At this point, I’m getting close to about a half-decade of these posts, so most of the conclusions should be old hat at this point. In short, hey, the projections are generally pretty good. Maybe they won’t get every player exactly right, but mix in the underestimates and overestimates, and they’ll be pretty close, on average. Sounds good and boring, right?
Well, truth be told — while that’s generally true, that was absolutely not true for the 2018 Atlanta Braves. You probably already instinctively realize this, because you saw it. The Braves were not projected to be good. The Braves ended up being good. That doesn’t happen without a lot more overperformance of projects than underperformance of projections.
One thing that heralded the Braves having “arrived” into a land of ambitious, meaningful baseball is that their lineup card stopped being a carousel featuring Adonis Garcia and Anthony Recker in lieu of horses. Rather, the Braves really only used 11 position players in 2018, and this undectet (yes, this is a real word for “group of 11”) was good for 93 of the team’s total (non-pitcher) plate appearances. Aside from Ryan Flaherty, whose 183 PAs unmercifully constituted over three percent of his team’s total plate appearances on the year, the remaining 12 players to grab a PA in a Braves uniform this year combined for just four percent of the team’s offensive playing time. I bring all of this up just to segue into the following. Across these 11 players, Steamer did not overestimate a single one. Across these 11 players ZiPS overestimated only Preston Tucker, and only by 0.1 WAR. Across these 11 players, IWAG actually overestimated four, but none by more than 0.7 WAR. “What happened with the 2018 Braves” is a multifarious question that can have many satisfying answers, but one of those answers is essentially: “They outplayed their projections.”
This post is going to have two separate sections. In one, I’m going to quickly go over the expectations for each of the 11 players and talk about where they ended up. In the other, I’ll do the same thing I did in prior years, which is briefly compare the three projection systems against one another. For reference, if you’re stumbling across this post without seeing things from past years or months, you may want to review the following:
- Last year’s projection retrospective for position players; and
- The pre-2018 projections for the position players.
The big table of preseason projections and actual results is below. Under that is a different table, that tells you the difference between the projections and outcomes. Note that in this second table, the values are dragged across how many PAs the player actually got. So, for example, Nick Markakis was projected for 0.3 WAR/600 by Steamer, and -0.1 by ZiPS. Instead, he put up 2.2 WAR/600, but it was over 705 PAs, or 2.6 WAR total. Hence, Steamer missed not by 1.9 WAR (because that’s WAR/600), but by 2.2 WAR (because that takes the 0.3 WAR, spreads it over 705 PAs for 0.4, and then subtracts that from his actual 2.6 WAR figure. Rio Ruiz and Lane Adams are not included in the second table; they got so few PAs that including them in a comparison would be kind of pointless and bias the results where averages/etc. are concerned.
A brief player-by-player review of the projections
Tyler Flowers was projected to be either somewhat above average on a rate basis (Steamer, IWAG) or somewhat below average (ZiPS). He only managed to play half a season, but managed to indeed be pretty much “somewhat above average,” finishing at 2.4 fWAR/600. His 95 wRC+, the product of some unfortunate xwOBA underperformance that befell him, was lower than projected by Steamer and IWAG, but still higher than projected by ZiPS (87). Flowers was one of the few players not really routinely underestimated in this review. Steamer ended up on the money, and IWAG overestimated him by a fraction.
Relative to the IWAG distributions, Flowers ended up more or less in the midpoint of his projections. His 1.2 fWAR was around the 43rd percentile; his 2.4 fWAR/600 was around the 48th percentile.
The other piece of the Braves’ catching tandem had some variety in some preseason projections. IWAG expected less of a decline from his remarkable 2017 with a 100 wRC+ and 2.5 WAR/600; Steamer and ZiPS were far more pessimistic with a below-average rate-based performance and wRC+s of 90 and 82, respectively. Instead, Suzuki once again defied these expectations: 108 wRC+, 2 fWAR in 388 PAs, and a shiny 3.1 fWAR/600 mark. As a result, Steamer and ZiPS underestimated his eventual production by a win or more across those 388 PAs.
Relative to the IWAG distributions, Suzuki’s outcomes were somewhat in excess of the projection midpoint. His 2.0 fWAR was around the 66th percentile; his 3.1 fWAR/600 was around the 72nd percentile. As part of a somewhat interesting pattern, Suzuki’s outcomes ended up being in the middle of the bimodal distribution that IWAG ended up drawing for him. Specifically, IWAG expected his fWAR/600 to either have him be well below average (i.e., somewhere around 1.0 fWAR/600) or crazily above (somewhere around 4.0-5.0 fWAR/600, albeit with a much lower probability). Instead, he settled in around the valley between these two modal “humps,” something that made a lot of sense on an average basis but not in terms of its specific likelihood.
The projections for Albies were one of the most fun parts of the leadup to the season, though superseded by the 21-year-old’s rollicking April. Steamer saw him as league-average thanks to a 95 wRC+ and above-average defense at the keystone; ZiPS had him hitting worse (89 wRC+) but with stellar defense en route to 2.3 fWAR/600; IWAG nudged the batting line up to a 101 wRC+, which combined with above-average-but-not-like-ZiPS-projected defense for 2.8 fWAR/600.
Albies ended up with a weird season, as he fell apart in the season’s final months after a bombastic (and bombtastic) April. His final line was a dead-even 100 wRC+, but that came with an awful 67 mark after the All-Star Break. Combined with standout defense more in line with the ZiPS projection and another half a win thanks to baserunning, and he ended his 2018 campaign with nearly a four-win season and 3.3 fWAR/600. Steamer and ZiPS ended up underestimating his production by over a win; IWAG by over half a win.
The IWAG distributions for Albies were relatively non-bimodal (the second modal hump was sub-2.0 fWAR) and mostly saw something normal distribution-y centered around 2.8 fWAR/600. His outcomes were beyond this, finishing around the 70th and 80th percentile projections for his fWAR and fWAR/600, respectively. (Albies getting as many PAs as he did was fairly unexpected.)
After a very odd and successful-but-not-sustainable 2017 season, the projections for Johan Camargo were also pretty varied. Steamer and IWAG saw him as a below-average bat with a mid-80s wRC+ and below-average production despite positive defense (1.2 fWAR/600 and 1.8 fWAR/600), respectively. ZiPS, meanwhile, saw little of use in him, tagging him with a 70 wRC+ and 0.3 fWAR/600 projection. Camargo ended up missing some time early in the year with injury, but finished dramatically above all of these, with a 3.3 fWAR campaign and 3.8 fWAR/600 on a rate basis. This was one of the cases where all three systems whiffed a ton, and one of the main reasons (along with the emergence of Ronald Acuña Jr. and the resurgence of Nick Markakis) that the Braves were as successful as they were in 2018. Among the 11 position players discussed here, the projection systems collectively whiffed on Camargo the most, especially on his bat. While Camargo did end up outperforming his xwOBA pretty heftily by the time the year ended (top 25 outperformance in baseball among the 350 players with 200+ PAs), even a season where he matched his .310 xwOBA in terms of results would have featured an underestimate by the projection systems: a .310 wOBA is a wRC+ around 90, rather than mid-80s. The xwOBA outperformance just made the projections look ridiculous, and the projection systems also underrated his defense to boot.
The IWAG distributions saw Camargo as either a mediocre bench player (0.5 fWAR/600) or a legitimately above-average player (3.0 fWAR/600), with relatively less likelihood of a middle-ground outcome. He did indeed make good on the latter in 2018, but as noted, that his success relied on heavy xwOBA overperformance is a storm cloud on the horizon. Still, with said overperformance, he ended up outside the “reasonable likely range” of the IWAG projections — which again, just circles back to the whole “the Braves outperformed projections by a lot!” deal mentioned in the introduction to this post.
Speaking of xwOBA overperformance... Charlie Culberson was expected to be pretty terrible, among the worst performers in MLB. All three projection systems featured a wRC+ of 60 or below and below-replacement-level performance, though ZiPS thought he’d be really good at fielding (13.2 Def/600!), while Steamer had him at over a win below replacement per 600 PAs. Of course, we all know how this story went: Culberson ended up posting a 108 wRC+ thanks to a historic wOBA-xwOBA gap, despite an actual xwOBA nearly in the bottom five percent of all hitters with 200+ PAs in 2018. Even so, his defense was not well-rated (-4.4 Def; a gross -8.2 Def/600) and he finished with just 1.0 fWAR and 1.9 fWAR/600. Of course, even those numbers meant his results were grossly underestimated by the projection systems, as ZiPS ended up being the closest to his actual value while being off by 1.3 wins.
Culberson didn’t quite exceed the upper bounds of his IWAG distributions (with little track record, the bounds get wider, though Camargo managed to exceed his bounds anyway), but still finished around the 94th and 98th percentiles for fWAR and fWAR/600, respectively.
Freddie Freeman was projected between a 139 and 148 wRC+, and between 4.0 and 5.0 fWAR/600 coming into the year. He finished a little below this range offensively (137 wRC+) and within it overall (4.4 fWAR/600), thanks to a strong defensive year. Steamer and ZiPS both ended up underestimating him by about half a win in the end, with overestimates offensively and underestimates defensively; IWAG ended up overestimating him by 0.7 wins (the biggest overestimate for any player and projection system in this review), with an even greater offensive overestimate but a smaller defensive underestimate.
Freeman had a boatload of PAs in 2018; given his previous injury issues, IWAG’s central estimate was only 566 PAs on the year. As a result, his actual fWAR total of 5.2 ended up around the 69th percentile of his IWAG distribution. However, his fWAR/600 rate of 4.4 was on the lower end, around the 23rd percentile.
If I had told you before the season that the Braves would make the playoffs, you would have probably figured that Freddie Freeman finding yet another gear and going completely berserk on the league would have had something to do with it. Instead, Freeman was one of the few places where the Braves really didn’t exceed expectations... and yet they enjoyed bewildering success anyway. Great times.
Swanson was one place where the projection systems all agreed, perhaps eerily so. They all gave him an identical 87 wRC+ with only slight variations in defense. His fWAR/600 across all three varied by just 0.2, falling between 1.6 and 1.8. In reality, Swanson hit worse than expected (80 wRC+) but had a better defensive season than projected, leading to 1.9 fWAR over 533 PAs, or 2.1 fWAR/600. As a result, all three projection systems undersold him a bit, but not by much.
IWAG’s distribution for Swanson was a very neat bimodal curve that pegged him as either a bench type around 0.5 fWAR, or a guy that made good on his promise and ascended to an above-average performance. In reality, he ended up in the middle — 57th percentile by WAR, 52nd by WAR/600.
Ronald Acuña Jr.
Like Camargo, Acuña taking the baseball world not by storm, but by typhoon-caused-by-emegence-of-Godzilla was another reason why the Braves ended up where they were. The projections for baseball’s top prospect weren’t gaudy at all: Steamer had a very pessimistic 1.3 fWAR/600, ZiPS had 2.2 fWAR/600, both had essentially league-average batting lines with ZiPS giving the kid a lot of defensive credit. IWAG was more aggressive, with a 112 wRC+ and 2.5 fWAR/600.
Of course, we all know that Acuña took a baseball bat to these projections, and everything else. He finished the year with a 143 wRC+ and 3.7 fWAR in 487 PAs, good for a 4.6 fWAR/600, nearly double the most optimistic of the three projections. Over those 487 PAs, Steamer missed by 2.6 wins, ZiPS by 1.9, and IWAG by 1.7. While all three systems underestimated his bat to an extreme degree, they overestimated where he’d end up defensively, which helped keep the forecasting error from getting too egregious.
Since Acuña missed time with injury, his 3.7 fWAR on the year was actually pretty middling as far as the IWAG distribution went — right around the 55th percentile. But the distributions didn’t expect him to put up that much value in so few PAs, and his actual fWAR/600 fell outside of the distribution’s “likely” bounds (though only by a few runs). As noted for Camargo, Acuña’s dominant rookie performance is one of the biggest reasons why the Braves did what they did in 2018.
I haven’t looked back at Steamer historically, so I’m not certain that this has been the case every year, but that particular projection system seems to dance an interesting tango with Inciarte. Every year, it says, “This is the year you stop defying your peripherals and end up a below-average hitter!” And every year, Inciarte obliged by defying his peripherals and ending up with an average batting line en route to an above-average season. Well, every year until 2018. Both Steamer and ZiPS projected Inciarte for a wRC+ in the low 90s, giving him a low-to-mid-2.0s fWAR/600. IWAG did not do so, figuring another league-average wRC+ and 3.1 fWAR/600. In the past, this worked out somewhat better for IWAG (in 2017, he suffered a weirdly low defensive value, but offensively he was average again), but this season, Inciarte’s ability to outperform his quality of contact evaporated. As a result, he ended up with a 90 wRC+, and though his defense rebounded, he managed a fWAR/600 of 2.6, somewhat below the 3.0 threshold. In the end, Steamer and ZiPS underestimated him but only by half a win or so; IWAG overestimated him by the same amount.
Given Inciarte’s relatively consistent performance, IWAG saw him as having a relatively uniform distribution of outcomes for value, i.e., a fairly equivalent chance of posting anywhere between 2.0 and 4.0 fWAR/600 on the year. But, for hitting results, it forecasted a pretty normal distribution between 95 and 103 wRC+, which made it more surprising when Inciarte finally failed to drag his batting line into that range. Overall, value-wise, it was a pretty disappointing result for the center fielder, as both his fWAR and fWAR/600 fell around the 20th to 25th percentile of the forecasted distribution. Will he go back to outperforming his xwOBA by an appreciable amount next year? Stay tuned.
Markakis is the third component of the triumvirate of “men wot exceeded expectations fortuitously” for the Braves’ 2018 position players. The projections for him were glum, to say the least. Aggressively dour, perhaps. Steamer saw below-average hitting, abysmal fielding, and 0.3 fWAR/600. ZiPS saw fire, brimstone, and misery in the form of replacement-level performance: an 86 wRC+ and -0.1 fWAR/600. IWAG looked positively enamored with him, forecasting a much more livable 98 wRC+ and 1.0 fWAR with average corner outfield defense. Of course, you know what happened: Markakis went Super Saiyan/1980s sports film training montage for the first four months of the season before cratering to earth in August and September, finishing with his highest fWAR total (2.6) since his crazy career year back in 2008. Even on a rate basis, his 2.2 fWAR/600 was his best since 2012. The end result was that all three systems underestimated Markakis by a lot — 2.2 fWAR, 2.7 fWAR, and 1.4 fWAR across Steamer, ZiPS, and IWAG, respectively.
The IWAG distribution for Markakis was essentially normal and narrowly constrained around a range of 0.5 to 1.5 fWAR/600 or so. His fWAR of 2.6 and fWAR/600 of 2.2, placed him around the 98th and 95th percentiles of said distribution — not quite out of bounds, but pretty close.
The projections were fairly confident that Preston Tucker couldn’t field, nor could he hit enough to make up for his fielding deficiencies. His projected wRC+s ranged from 86 (Steamer) to 95 (IWAG), his projected Def/600 ranged from -15.1 (IWAG) to -13.6 (Steamer), and IWAG was the most optimistic overall at a measly 0.1 fWAR/600. In reality, the most pessimistic projection was the closest, as Tucker managed only an 88 wRC+ and -15.3 Def/600, en route to -0.7 fWAR/600. He ended up being the only player of the 11 discussed here that ZiPS overestimated (however slightly), and one of only two that Steamer did not underestimate. So, it wasn’t all positive for the Braves, but the reality is that Tucker only got 142 PAs in a Braves uniform, so he wasn’t exactly counterbalancing anyone else’s projection.
Tucker’s performance was fairly poor relative to his IWAG projected distribution of outcomes, too. IWAG saw a path for him to be an okay bench piece if he could hit around a 110 wRC+. Instead, his -0.2 fWAR was around the 25th percentile outcome, and his -0.7 fWAR/600 was near the third percentile.
“Bonus” — Ryan Flaherty
I didn’t look up the Steamer and ZiPS projections for Ryan Flaherty, but IWAG projected him for a 60 wRC+ and -0.1 fWAR/600. Instead, he ended up with a 57 wRC+ and -1.3 fWAR/600. Like Tucker, however, he didn’t really get enough PAs to make his terrible performance counterbalance all the good fortune that the Braves got relative to expectations.
Project system comparison
At this point, and especially in a season where the big story was systematic underestimation, comparing the three systems feels somewhat beside the point. But, for the sake of interest...
The sum total differences in position player WAR across the PAs actually accumulated by these 11 players were:
- For Steamer, an underestimate of 12.5 WAR;
- For ZiPS, an underestimate of 13.0 WAR;
- For IWAG, an underestimate of 6.0 WAR.
In a nutshell, this is a huge part of how the Braves ended up where they did. When you beat your projections by that many wins, even if only for position players things have a good chance of going great for you.
Specifically for batting runs:
- Steamer, ZiPS, and IWAG averaged an underestimate of 7.7 runs, 10.4 runs, and 4.1 runs, respectively.
- On an absolute value basis, the average error was 9.7, 12.5, and 9.3 runs, respectively.
- Root mean square error: 11.9, 14.5, and 10.9, respectively.
Specifically for Def:
- Steamer, ZiPS, and IWAG averaged an underestimate of 1.9 runs, an overestimate of 0.6 runs, and an underestimate of 0.8 runs, respectively.
- On an absolute value basis, the average error was 3.0 runs, 4.3 runs, and 2.8 runs, respectively.
- Root mean square error: 3.4 runs, 5.3 runs, 3.0 runs, respectively.
- Steamer, ZiPS, and IWAG averaged an underestimate of 1.1 wins, 1.2 wins, and 0.5 wins. Note that, for example, last year, the average error was 0.2 or 0.3 wins, so this is just another way of expressing the systematic underestimation that occurred this year.
- On an absolute value basis, the average error was 1.1, 1.2, and 0.8 wins, respectively.
- Root mean square error: 1.3 wins, 1.4 wins, 0.9 wins, respectively.
In terms of closest projections, they broke down like this:
- Batting runs: Steamer = five players; ZiPS = two players; IWAG = seven players
- Def: Steamer = five players; ZiPS = one player; IWAG = five players
- WAR: Steamer = three players; ZiPS = two players; IWAG = six players
The best projections were Steamer for Flowers and Tucker. The worst projection was ZiPS for Johan Camargo. The player most easily pegged was Preston Tucker (he was who we thought he was), but Freddie Freeman also ended up most resembling the average of his projections. As already noted, Johan Camargo ended up being the prime projection-buster of this group.
No charts are included because they’d look pretty similar, but can be produced on demand. Stay tuned for the same analysis for pitchers, coming eventually.