Cluster and performance luck 2015: An update

June was a long time ago. In June, the Rays were in first place in the AL East, the Twins were in playoff position, and the Mets were leading the Nationals in the NL East. A lot has changed since June. I wrote an article around then discussing the impact of luck on every team, breaking it down into the familiar cluster (or timing) luck and the less intuitive performance luck. That was two months and about 60 games ago, so I figured an update was in order.

As I mentioned in that article, cluster luck is well understood -- it's the difference in results due to timing rather than performance. If one team goes HR, single, single, K, K, K, and a second team goes K, K, single, single, HR, K, both teams look the same from a context-neutral perspective, but the results for the first are much worse than the results for the second. BaseRuns (developed by David Smyth, available at FanGraphs) calculates team records based on their context-neutral performance.

The second type is a little more nebulous, and here I'm just going to quote my original description, since I thought it was pretty good:

Some readers might balk at describing [this] as "luck" at all. It's when a player performs much better than he is expected to going forward, such as when a 30-year-old with a career wRC+ of 90 hits for a month or two at a wRC+ of 120 ... "Luck" is maybe not an ideal name for this, since under- or over-performance like this happens for a reason, and not knowing that reason is different from that reason not existing. But these performances are, by definition, unforeseeable with our current tools, and so effectively random from an analytical perspective, which is why I'm alright describing them as luck.

Has Carlos Correa's blistering performance as a 20-year-old been "lucky"? No, not in the sense that he somehow doesn't deserve it. Have the Astros been lucky to have it? Absolutely! I'm coming at this from more of an organizational perspective, and so I think categorizing both of these as "luck" is fine.

In that first article, I used the difference in BaseRuns predicted win% and actual win% for cluster luck, and the gap between preseason projected win% and actual win% for performance luck. Sabermetrician and author Mitchel Lichtman made the following comment:

It's a little more work but surely you want to use individual player projections prorated to their actual playing time to determine expected performance and not just pre-season team projections. There are too many injuries and playing time differences for that to be used in an analysis like this.

At the time, I actually stood by the preseason method -- as MGL alludes to, it captures differences in playing time due to injury, which seemed to me a valuable thing to include. Now that the trading deadline has passed, however, many teams look substantially different from their early iterations, and I think it does make more sense to use individual projections. For each player, I used their projected WAR from preseason ZiPS prorated to their number of plate appearances. For players that didn't have a projection, I assumed they were projected at their actual performance from this year, which is an arbitrary decision but one that won't matter much, since those players tend not to get much playing time. All this is through 8/29.

Clustering

Like I said, I think cluster luck is fairly well-understood, so it's likely none of this is super surprising. In order to have the best record, a team has to get at least somewhat lucky, so seeing the Cardinals at number three and the Royals at number one, beneficiaries of about 9 and 11 wins respectively due to sequencing, makes a lot of sense. On the other hand, the Twins and Rangers are fighting for the second Wild Card spot, but by BaseRuns they're the worst and fourth-worst teams in the AL.

The Athletics' underperformance is just as well-known as the Twins overperformance, and while their gap in win% and BaseRuns win% has shrunk from .163 in June to .101 now, they still "lead" the league and have played as the fourth-best team in the AL from a context-neutral perspective. Several teams are in very good position entering September and have actually underperformed by sequencing: the Dodgers, Blue Jays, and Astros, all by between five and six wins. That strong performance is part of the reason those three teams have the three highest World Series odds at FanGraphs, despite being fifth, sixth, and tenth by records (through 8/29).

Performance

This table probably needs a little more explanation. The "total WAR" column is exactly that, using FanGraphs figures, though there might be some small differences from the team totals available there due to rounding. The "projected WAR", as I discussed in the introduction, is the total projected WAR of each individual player on a given team prorated to their actual playing time for that team. The fourth column is the first minus the second, and the fifth column is a rough estimate of how that translates to winning percentage, to make it comparable with the prior table. Team WAR doesn't correlate perfectly with expected team performance, but it's pretty good -- the R-squared of that and BaseRuns win% is .9 -- so while it's not perfect, I think it's a fine corner to cut for the purposes of this article.

Again: please don't think I'm saying the Astros don't deserve to be where they are. But this season's performance has certainly been unexpected and fits this definition of "lucky". Carlos Correa has been the perhaps the most visible surprise for Houston (1.1 projected WAR, 2.8 actual), but that ranks "only" 24th among all batters in outperformance of projections, and the Astros' total batting WAR has actually been slightly lower than projected (18.9 projected, 15.4 actual).

Instead, it's been their pitching staff that has blown away expectations, led by Lance McCullers (-1.0 projected, 2.3 actual), Dallas Keuchel (2.5 projected, 5.4 actual), and Vincent Velasquez (-0.1 projected, 1.1 actual). Just as important, they've seen almost no underperformance -- the biggest negative surprise has been Jake Buchanan, owner of a whopping 9 innings, projected at 0.0 WAR and actually giving -0.1. Of the 23 pitchers throwing any number of innings for the Astros, only three have underperformed preseason projections. If there's any one reason Houston's rebuilding timeline has seemingly leapt a year ahead of schedule, it's been the surprisingly excellent performance of their pitching staff.

The Blue Jays offer a much more balanced picture, with their fair share of unexpectedly good and bad performances on the hitting and pitching side but still a net positive of more than 7 WAR. Josh Donaldson has led the pack with his MVP-caliber performance (4.8 projected, 7.3 actual (!!!)), but Devon Travis, Kevin Pillar, and Ryan Goins have also beaten their projections by at least 1 WAR. Jose Bautista is putting in a solid season, though not quite as good as expected (4.6 projected, 3.3 actual), but he's the only Blue Jay batter more than a win under his projection. The pitching side looks similar: very minor underperformance, with the biggest negative gap coming from R.A. Dickey (1.4 projected, 1.1 actual) and substantial upside, with Liam Hendriks, Marco Estrada, and Drew Hutchison all delivering close to one WAR more than expected.

The negative side of the spectrum is a little bleaker, so I won't spend as much time on it, but it's worth a reminder that this is based on playing time and so doesn't include the impact of injuries. As a result, the Rangers look slightly lucky, despite running into way more than their fair share of injuries. Some of the biggest disappointments of the year occupy the bottom of the list, as expected: the Red Sox and Angels have both underperformed to a large extent, and the former has seen their playoff odds evaporate completely while the latter is clinging to their dwindling remains.

The White Sox, however, blow both of them away, with their gap of 10.5 wins the largest of either the positive or negative teams. Crazily, that's with a better-than-expected pitching staff -- only Daniel Webb has underperformed, and even then by a rounding error away from 0, leading to an overall positive difference of 3.6 WAR.

But the batters; oh lordy, the batters. Geovany Soto has been better than expected, by 1.3 WAR, and Trayce Thompson has also been a pleasant surprise to the tune of about 1 win. Beyond that, it gets real ugly. Jose Abreu doesn't have much a history to build a projection on, but his 2.9 WAR has been almost a win less than expected. There are six -- six! -- players who have underperformed more than that: Tyler Flowers (1.3 projected, -0.1 actual), Conor Gillaspie (0.5 projected, -1.0 actual), Avisail Garcia (0.8 projected, -0.8 actual), Melky Cabrera (2.0 projected, -0.1 actual), Alexei Ramirez (1.8 projected, -0.6 actual), and Adam LaRoche (1.8 projected, -0.8 actual). Much has been made of the White Sox position players' historic levels of awfulness, but this suggests that it might not be Rick Hahn's fault. You might not have been a fan of his decision to push the chips in this year, but by projected WAR, they rank 12th, or right on the bubble for a Wild Card slot, compared to 23rd by performance.

Finally, a chart to summarize, with over- or under-performance of context-neutral measures (i.e., BaseRuns) on the horizontal axis and over- or under-performance of projections on the vertical. Teams above or to the right of the line that runs through the origin have been overall lucky, when combining both these measures; teams below or to the left, unlucky.

This chart is useful for a couple reasons. First, it gives a combined view of both these metrics, making their interplay clear. Congratulations, Royals fans; my condolences, A's fans. It also helps put the impact of the two types in context. The difference in BaseRuns performance and actual win/loss is spread over a much wider range than the difference in projected and actual performance. You might not be a fan of individual projections, but on the whole, they do a very good job predicting team performance. Sequencing, however, is basically impossible to predict, and as this chart shows, it has a much larger impact on team performance than projection error (in general).

Comparing this to the prior edition's chart also brings up a few interesting points. First, as expected, the percentage gaps are smaller, as teams regress toward the mean; both types of gaps in this edition are basically bounded by -.100 or .100 in winning percentage, compared to -.150 and .150 in June. At that time, I noted "something of a negative correlation" with BaseRuns overperformance and projections overperformance, to the tune of an R-squared of about .15. I couldn't think of any reason the first would impact the second, or vice versa, so I basically dismissed it, and in this iteration, the correlation has almost entirely disappeared, with the R-squared falling to .02.

As I said in June, lumping these together as "luck" isn't ideal, and hopefully some nuance begins to work its way into baseball's vocabulary to distinguish between the two.

. . .

Henry Druschel is a Contributor at Beyond the Box Score. You can follow him on Twitter at @henrydruschel.