So we’re agreed: it’s interesting to see where different projection systems disagree. After all, the projections are based on the same information, for the most part, so it could be telling where there might be significantly different interpretations. Monday, I compared and contrasted 2014 Steamer and 2014 ZiPS for position players. This is the natural and obvious follow-up, for pitchers.
You’d think this would be just as simple as the position-player version. In truth, it’s more complicated, and analysis required a few more judgment calls. I’m okay with them — I’m the one who made them — but if you’re dissatisfied, you’re free to go into the spreadsheets and run the numbers the way you’d prefer to. One of the neat things about our hosting the projection data is that it’s all so easily exportable. Now then, let’s walk through the process so we can get to the end of walking through the process.
As was the case with the position players, I wanted first to narrow down the player pool to eliminate guys who aren’t going to see very many innings. Steamer projects many thousands of pitchers. ZiPS projects too many hundreds. That’s wonderful for the systems, in that they’re very thorough, but for me, all that extra stuff is irrelevant. The first thing I did was eliminate everyone who isn’t projected to throw at least 50 innings by both systems.
Then I had to make a decision: how much do I care about relievers? Do I care enough to offset some of the problems when you encounter a guy projected to both start and relieve? Aren’t we all ultimately more interested in guys likely to start? I decided to narrow some more, eliminating everyone who isn’t projected to start at least ten games by both systems. This left me with a sample of 169 starting pitchers, and there was just one hard decision left to make.
I didn’t want to look at projected WAR over an innings denominator. I wanted to compare a simpler, more natural rate stat, so I had to think, projected ERA, or projected FIP? There’s a good argument to be made for ERA, since that’s a measure of runs allowed and since projections shouldn’t include good or bad luck. But in the end I had to go with FIP, because what we’re really curious about are projected differences in walks, strikeouts, or dingers, and ERA can muddle everything up. We don’t have a real great understanding of beating or undershooting peripherals, so for these purposes that whole area should just be avoided entirely.
What’s following is stuff I was left with. You’re going to see tables with projected FIPs. Look, right now, here’s the first one. Here are the ten pitchers Steamer likes more than ZiPS the most, according to the criteria and filters above:
Just missing: Jake Arrieta, then Jerome Williams, then Carlos Carrasco, then Hyun-jin Ryu. The biggest differences are in the dinger column, because of course it’s the dinger column that’s going to make the biggest difference. This will be a theme.
This table by no means features an assortment of stars. But then, it’s the stars who tend to be pretty well understood. Other guys can be more mysterious, and here, you see Hernandez and Alvarez with projected FIPs that differ by more than a full run. How does that happen? It’s pretty easy to explain.
Steamer projects Hernandez for an FIP in line with last year’s xFIP. ZiPS figures he’ll keep getting blown up by home runs. There’s very little difference in other categories. Alvarez, also, gets a Steamer FIP that’s a lot like last year’s xFIP. ZiPS figures he’ll strike out fewer, and walk more, and allow a ton of home runs, which is something he did last year in the majors, but which is something he didn’t do last year in the minors. Steamer seems to want to regress homer rates very heavily. ZiPS would appear more reluctant.
Not that these are all dinger cases, but there are several of them, and there will be several more to follow in the opposite direction. ZiPS thinks Bauer’s going to keep walking the world. With Deduno, both systems expect some walk-rate regression, but Steamer likes his 2013 dinger rate while ZiPS keeps in mind his 2012-2013 dinger rate. Dingers are the story with Delgado as well, with ZiPS unable to ignore last year’s 24 homers in 20 appearances.
Keep going. ZiPS sees Oberholtzer getting destroyed by the dinger, as he’s a fly-ball pitcher in a hitter-friendly environment. Steamer and ZiPS have the obvious disagreement over Blanton’s homer-proneness. Same deal with Santana, as the systems project nearly identical strikeout and walk rates. I feel like maybe I don’t need to keep explaining this. ZiPS thinks Burnett will get back into some home-run trouble. Finally, with Roark, there’s something more interesting — ZiPS projects a far lower strikeout rate. It could have something to do with the fact that, last year, in the majors, Roark allowed a contact rate of 86%. Steamer thinks he’ll be able to keep fooling batters, while ZiPS foresees regression.
To the other extreme. Here are the ten pitchers ZiPS likes more than Steamer the most:
Let’s knock out the unusual one first. Tanaka, of course, has no major league or minor league track record. ZiPS does think he’ll allow fewer homers than Steamer, but it also projects a much higher strikeout rate. Steamer thinks Tanaka will strike batters out like 2013 Jose Quintana. ZiPS thinks he’ll strike batters out like 2013 Gio Gonzalez. ZiPS projects him at his 2012 Japanese level; Steamer projects him below his 2013 Japanese level.
Now, you should be able to walk yourself through a lot of this, based on what was discussed above. Weaver has demonstrated, over his career, that he can suppress home runs. Steamer sees him not doing that anymore, while ZiPS thinks it’ll keep up. ZiPS also sees higher strikeouts, while Steamer sees a declining fastball and a pitcher who might be nearing the edge.
With McAllister, it’s about ZiPS seeing greater dinger suppression and fewer walks. The Verlander story is very similar to the Weaver story. Notably, however, ZiPS also projects Verlander’s highest strikeout rate since 2009, with Steamer lower by almost a strikeout an inning. With Marcum, Steamer seems afraid of his declining stuff and injuries. With Moore — you guessed it — Steamer regresses the homers up, while ZiPS keeps them down.
It’s different with Skaggs. Skaggs has a high big-league homer rate, but ZiPS projects something even lower than Steamer does. In this case, ZiPS seems to be recognizing what Skaggs did last year in Triple-A. With Teheran, ZiPS is forecasting considerably fewer walks. Kershaw’s case is a lot like the Weaver case again. Kershaw doesn’t give up homers. ZiPS gives him credit for that, looking forward. Steamer thinks the clock’ll strike midnight, and Kershaw will be left as something simply terrific. At last, there’s Jimenez. ZiPS is a big believer in his most recent strikeout boost. Steamer, less so. That difference is almost entirely wrapped up in the strikeouts, instead of in the homers. ZiPS sees 9.6 strikeouts per nine innings; Steamer sees 8.1. The walks and homers are almost identical.
The big and fairly obvious takeaway: Steamer is pretty aggressive with its homer-rate regression. Steamer, in other words, is a pretty big believer in xFIP, while ZiPS is more willing to give or subtract points based on homers actually allowed. Personally, I like the former approach, but only to a point, and going that way means you’ll miss out on guys like Kershaw and Weaver, who have long track records of being unusual. It’s also worth noting that ZiPS might be more willing to be aggressive with strikeout-rate projections. But that might not actually be a whole real trend.
The closing advice is the usual advice: generally best to blend projections and average them. With certain players, you might have reason to believe one system over another. On the team level, an average should work well enough. On the one hand, it can seem silly to regress pitcher homers so heavily. On the other hand, history and math. So. It’s complicated.
Print This Post