Natstradamus Projections for 2015: REPLY HAZY TRY AGAIN

And now the moment you’ve all been waiting for: the Natstradamus projections for the 2015 season!

This year, the projection comes with a major caveat: If Ryan Zimmerman is no worse than a league-average defensive first baseman, the Washington Nationals are projected to win between 95 and 98 games.

Just to refresh your recollection (because Lord knows I need to refresh mine every year), I use a pretty simple projection system to come up with the Nats’ won/loss totals for the year. The whole thing is based off Bill James’s Pythagorean Expectation, and it’s a satisfyingly intuitive way to figure out how good your team is. In baseball, you win if you score more runs than you allow. The Pythagorean Win Expectation model reflects that.

Imagine the whole baseball season compressed into two halves of one inning at Nats park. First, we need to fill up the lineup card of players. Then, we need to know who these players are–I use a four-year trailing average as the basis for these calculations. Then, we need to figure who plays where and when. This is the greatest acknowledged weakness of my system, as I have somewhat arbitrarily assigned playing time based on my impressions of injuries, etc.

At the top of the inning, the visiting teams come to bat. The result of that half-inning will be Runs allowed. Any upper-deck crank will tell you that there are two ways you can allow a run, generally: by pitching badly (giving up tons of walks and homers) or by fielding badly (not getting to balls hit in the gap, dropping fly balls, committing errors). The same upper-deck crank will tell you that you can get out of the inning with good pitching (striking everyone out) and great fielding (robbing home runs, showing ridiculous range, gunning down runners with your arm). In my model, base pitching runs allowed off a pitcher’s FIP (I also use xFIP as an alternative, which normalizes pitcher home runs allowed to a league average home-run/fly-ball rate). Defense is handled by UZR, which handily expresses defense as the number of extra runs allowed or saved.

At the bottom of the inning, the Nats come to bat: time to score some runs. I use Weighted Runs Created for each batter. Since that’s a counting stat, I divide that by the number of plate appearances over the last four years to get the number of runs created per plate appearance. I multiply this by the number of projected plate appearances (an everyday player will get about 600 plate appearances). That’s the number of runs on the board.

When that’s over, I come to some conclusions.

The 2015 Nats pitching staff is projected to allow between 530 (using FIP) and 562 (using xFIP) runs. The 2014 Nats actually allowed 555 runs–and we were already amazed at how good the pitching was last year.

This is a good thing because there is too much uncertainty about the defense to have any real confidence. UZR is notorious in that it needs a pretty big sample size to stabilize–the rule of thumb is that 3 years’ worth of data for an everyday player is what you’d need for the stat to be of any real use. Unfortunately, Ryan Zimmerman, first baseman, is a relatively new creation. His limited time at first base resulted in a comically bad UZR/150 (i.e., what UZR would be if he played 150 games at first base for a year) of -109.1. If true, it would mean that Ryan Zimmerman’s first base defense would be costing the Nats over 20 more runs than he would stand to get them at the plate (~88, by my calculations). That’s hard to stomach. If we follow the model blindly, though, we end up with the defense costing the Nats’ excellent pitching just over 97 runs. If we back off and assume Ryan Zimmerman is at least a league-average first baseman, the defense improves significantly, actually saving just under 12 runs.

So, if you think Ryan Zimmerman is a 100-run liability at first base (and I doubt very much that this is the case), the pitching and defense combined concede between 627-659 runs (totals not seen since 2011, when the Nats allowed 643 runs). If you think Ryan Zimmerman is a league-average first baseman, the pitching and defense combine to allow between 518 and 550 runs (As good or better than the 2014 Nats).

Turning now to the batting, things are more straightforward. The model projects the Nats will score 652 runs. This is lower than last year’s observed total of 686 runs. The projection reflects my pessimism regarding Rendon’s playing time and the speed at which Span and Werth can return to the lineup. I will be very happily proven wrong on this point, though.

Add it all together, and you end up with 95 to 98 wins if Zim is at least a league-average first baseman. If he is the nightmare that the tiny and highly unreliable sample of data UZR has to work with, things are much less rosy, with the Nats winning between 80 and 84 games, and likely missing the playoffs.

What the hell is the matter with Bryce Harper?


Oh, you wanted more? Fine:

Bryce Harper doesn’t know how to play outfield very well yet. He makes up for this by being unbelievably fast. But where his bad route to a ball intersects with a wall, that same speed results in painful collisions.

That’s heresy, right? How dare I impugn the defensive skills of the National League’s fifth-best outfielder (by UZR) in 2012? Yes, Bryce Harper posted a ridiculous 9.5 UZR in 2012–that is, his outfield defense prevented 9.5 runs from scoring on the 2012 Nats. That’s pretty good, right?

Sure. But let’s remember that UZR is a highly unstable measure of defensive ability–that is, we need a pretty big sample size to be sure of what we’re looking at:

How many UZR opportunities do you need for UZR to be reliable? There isn’t any magic number. If I asked you how many AB you need before a player’s BA becomes reliable, you would likely answer, “I don’t know. The more the merrier I guess.” That is true with UZR and with all metrics. Of course, for some metrics, you need more or less data than for other metrics for an equivalent reliability. It depends on the sampling error and the spread in underlying talent, and other things that are inherent in that metric. Most of you are familiar with OPS, on base percentage plus slugging average. That is a very reliable metric even after one season of performance, or around 600 PA. In fact, the year-to-year correlation of OPS for full-time players, somewhat of a proxy for reliability, is almost .7. UZR, in contrast, depending on the position, has a year-to-year correlation of around .5. So a year of OPS data is roughly equivalent to a year and half to two years of UZR.

This makes intuitive sense, in a way. To gather information about a player’s defense, we have to put a ball in play somewhere near that player and give him a chance to make a defensive play. In some cases, that happens pretty often–think of a second baseman or a shortstop taking ground balls. Other times, it’s less often–think of an outfielder (like, say, Harper!) standing around as his starting pitcher (like, say, Strasburg!) strikes out batter after batter.

All of this is to say that even though UZR says Harper likely saved 9.5 runs for the 2012 Nats, that might not really be the truest measure of Harper’s defensive prowess in the outfield–although, again, it’s the best we can do for now.

But let’s just take the one year of data and look at it  more closely, OK?

Now, UZR is broken up into components, each of which makes sense if you imagine yourself playing baseball. First, as the ball is put in play, you have to react to the ball, get to where it’s going, and put yourself in a position to make the play. The distance you cover to get to that ball is your range. Thus, the runs that you save because you can get to the ball (instead of letting it go by you) are Range runs, denoted RngR. Bigger is better here–this means you’re actually getting to the ball and getting a glove on it. That’s good news for your ball club.

Next, once you’ve got the ball, you might need to throw it somewhere in a hurry. Maybe you need to turn a double play, or maybe you need to hit a cutoff man, or maybe you’re trying to gun down a runner at the plate. You need a pretty good and accurate arm to do any of those things. The runs you save because of your good and accurate arm are Arm runs, denoted ARM.

Finally, things don’t always go your way. Maybe you get to the ball, then boot it. Or maybe your arm is strong, but not accurate; or accurate, but nowhere near strong enough. To err is human, of course. Runs you cost your team because of your errors are–shocker–error runs, denoted ErrR.

Now, let’s look at each of those components for 2012 Bryce Harper. Harper has 5.4 RngR, which tells us that he’s got pretty good range for an outfielder. His arm is absurd: 6.2 ARM,  best in the National League in 2012. He does goof every so often, though, giving up -2.1 ErrR. That all adds up to his 9.5 UZR.

Now let’s do the hack thing and compare Harper to Mike Trout, another phenomenal young outfielder.

Trout posts a higher UZR of 13.3, fourth-best in the American League in 2012. What’s interesting is that he does this despite a not-so-great ARM (-3.8). So, if Trout costs his team runs with a weak/inaccurate arm, whence comes this outrageous defensive skill? Well, Trout doesn’t make many mistakes. In fact, he makes fewer mistakes than average, so that’s worth 0.4 ErrR. The real story is that Trout has absurd range, with RngR of 16.7–best in the American League!

Now, think about what that means, for a second. What does it mean when we say that an outfielder has good range? It means he gets to balls that other outfielders might not be able to reach. There are three parts to fielding a ball in play in the outfield: first, you have to know the ball is coming to you. Then, you need to figure out where that ball is going to be, and how best to get there. Finally, you have to run to that spot and make the play.

Mike Trout has been playing the outfield for quite some time. He played the outfield as an amateur. He has, in his short life thus far, seen many more balls hit towards him in the outfield than Bryce Harper has. No wonder, too– Harper was an catcher as an amateur, and was only turned into an outfielder after he turned professional.

Now, Trout and Harper are built similarly. I don’t have the data, but let’s assume that they have similar reaction times. They can see equally well. They run more or less the same speed (fast!), jump more or less the same height (high!). I submit that the vast difference between Harper and Trout’s range has nothing to do with the raw physical part of fielding–the running to where the ball is going to be. It has everything to do with the first and second parts of the process–seeing the ball and picking the best route to the ball.

This is a long way of saying that Harper’s propensity to run into, at, or through outfield walls has nothing at all to do with his willingness to play hard, or play “the right way,” or whatever. Harper runs into walls, or pulls up at warning tracks, or sprints towards fly balls in the gap because he just doesn’t know where he is on the outfield. He makes up for his lack of skill by employing that prodigious physical gift of speed. It is a testament to Harper’s raw speed that his range is as good as it is at all.

The trouble is, of course, when those sub-optimal routes, taken at breath-taking speed, intersect with walls. That’s why he has bursitis now.

The good news here is that there is every indication that Harper will learn to be a better outfielder as he gains more experience. This is exciting, because if he can get better jumps on balls and make fewer mistakes, he can bring his absurdly powerful arm into play even more often.

Bryce Harper is good at baseball. He is not yet good at playing outfield. He is probably going to become very good at that soon, though–and that will be fun to watch.

Again, notice: this isn’t about Harper’s mentality, or whether he’s playing “too hard,” or whether he believes he can blast through walls like the goddamn Kool-Aid Man. This is just about a 20 year old kid learning to play baseball better tomorrow than he did today.

Projecting the 2013 Nationals, Part 2: Pitching and Defense

In Part 1, we announced the starting line-up. Let’s see how many runs the pitching allows in 2013. My model conservatively estimates that in 2013, Nats pitching will account for 609 runs scored against the Nats, but defense will “save” 18 runs. Thus, the model conservatively predicts that 591 runs will be scored against the 2013 Nationals.

Here’s the table for pitching:

Pitcher Name Projected IP 4-Yr Moving Avg xFIP Projected Runs Allowed TOTAL RUNS ALLOWED
Stephen Strasburg 180 2.56 51.20
Gio Gonzalez 190 3.81 80.43
Jordan Zimmermann 190 3.71 78.32
Ross Detwiler 180 4.44 88.80
Dan Haren 190 3.37 71.14
Rafael Soriano 70 3.6 28.00
Drew Storen 70 3.46 26.91
Tyler Clippard 70 3.54 27.53
Ryan Mattheus 70 4.48 34.84
Craig Stammen 110 3.96 48.40
Zach Duke 90 4.34 43.40
Bill Bray 65 4.19 30.26 609.25

You will notice that my initial guesses for innings pitched for starting pitchers are quite low. We’ll tweak those later, but for now, I’m going to assume that these are good enough to go by.

A similar table of the defensive statistics would be tedious to recount, so let me sum it up with a few general notes:

  • According to these projections, the three biggest defensive assets on the 2013 Nationals are Ryan Zimmerman, Denard Span, and Danny Espinosa.
  • Ryan Zimmerman should save 7.6 runs–best on the team. The high number of defensive runs saved here underscores just how important it is for the Nats to keep him healthy.
  • Danny Espinosa has been the target of a lot of fan frustration lately, especially given his struggles at the plate. His defense, however, is outstanding. The model projects that he will save 5.2 runs.
  • The newest addition to the Nats defense, center fielder (and noted icthyophobe) Denard Span, is projected to save 4.6 runs. Bryce Harper had a UZR of 9.7 as a center fielder last year, so just looking at that, you might think that Span is a lousy center fielder compared to Harper. You’d be wrong. UZR is notoriously unstable–we need at least 3 years of data to get a good sample. Span actually posted a UZR of 9.0 as a center fielder for the Twins in 2011; likewise, as Twins CF in 2012, he posted a UZR of 8.5. As you can see, the projection for Span seems very conservative–but it takes into account some bad defensive years for Span (2008 and 2009). I would expect Span actually to outperform this projection.

Right, that wraps up the top of the inning. Tune in to Part 3, where we’ll discuss how the offense looks.

Projecting the 2013 Nationals, Part 1: Ground Rules & Starting Line-ups

Spring Training is well underway down in Viera. This is the season, then, of portents and omens–the latest and most amusing of which was the story of an osprey dropping a fish onto the Nats outfield. Not having any expertise in the art of augury, I don’t think I can really comment about the auspiciousness or inauspiciousness of such an omen for the upcoming season.

What I can offer you, however, is the results of my own admittedly crude projection system. Long-time readers will know that I like to think of the baseball season as a single inning of a baseball game writ very very large. In the top of the inning, we see the home team take the field, and see how good the pitching and the defense are at getting opposing batters out. In the bottom of the inning, we watch the home team at bat, and see how well they drive in runs. Then we count the runs allowed in the top of the inning and the runs scored in the bottom of the inning–if the home team scored more runs than the other team, they win.

If you want the nuts and bolts of my projection system, please, read the post I’ve linked above. It describes the general outline of the system as clearly as I can.

This year, however, I’m making a few changes to the Natstradamus projection system.

First, in pitching, I have replaced FIP with xFIP. I don’t know enough about home run/fly ball rates to tell, really, which pitchers are “lucky” or “unlucky” with respect to how many home runs they give up on fly balls. xFIP fixes that for me by normalizing runs allowed by a pitcher to a league-average home run/fly ball rate. Some pitchers get better; other pitchers get worse; but I think over all that might be a more fair way of evaluating pitchers for the purposes of this projection system.

Second, I have tweaked the defensive calculations slightly. Instead of using UZR, I have calculated a UZR/game, and then multiply that  by the number of games in which I expect each player to appear. Again, this is crude, and defensive metrics are highly unstable anyway, but hey, it’s all I’ve got.

Remember, my projections are based on four-year trailing averages for each stat. That is, they’re the averages of the past four years.

With those preliminaries out of the way, let’s start this year’s predictions off by going through the 2013 Nationals’ projected 25 man roster:

Starting Rotation

  • Stephen Strasburg, xFIP 2.56
  • Gio Gonzales, xFIP 3.81. I do not believe Gio will be subject to a suspension for his alleged involvement in the Biogenesis scandal. I explained my view on the situation here.
  • Jordan Zimmermann, xFIP 3.71.
  • Ross Detwiler, xFIP 4.44
  • Dan Haren, xFIP 3.37.

Starting Position Players

  • Adam LaRoche, 1B.
  • Danny Espinosa, 2B
  • Ryan Zimmerman, 3B
  • Ian Desmond, SS
  • Bryce Harper, LF
  • Denard Span, CF.
  • Jayson Werth, RF
  • Wilson Ramos, C
  • Kurt Suzuki, C. I have Ramos and Suzuki splitting playing time evenly.


  • Chad Tracy, OF/3B
  • Tyler Moore, OF/1B
  • Steve Lombardozzi, IF/OF
  • Roger Bernadina, OF


  • Rafael Soriano, xFIP 3.6, Primary Closer
  • Drew Storen, xFIP 3.46, Primary Set-up, Back-up Closer
  • Tyler Clippard, xFIP 3.54
  • Ryan Mattheus, xFIP 4.48
  • Zach Duke, xFIP 4.34, Left-handed long reliever/Spot starter
  • Craig Stammen, xFIP 3.96, Right-handed long reliever/Spot Starter
  • Bill Bray, xFIP 4.19, Left-handed one-out guy. This is probably the most controversial pick; others might put Henry Rodriguez or Christian Garcia here instead. But I’m going to assume Bray heads north with the club.

No surprises, then. Stay tuned as we discuss pitching and defense in Part 2 of our projections.

Baseball Eve!

The Boys Are Back in Town!

No real insights for you today on the day before Pitchers & Catchers report to Viera. Federal Baseball already has some early photographic evidence of baseball returning to Viera. Highlights include Jordan Zimmermann and Drew Storen rocking the quasi-official Beastmode T-Shirt introduced to the ’11 Nats by Ian Desmond and made famous by Michael Morse. But who’s that shaking hands with Tyler Clippard? The #tigerbeatbaseball girls want to know. (It’s not Ryan Tatusko, though. I checked that already.)

Two statistically-related things that I’ve been thinking about lately, though:

Lost in Translation

Given the number of major league players and prospects who play in the Latin American winter-ball leagues in Venezuela, the Dominican Republic, and Mexico, it’s remarkable to me how hard it is to get reliable statistical information out of those leagues. The leagues have their own stats pages, to be sure. For instance, the Venezuelan League’s stats pages are pretty comprehensive. But it’s not exactly easy to find the player you’re looking for. Moreover, calculating advanced statistics like wOBA and wRC is pretty much impossible. The worst has got to be wRC, because it depends on calculating a league average wOBA. To do that for the Venezuelan league, I’d have to key in all the data for all players into another spreadsheet and run the calculations from there. The calculating isn’t too bad, but the data entry will take more time than I’m willing to commit (it’s not like sabermetrics is my job, y’know–and if it were, I’d be pretty terrible at it).

As an aside: reading statistical tables and box scores in Spanish reminded me that my Spanish isn’t as good as it ought to be. Baseball stats are cryptic enough in English, but they can be pretty opaque in Spanish. Glossaries do exist, but I’ve had to bring in an outside consultant for help with a few.

If you’re at all interested in Latin American baseball stats, PuraPelota has the most complete database I’ve been able to find, but they can be a bit slow on the update cycle. I haven’t been able to find anything nearly as complete or helpful for any of the Asian leagues (Japan, Korea, Taiwan). I can’t understand why that would be so–surely the Sabermetric revolution has spread all across the baseball world? Nothing makes you appreciate the excellent work that Baseball Reference and Fangraphs do quite like dealing with the sparse data available for foreign baseball leagues.

Eye in the Sky

I’ve already written about this post at Línea de Fair, but I can’t help but take a closer look at one of the author’s objections to UZR:

…UZR, the measure employed to determine whether a fielder has more range than his teammates, and whether, on the whole, he can prevent opponents from from creating more runs. Joey Cora used to remind me how an infielder could be better depending on which pitcher was on the mound. This was due not only to the pitches, but also to the control the pitcher has over them. “What happens if a catcher calls for a sinker inside,” Cora asked. The shortstop moves a little, almost imperceptibly, towards the hole if the batter is right-handed. But if the pitcher leaves the ball outside, the roller could go up the middle of the infield. Result? A higher probability that the batted ball goes up the middle of the field and finds the shortstop further away from it–thus raising his UZR.

My initial reaction is that complaining that UZR may not describe that particular defensive alignment and situation like this is like complaining that the Ideal Gas Law won’t tell you exactly where to look for one particular carbon dioxide molecule in a tank full of compressed air.

Part of the problem, I think, is that UZR is the one baseball statistic in (quasi-) common use that is flat-out impossible to derive from other published statistics. As far as I can tell, the whole process depends on individual human beings watching game footage, noting where fielders are positioned, and noting where the fielder meets (or doesn’t meet) the ball.

Because I’m lazy, I figure that there must be a better way to do things–or at least one that isn’t so unbearbly tedious. We already have fairly sophisticated software that can track the location of, say, baseballs and baseball gloves as they move across a camera’s field of view. It should be a fairly simple matter to fix a wide-angle camera (or several) across a baseball field, record the whole game, and only have human intervention whenever the ball strikes the bat. An observer might tap one button when he sees the impact of the ball on the bat, and then tap another when the ball comes to rest (either in the glove of the fielder, or out of play). The end result might look something like the FlipFlopFlyball‘s defensive positioning infographic.

The genius of computing, however, would allow us to track each defensive move as a vector, with an origin point at wherever the defender started when the ball was put in play, and an endpoint at wherever he was standing when the play was over. I’m not so great at mathematics, but I imagine the resulting graphical representations (and statistical inferences!) that could be made from those data would be extremely useful in evaluating the range of any individual defender. Heck, maybe it wouldn’t be too hard to explain– if I only had a brain!

[Something I didn’t notice when I first saw that video in high school: MC 900ft Jesus is wearing a 1926 Washington Senators cap!]

The Limits of Prescience

A thread over at the Washington Nationals Fan Forums pushed back against some of my projections here and raised a few points that I neglected to address in my 2012 projections.

Margins of Error

Interesting projections but the missing piece would be an estimate of how much of a margin of error there would be for both the offensive and defensive estimates that would provide a range for the expected number of wins as opposed to a hard number.

This was a serious omission on my part. All projections have a certain degree of uncertainty built into them, and I really should have discussed the degree of uncertainty built into mine.

I took my method for calculating the projected runs allowed by pitching and defense from this site. The author tested this method against 7 years of complete season data from 2002 through 2008. As he writes:

I found the R^2 value. Not to oversimplify things too much, but this value basically shows what percentage of the variation can be accounted for by the model. The value ranges from 0 (worthless) to 1 (perfect). For my 210 data points, I had an R^2 value of about 0.78 (i.e. 78% of the variation).

That means that my defense and pitching runs allowed projections should be good for plus or minus 22%. That gives a lower bound of 482.84 runs allowed and an upper bound of 755.20 runs allowed.

If we assume that my offensive predictions are correct (a problem I’ll get to in a second), that means the 2012 Nats will win anywhere between 68 and 103 games

I know that’s an immense difference. I’m not sure how I could close that gap. UZR doesn’t account for pitcher or catcher defense, for instance. But even then, I think the method at least gets us in the ballpark.

The offense numbers are a lot more troublesome. I haven’t been able to do any real regression analysis to determine how good my model is–I simply haven’t had the time.

On the other hand our offense has way too many question marks to estimate the total number of runs scored with enough precision to come up with a meaningful value that can be used in a secondary projection as you did in calculating our win total.

Any type of future projection is likely to involve more than a little handwaving. Here, I’ve drawn an arbitrary line: all players included in this analysis are players on the Nats’ 25-man roster as of January 27, 2012, some 23 days before pitchers and catchers are due to report at Viera.

Individual Players and the Projections

Will Werth stay Werthless?

2011 Jayson Werth was astonishingly bad. I’m going to believe that his 2011 numbers are aberrations and not indicative of a “new normal.” I’m fairly confident that the 4-year average from 2008-2011 is a fair picture of what kind of player Werth is now–somewhere between his Philly days and the debacle of 2011.

Will Desmond, Ramos, and Espi improve or stagnate?

As far as Desmond and Espinosa, I have no idea. I don’t think I have nearly enough data about them to make any predictions going forward. Ramos, however, gets a nice bump from more playing time and more PAs. His wRC/PA isn’t terrible, so that’s to be expected.

Will Morse fall back to Earth?

I’m going to go ahead and say No. As I said in Part 3, Morse’s modest offensive outputs in 2008-2010 might make you think that he’s going to crash down to Earth in 2012. But, remember, I’ve taken a four year average of his wRC/PA over the same period. Giving Morse 600 plate appearances in 2012 gives a projected wRC of 97.00: exactly the same as his breakout 2011 “beastmode” year. Indeed, even if we throw out Morse’s 2011 season, running the same calculation over data from 2008-2010 yields a projected wRC of 90.00: Seven runs short of our prior projection and of the 2011 total, but still enough to make him almost as good as Ryan Zimmerman (projected for 90.69 wRC). Indeed, all of this taken together seems like pretty persuasive evidence that “beastmode” has been lurking inside him the whole time, and only needed to see enough PAs.

Will Zimmm get hurt again? Will LaRoche bounce back?

My response: Dammit, Jim, I’m a baseball fan, not a doctor!. I have really no good way of figuring out La Roche’s prognosis post-surgery, nor can I really know anything about the state of Zimmerman’s joints and muscles. The only real response I have here is that the four-year interval I picked should be fair to both men in terms of their expected production.

Who plays centerfield?

Again, I had to draw an arbitrary line and go with who was in the organization as of the day I began compiling the statistics. That means that for now, we’re looking at a DeRosa/Bernadina platoon in center field. This might not be ideal, but I didn’t want to mix players who weren’t officially in the organization into these projections. Blown Save, Win, however, has attempted to address the center field question in a recent post, where he suggests that perhaps the short-term answer is Rick Ankiel. I’ll have to go back and study this, obviously.