Actually, Matt Pouliot wrote that, but it's not any less scary.

Why is that list scary? Similarity scores have zero predictive value.

Here's a less scary list: right-fielders centered on Ethier in terms of OPS+ and career PAs through age 29, ranked by WAR Fielding Runs:



A bunch of good hitters, not all of whom got the earliest start on ML careers. There are some pretty durable guys on the list (Justice, Griffey Sr), and even the notably rickety ones (Drew, Gibson) had a lot ahead of them at age 29. Frank Howard had the whole famous part of his career ahead, not that Ethier is going to develop that way.

Actually, Matt Pouliot wrote that

Scaryingly fixed.

The important part of the article is:

the Dodgers just gave Ethier a five-year, $85 million contract with what apparently is a pretty easy vesting option for 2018.

Why is that list scary? Similarity scores have zero predictive value.

You're no fun.

Why is that list scary? Similarity scores have zero predictive value.

Well, they don't, but they can be indicative. In this case, they're indicative of the fact that Andre Ethier is not that great a major league ballplayer.

He's been a regular for six seasons in the majors, ages 24-29, and he's put up B-R WAR numbers of 2.2, 0.9, 2.0, 2.7, 1.7, and 2.1. That's with an average defensive rating of about four runs below average per season - nothing terribly extreme. Fangraphs likes him a bit more, but still places his value per season a bit under 3 WAR.

Ethier is a good but not great hitter who plays indifferent defense in a corner and doesn't run well. This contract would have been fair for his 20s, it's a huge overpay for his 30s.

@7: Cosigned.

[7] Well stated. Isn't that kind of like saying that PECOTA or SCHOENE or whatever the forecasting tool du jour is isn't predictive? They're both using the same type of data, aren't they?

[9] No, they aren't. The similarity scores on BB-Ref are based on Bill James's concept, explanation here. They have little relation to player value and no predictive value. The projection systems use their own proprietary similarity scores. The actual formulas are not public but the stuff they have revealed about them suggests they are not at all like the Bill James similarity scores. You'll note that in [7], MCoA mentions the sims scores, but then goes to WAR to make his point...

Also, to be clear, I don't disagree that this contract probably will overpay Ethier, but the sim scores are irrelevant.

To add on to Randy's point, the major difference between similarity scores and the inputs used for the more popular projection systems is that they're adjusted for offensive context.

Even the simplest of projection systems (Brock 2 and Marcel) use the 3 most recent seasons as their inputs. IOW career totals don't matter to even the simplest of projection systems.

I think "irrelevant" is too strong. They're not a good measurement, given the problems of league and era difference, and some of the more arbitrary weights. But when the sim score numbers return lots of pretty eh players, it's often an indication that the original guy in the comparison is pretty eh, too. It's something to take a look at. And when you take a look at Ethier it turns up, yup, a pretty eh player in the aggregate.

One other thing on sim scores - with the fading of the sillyball era, run scoring is now much closer to historical norms. This will make sim scores more useful for off-the-cuff comparisons.

EDIT: And, of course, Ron is exactly right that the other really big problem with using sim scores for projections is that they don't use weighting of more and less recent seasons. The sim scores in ZiPS and PECOTA are based on heavily weighted recent performance.

EDIT2: To be clear, I'm not saying that James' sim scores will become useful for projections, but that's not what they were ever designed to do. It's for comparing players over their whole careers, which is a fun thing to do, and will become easier with the fading of sillyball.

The similarity list is garbage, because it doesn't adjust for context. Ethier has a career 124 OPS+, that comp list is at 116.

Let's say, guys through age 30 within 750 PA of Ethier either way, 5 OPS+ points either way, with 50%+ games at a corner OF position. That gets me 39 guys. Ethier doesn't steal bases often or well, let's carve out guys with a SB:CS of less than 1.5:1. 7 guys. Good enough:



Provided by Baseball-Reference.com: View Play Index Tool Used
Generated 6/12/2012.

So a list that says Ethier is similar to Roy Cullenbine and Joe Cunningham is somehow better than the one in the excerpt? He's a lot closer to Bobby Higginson and Jacque Jones than those guys.

#13 What you really want to do is search for corner outfielder with 1500+ PAs and an OPS+ between 124 and 134 for ages 27-29. As noted above, earlier play is of no particular interest in projecting forward.

Incidentally we don't have to guess what ZiPS saw as the most comparable players.


Interesting. I guess ZiPS treats corner OF and 1B as an almost interchangeable talent group and also sees handedness as significant.

Ett-yay no JD Drew.

He can't hit LHP for ####.

So a list that says Ethier is similar to Roy Cullenbine and Joe Cunningham is somehow better than the one in the excerpt? He's a lot closer to Bobby Higginson and Jacque Jones than those guys.

Higginson, maybe. Jones, not even close. He's far better than Jones at the same age. A much, much better hitter, and much, much better arm. They're not close at all. Jones' career-high OPS+ through this age was below Ethier's career average.