meaning in this context is because you need to under stand unearned runs to determine runs allowed.

Roger said...

Jon, the point of the article and much of the subsequent discussion is that Fangraphs' projections on unearned runs are patently incorrect to the tune of two wins for the O's. I am offering an alternative method to predict unearned runs. If the result (i.e. unearned prediction) is incorrect, what is wrong? The equation, the variables, or the data?

I realize that errors are a part of fielding and, of course the fielding analytic is a good predictor of overall team performance with regards to wins. But in the case of unearned runs, in particular, the fielding metric does not seem to help and my assertion is that it includes variables that conflict with unearned run predictive value. I am suggesting to simplify the metric and then expand on it with more detail to create a stronger predictor.

Matt Perez said...

"Jon, the point of the article and much of the subsequent discussion is that Fangraphs' projections on unearned runs are patently incorrect to the tune of two wins for the O's."

Eh. The point is that their model suggests that as fielding improves, so do unearned runs. That's the problem. The two wins is largely cosmetic.

"Matt, the data exists on errors and passed balls and unearned runs in basic reported data - does not require Steamer or ZIPS. Fielding data has too many conflicting variables related to unearned runs."

If ZIPS or Steamer doesn't project the number of errors that a player will allow in 2016, then how am I supposed to use errors to project the amount of unearned runs a team will allow in 2016?

Fine, I can get error data for 2015 but that tells me nothing about Fangraphs projections for 2016.

"The issue with the Fangraphs projections is that they ascribe an improbably high number of unearned runs to the Orioles based upon what we know the likelihood really is. Why is that? Is it the way analytics uses fielding data or is it because no one is actually predicting errors specifically."

The way analytics uses fielding data? I don't know what that means. I can use analytics to build one model. Jon can use analytics to build another model. Fangraphs can use analytics to build another model.

You're mistaking a means for an end.

"But in the case of unearned runs, in particular, the fielding metric does not seem to help"

I have no idea why you'd think that. I've discussed in detail why it does help. There's a correlation between fielding and unearned runs allowed in both real life and in the projections.