Go to the new http://midnight-baseball.com for updates from now on.
Matt Kemp recently voiced his distaste for being a potential fourth outfielder. He’s presently dealing with ankle issues, but I think most would agree that when he’s healthy he has the talent of one of the best outfielders in the league. He’s certainly wasted as a bench player, or a late-game defensive specialist, or a guy you put in when the other guy has a hangover. But Los Angeles is paying four outfielders something like a quarter of a billion dollars, so someone has to stomach this demeaning role.
OR DOES SOMEONE?!
In short, no. A study of all the outfielders to make a plate appearance in the last decade demonstrates that teams get 2389 PA for their outfielders, on average. The player with the fourth-most plate appearances got on average 304 of those PA, a mere 12.7 percent. That would indeed be a pitiful use of Matt Kemp, but again that’s just the average. A few teams have demonstrated a better balance of playing time, none more so than the 2007 Yankees. Let’s take a look at their example.
First, it helps that those Yankees got 2668 PA from their outfielders, well above the average, leaving more pie for everyone. That’s no problem; the 2014 Dodgers should have a powerful offense that generates more batting chances than the average offense does. In fact, last year’s Dodgers got 2524 PA from their outfielders, and they were a middle-of-the-pack team in terms of run scoring. We should expect at least that many PA from this year’s outfielders, given that injuries last year gave insane playing time to inferior players. Take a look: 10 Dodgers suited up in the outfield last year. That’s a sign something went wrong.
This year, health permitting, the Dodgers can mimic the 2007 Yankees and give 95 percent of their PA to their top four guys. Those Yankees were not only healthy and top-heavy, they were incredibly even.
Their top four guys each got at least 23 percent of total outfield PAs, leaving just 4 percent for the scrubs and replacement players at the very end of the bench. If you apply those ratios to last year’s Dodgers (2524 PA), then the consensus top four guys each get at least 580 PA, which is plenty, just shy of a typical starters’ amount.
That is of course just one possibility for 2014, one that treats the four Dodgers outfielders equally. If you subscribe to the idea that Andre Ethier is just a platoon player at this point, you could subtract say 150 PA from his total and spread those around among Kemp, Puig and Crawford. At that point Ethier becomes a $15 million part-time player, but he’s still getting a sizable part, and that’s the luxury of having the highest payroll.
Speed is listed as one of the five tools, but metrics for it are either lacking or impenetrable. Stolen bases have been around forever, and in tracking them we’ve learned a lot about speed on the base paths and who is the greatest of all time. Stolen bases start and dominate the conversation, but they don’t represent speed as much as they are a result of multiple factors, speed among them, of course, but also aggressiveness, timing, preparation (studying tape to learn pitchers’ pickoff moves), and opportunity. Some managers enforce conservatism on the base paths, and some players can slug just as well as they can run, so they bat in the heart of the order, more likely to reach base with a player ahead of them (can’t steal unless it’s a double steal). Also less likely to risk getting caught stealing when the next guy in the lineup is a good power hitter. Because Dingers and Ribbies are best enjoyed together, some players with good speed and the inclination to steal will miss out on opportunities.
We could spend all day noting the imperfections of stolen bases, the pappy of all speed statistics, but that’s unproductive, about as unproductive as writing down all the flaws in your real pappy. My goal is to find another statistic that complements stolen bases to help us better understand which players have the best speed, or whose speed impacts the game the most. (Those might be the same question.) To start, I looked at double plays.
Above you see how the percent of ground balls that became double plays has fluctuated over the last decade. They seem to have rebounded back to normal last season. You can also see that for the last eight years the AL has been more conducive to ground ball double plays (GDP). The difference is only a few tenths of a percentage point, but it is interesting, if you have an uneventful life like mine. I would guess the DH/pitcher disparity is the difference here. If pitchers are batting with less than two outs (i.e. the only time there could be a double play) and a man on first or multiple men on base, that pitcher is probably going to bunt. Meanwhile, a DH is usually selected for his power. If he was speedy they’d put him in the outfield. So a DH is probably more likely to come up in double play situations and ground into double plays during those situations.
So, with the league average rate of ground balls going for double plays, I compared the rates of individual players. In 2013, 6.5% of ground balls resulted in double plays (GDP/GB = DP%). Matt Holliday, in an extreme case, grounded into 31 double plays, out of 200 total ground balls hit. 31/200 = 15.5%. By multiplying Holliday’s total of 200 by the league average rate of 6.5%, we can say the league average hitter would have made only 13 double plays (xDP, x for expected). With the same amount of ground balls, Holliday made 18 more double plays than we would expect of an average hitter, thus “costing” his team 18 outs.
On the other hand, Norichika Aoki hit into only 9 double plays out of 328 ground balls, good for a 2.7% DP%. A league average hitter would be expected to hit 21.3 double plays, so Aoki “saved” his team 12.3 outs. Aoki is a fast player, Holliday notoriously slow and gimpy and getting old. These extreme cases jive with what we’d expect. In the narrow view speed doesn’t always break up a double play, but in the wide view it should, so let’s look at the players alongside Aoki and Holliday at the ends of the spectrum.
Again, xDP is the expected number of double plays based on the league average DP% of 6.5. DP- is the difference between a player’s actual and expected double plays (GDP – xDP = DP-). DP- gives good players negative scores, the lower the better, as in ERA- or FIP-.
The leaderboard was filled with speedy guys, the anti-leaderboard with catchers and portly players. It does seem like speed is the driving issue behind the differences at the extremes. Of course, guys at the top of the second graph had a lot of opportunities to bat with runners on base, which may have inflated their DP- a little. Even so, Occam’s Razor wins out: The reason fast guys populated the first board and slow guys the second is that DP- measures pretty accurately the difference between fast and slow players in terms of double plays. In general, slow guys hit into more double plays and fast guys fewer, regardless of external circumstances; those always even out in the long run.
Stolen bases are choices. Not everyone decides to steal, but everyone must run to first as quickly as possible when they hit a ground ball. In ground ball double plays, a few tenths of a second make all the difference–fast players get to first base before the ball does, slow players can’t. Double plays above or below average are therefore a good index of speed; and therefore DP- is a good index of speed, and a purer one than stolen bases, for there is no element of choice. Over the course of a season, a individual’s speed when running against a potential double play can make a difference of 15 outs or so. This is just one small way speed shows up, the start of an inquiry.
Fifteen extra outs may not sound like a lot, especially because that number represents the extremes, but I’ve started to mess with Tom Tango’s linear weights in order to convert the outs gained and lost to runs gained and lost. With what I have so far, Matt Holliday cost the Cardinals 3 runs and Norichika Aoki saved the Brewers 2. From there, it’s easy to get to -0.3 wins for Holliday and +0.2 wins for Aoki. So only fractions of a win, with admittedly crude calculations, but again, this is just a first look of one kind of situation in which speed is manifest.
Yesterday, for fun, I looked up all the National League position players who have accumulated at least 6 WAR in a season since 2005, and I mapped them out on a chart. Nothing fancy, just what you see below.
I was going to start the time line at 2000, but that would have made this thing monstrously large, since all Barry Bonds did was rack up WARs over 10 and up to 12. The closest anyone has gotten to 10 WAR since 2005 is Albert Pujols in 2009, who got 8.7 WAR.
And since that year, only two other NL position players have even cleared 8 WAR. Matt Kemp in this context looks a little more like a flash-in-the-pan, whereas Andrew McCutchen is probably just getting started.
These men are the best the National League has had to offer since megaBonds stopped producing at megalevels. It’s been nine years now, so we’re basically looking at an era. Any player who played through the bulk of these seasons better show up on this list at least twice if he wants to be a serious Hall of Fame candidate. Let’s look at the one-timers above.
Derrek Lee was a good-hitting first baseman who put together a great year. I don’t think you can describe him as anything more than that. Jim Edmonds shows up here in the twilight of his career. This graph wouldn’t be very useful in evaluating him, you’d have to go back to the ’90s. Even without that graph, I feel confident saying he’s not a candidate.
Oops, seems like I skipped Morgan Ensberg.
Miguel Cabrera went to the American League and continued to mash.
Alfonso Soriano was always an exciting talent, but his flaws with the bat and in the field kept him from performing at elite levels consistently. Jimmy Rollins was the best shortstop in the NL for a time, but that’s not enough, especially since his double-play buddy Chase Utley was setting a higher example the entire time. The Big Puma Lance Berkman was once as feared as a big puma should be. He had some great seasons before 2005, and deserves consideration. One thing helping the case of Adrian Gonzalez is that he switched leagues soon after 2009 and posted 6.3 WAR for the Red Sox in 2011. Now that he’s with the Dodgers, he can add to this chart and to his resume.
Andres Torres is the least likely name here except for maybe Morgan Ensberg. Torres played out of his mind for the eventual champion Giants. He patrolled center better than anyone else that season, and his bat was hot for four months and cool for two. His speed and defense were so stellar he ranked as the fourth-best player in a weak NL. He wasn’t even a regular before that season, and he hasn’t been one since.
Justin Upton, Buster Posey, Jason Heyward, and Paul Goldschmidt are all quite young, and I think most of us would expect all of them to show up again. Matt Carpenter is more of a surprise, and an iffier bet to repeat, because his line drives came at Votto-esque levels. Could we really have two Vottos at once? What makes us so special?
Yadier Molina needed his best offensive season to climb over 6 WAR in 2012. (That’s the only time he’s hit 20+ dingers.) He will need one or two more seasons like that, or he will be remembered as merely the best defensive backstop of his time. I’m not sure if the defense alone will get him in. By the time he’s on the ballot, we might have new ways of quantifying catchers’ defensive contributions. They could ultimately temper or reinforce our belief in his defense
I think our last three guys–Chase Headley, Michael Bourn and Carlos Gomez–are not likely to return to this level of performance. All three could fit in the category “Guys who had great defense at important positions and who, in their prime years, put together a truly worthy offensive season to go alongside that defense.” Gomez is young, though, and exciting. Maybe the only reason he fit into the category is because I tried so hard to make him fit.
That leaves us with: Albert Pujols, Andruw Jones, Chase Utley, Carlos Beltran, David Wright, Matt Holliday, Chipper Jones, Hanley Ramirez, Ryan Zimmerman, Joey Votto, Ryan Braun and Andrew McCutchen as the cream of the NL from 2005 to 2013. When you think of the best players of this era, your thoughts should start with these guys. Some of them made their name before, some of them still have their best to come, but during these nine years there was no one better. Except maybe guys in the American League.
Recently on the subreddit for fantasy baseball, there was a thread about which stats you look at when trying to gain an edge over your league. A lot of the responses concerned pitchers, those most volatile of commodities and therefore the players who merit the most waiver-wire attention. The metrics most people cited as useful for pitchers were predictably strikeout-based: K/9, K%, K/BB.
Those stats are fine and useful in all sorts of situations, but I’m not convinced they really provide an edge. After all, if we suppose the people who post on the fantasy baseball subreddit are representative of serious fantasy players, it follows that the only edge K/9, K% and K/BB would provide is over casual players, the guys or gals who are there just to give the league enough members.
We must necessarily look deeper to find an edge over other serious players. Swinging-strike percentage (SwStr%) is deep! Instead of measuring strikeouts as a function of innings or plate appearances, it measures swinging strikes out of total swings. You can also call it Whiff% (Brooks Baseball does).
In the way of finding proof that SwStr% is great, I looked at all starting pitchers from the last decade (2004-2013) who had at least 50 IP in a season. That gave me like 1700 player-seasons, or near enough as makes no matter. TruStats from that badass data set after the musical break.
Fact: I am trying to rewrite that song for Jeremy Giambi.
First look how well SwStr% correlates with K%:
I wouldn’t have shown it to you if the correlation weren’t high, which .65 is. I would have instead felt ashamed that my hypothesis was wrong and kept the data forever secret. Then I would self-medicate. Smut peddlers of the world would rejoice.
Instead, intuition is borne out in the numbers. Whiffs beget strikeouts, perhaps more so than called strikes do. Now let’s see the difference between a whiff machine, an above-average whiff guy, etc. The following numerical endpoints are arbitrary, because whole numbers have that ancient allure which standard deviations lack. I like feeling connected to ancients. I’ll be an ancient someday and so will you.
Pitchers with higher SwStr% pitch more innings with both lower ERA and FIP. Naturally, their K% exceeds their peers’ and their WAR is higher, partly because FanGraphs relies on FIP to calculate WAR. WAR/180 is just WAR prorated to 180 innings, a reasonable standard for the modern starter. Hits are unaffected, score another for DIPS theory. Walks are also largely constant; it seems that getting whiffs and avoiding walks are two distinct skills.
Finally (for today), here’s one big graphic showing the year-to-year correlation for four related stats. This dataset was a little smaller than the previous one of ~1700, because not every pitcher pitched two consecutive seasons of at least 50 innings. Here I have 1182 such pairs of player-seasons. You’ll definitely want to click to engorge.
I was mildly surprised to find K/9 had the most consistency between years. I would have guessed SwStr%, then K%. Anyway, SwStr% is right up there with the two foremost strikeout stats in terms of yearly predictability. But say you need a stat in the middle of a season to predict the next few months’ worth of games. I still think SwStr% would be the best there, because it has more data points. It’s a matter of pitches versus entire at-bats, and tomorrow I’ll see if this hunch is true.
The previous post introduced some research we undertook at the behest of my fantasy star Paul Goldschmidt. In short, I kept Goldschmidt from 2012 to 2013 on the hunch that his doubles and home runs (43 and 20, respectively, in 2012) would even out some, toward a ratio closer to 1-to-1 than 2-to-1. (Huzzah, he hit 36 doubles and 36 home runs in 2013.) So I decided to follow that up with a 10-year study of whether doubles (rather, doubles plus home runs, or all extra-base hits) offered more predictive power than home runs. It was for my own fantasy purposes that I began this study, but it ballooned into something enormous thanks to my inability to accept that my hypothesis was wrong. The number of avenues I investigated was excessive, but some of them proved interesting, so I’ve started sharing. Thus ends the recap.
My hypothesis wasn’t wrong per se, just poorly worded. I thought at first that doubles and home runs (2B+HR on most of my graphs) would prove more stable over time than plain home runs, feeling that players generally displayed a consistent amount of power and that it was mostly luck who determined whether that long fly ball went over the fence or bounced just short of it. That was proven wrong; home runs are more consistent year-to-year than 2B+HR, no matter how you slice them.
If I had reworded my original hypothesis to account for aging, I would have gotten a satisfactory answer sooner. As a player enters his prime, and his power develops, home runs should eat a bigger share of his total extra-base hit pie. Here’s one visualization of that idea.
You can see the trends, but on this graph they look minor. Triples decline pretty much from the start, first counting for 10 percent of all XBH and declining with age to about five. Players get slow; that makes sense. As the triples vanish, home runs seem to replace them. However, the gains there are minimal, except at age 39 (see caption above). Doubles maintain their share of about 60 percent of all XBH.
But you are tired of hearing about dead ends. Did I find anything useful?
Doubles per home run. That’s what it came down to at the start. Goldschmidt had a more than two doubles for each home run in 2012, far too many for a player of his talent, at his age. This final graph delineates three distinct plateaus throughout a young player’s development. From ages 20 to 24, the ratio holds steady around 1.9, then it dips to around 1.7 for ages 25 to 28, then dips again to 1.6 for ages 29 and 30. So you may expect the ratio of doubles to homers to even out with age, but not as much as I originally hoped; the average ratio never dips below 1.5-to-1, except for the Bonds-skewed 39 year olds. Of course, these figures could be refined by categorizing batters by type and measuring the ratio for each type. We won’t shut the door on that possibility.
Here’s an appendix of interesting graphs that didn’t fit into the discussion above.
The mean age in this sample was 28.3 years old. The graph shows that the mode for ages is 26. The difference is explained above, in the gradual downward slope. Veterans stick around in baseball. Speed isn’t as important as in other sports, home run power lasts through the early and mid 30s, their production is generally more reliable thanks to all the data of their past seasons, and don’t forget that this window includes the PED era. PED use grinds normal aging to a halt, and sometimes can turn the process around.
Baseball is a kind climate for veterans above 30, but younger players are more profitable in almost every sense. The line representing WAR per 600 PA shows how much more valuable youngsters are than veterans, on average. Younger players are not only more productive, but cheaper, under team control, steadily improving, more handsome and likely less prone to injury. If there was enough young talent to stock major league rosters, we would already have seen that happen. Alas, the dozen or so minor and international leagues that feed into the majors are playing so far below big-league level that Juan Uribe just got two years, $15 million from the Dodgers.
This graph was from early in our process, which is why it only covers the last five years. Still, it demonstrates that the few players who are talented enough to break into the bigs at 20 and 21 are also talented enough to start the majority of games. The more traditional path to regular playing time starts at age 22 on this graph. Thirty-year-olds have the highest average games played, yet their figure is still less than 75 percent of all games in a season, underscoring the ever-present need for positional depth.
Hope you enjoyed this. These kinds of data-driven posts will be occasional features here at Midnight Baseball. We’d love any feedback and quality control you have to offer.
Many of you have started reading this week. Thanks!
I like to construct narratives around my fantasy team, which currently operates under the name Goldschmidt’s Gold Shit. This was the dumb logo.
Behemoth slugger Paul Goldschmidt didn’t ascend to team leadership until early August, when he kicked Josh Reddick (and his Red Dick [SFW]) out of the clubhouse in a verbal altercation, his booming baritone resonating throughout the bowels of the stadium and every fan in it. You see, Reddick had been spending too much time at nighttime clubs, wielding his favorite toy and namesake, leaving himself depleted come gametime. Things all came to a head when Marco Scutaro screamed at Reddick in frustration, expressing incidentally some long-held and deep-seated ethno-linguistic tensions felt by pretty much everyone on the team, normally neglected in the daily performance of badass manhood. In other words, Scutaro was fucking fed up with the nickname “Scooter.” Long story short, Jose Iglesias was pushed into a table, paralyzed gruesomely, and released within minutes, while Scutaro turned his back on the team to follow the long cold path of revenge. In the end, it was Goldschmidt who stepped up to refocus the team, leading a bold charge up the rotisserie points standings all the way up to…third.
Anyway, I acquired Goldschmidt in late 2012 with a specific hunch suggesting a breakout for 2013. The hunch was that some of his doubles would turn into home runs, since I’ve heard a lot that power develops later than other skills. And Goldschmidt’s 2-to-1 ratio of doubles to home runs (43 to 20, to be exact) seemed abnormally high for a slugger of his caliber. So I figured he’d get two bumps in homers, one from growing into power and the other from a corresponding decrease in doubles. Even supposing Goldschmidt couldn’t age (as yet unconfirmed), I still would expect the doubles and home runs to average out.
Before 2012 it was reasonable to consider Kansas City DH Billy Butler a robot designed by some bored and impractical scientist to hit doubles with maximum efficiency in a believably human range. It was, I swear. Then he broke out for 29 dingers with only 32 doubles, evening out the ratio when previously it had been 2-to-1. The example of Butler, combined with the fulfilled promise of Goldschmidt this year, spurred me to study the league at large for a trend between doubles, home runs, and how power ages.
That study ballooned and ballooned as I struggled for find evidence for the hypothesis that home runs become a bigger share of extra base hits as hitters age. I didn’t always phrase the hypothesis that way, which was part of the problem. The answer of course was staring at me all along, but I’m glad I took myself on a wild goose chase, because I think I learned a lot of interesting things. See for yourself.
First I thought I should check the most consistent measures of power before getting caught up in the doubles search. So I got caught up in examining the year-to-year correlation of stats: HR, HR/PA, HR/Contact, HR/Air. “Contact” is shorthand for all plate appearances ending in contact. To get it I subtracted strikeouts, walks and HBPs from total plate appearances. “Air” is shorthand for all airborne batted balls. I used Fangraphs’ data to get fly balls and line drives. Might as well say all the data came from Fangraphs.
The pool of players I used started big and got bigger. I began by collecting all player-seasons from 2009 to 2013 with at least 100 PA, then filtered the pool so that only players who had 100 PA in consecutive seasons would be counted. Then I made a lot of what are in retrospect superfluous graphs that look like this example. I wasn’t satisfied, so I started over, doing the same thing for 2004-2013. I recorded all these R-squared figures to put into a summary bar graph.
It looks nice, I’ll give it that. The dark bars signify the 2004-2013 window, and the half-transparent bars 2009-2013. Over the long run, no combination of home runs plus other extra-base hits offered more predictive power than home runs alone. Isolated Slugging Percentage (ISO) was the next-best thing after home runs; though it comes just behind 2B+HR/Contact, a stat with a shamefully long name, ISO wins given how readily it can be found on the internet.
Within the subset of home run stats, HR/Contact was the most consistent year-to-year. My guess as to why would be that it isolates home runs from outcomes based on other batting skills. Walks and strikeouts are heavily influenced by a batter’s plate discipline, which is a separate skill (some say the sixth baseball tool) and itself an ever-evolving attribute. Thus HR/PA can be influenced by a sharp spike or drop of restraint at the plate in a way that HR/Contact is not. Whatever the underlying reason, HR/Contact ought to be of some use for projections. I for one will be using it for fantasy purposes.
I imagine the propensity to hit fly balls or ground balls stems from a batter’s swing mechanics, which are etched into a player’s muscle memory long before the major leagues. Such differences distinguish batters into types: the slap hitter, the slugger, the veteran bat-control guy, etc. Performance may wax and wane, but those kind of batting identities are ingrained. Plus, batted balls come in hundreds while home runs and doubles in tens; there’s simply more data. All these, reasons why Air/PA and Air/Contact are more consistent than the rest of the stats in that graph.
There are gains in the rates of walks and home runs, only they are tough to discern at this scale. I have other graphs where the differences are obvious, and I’ll make them public with all these graphs, even the ones that make no sense now that I think about it, on my Google Drive later this week.
Anyway, the decline in strikeouts is obvious, and lasts throughout a player’s thirties. Walks rise less dramatically, but never stop rising–at least until 40. Overall, strikeouts are reduced more than walks are increased, resulting in more plate appearances that end in contact as a player ages.
Moreover, there are basically no gains in HR/Contact with age. Mere tenths of a percent. Perhaps you wish to identify a plateau from ages 25 to 28 that is distinctly higher than a plateau from 21 to 24. The spike around thirty probably isn’t happenstance, given that my pool was 4394 player-seasons large. The spike at 39 is Barry Bonds again.
So far, I hadn’t found anything that convinced me of my theory behind home run surges. Could it really be the accumulation of all these little marginal differences? Is this the whole story: slighty fewer strikeouts, plus a few more walks, plus more plate appearances on average, plus small percentage points gained in home run rate, equals more home runs? It’s possible but I might have missed something, I thought. I couldn’t get the question off my mind, all thanks to Goldschmidt and Billy Butler. Tomorrow I’ll go into some of my later approaches.