On the Recent Alterations to Jose Abreu’s Steamer Projection
In the past week, the Steamer projections for Cuban emigre and presumptive White Sox first baseman Jose Abreu have changed a couple of times, from a line of .292/.381/.554 to .279/.364/.518 to .272/.356/.541. For many, this might be confusing. For some, these changes might only serve to underscore for them the fact that projections are, not unlike life itself, totally and irretrievably absurd. Many others probably didn’t even notice, have no opinion and have already stopped reading this paragraph.
Indeed, objective analysis of such things suggests that the Steamer does pretty well for itself, in terms of producing accurate projections. Both common sense and also math, probably, indicate that the easiest sort of player to project is one with a major-league track record. Players who’ve recorded substantive plate appearances or innings in the high minors are the next easiest. Those in the low minors are more difficult to project. The Japanese league falls somewhere after that, probably. And finally, for players who’ve only produced stats in the Cuban league, formulating projections presents what a less refined person than this author might call a “frigging panoply” of troubles.
Regarding why such an endeavor is so difficult, the absurdly coiffed Eno Sarris wrote a comprehensive piece last month — a piece in which he summarized those difficulties in the form of bullet points.
Bullet points like these:
• Inconsistent record-keeping.
• Small sample size of Cuban players coming to America.
• Non-existent sample size of American players going to Cuba.
• A recent change in ball size.
• Sheer number of parks (40 parks for 17 teams!).
• Inconsistent park conditions.
• Political influence on roster decisions.
• Large swings in talent level based on defections.
That list speaks to all of the potential variables in formulating a projection for a Cuban player. Regarding why Abreu’s Steamer projection, specifically, has changed of late, I corresponded with Steamer owner-operator Jared Cross for answers.
For Cross, there are essentially two main variables which have required his attention: an understanding both of (a) the level of competition in the Cuban league and also (b) the run environment in the Cuban league. While isolating these variables would initially appear to resemble more science than art, the small sample of players who’ve made the transition from Cuba to the States (and the complete absence of players who’ve gone the other way) requires the use of capital-J Judgment, as well.
Here’s what Cross writes with regard to his methodology for estimating the level of competition in Cuba:
It’s… based on past players. I have 12 hitters (this century) who played in Cuba and MLB and another seven who played in Cuba and the minors and I used those 19 players. I figured that just using the 12 who played in MLB would be a biased sample since these are, relatively speaking, success stories (or non-failure stories, at least).
With regard to the run environments in Cuba, he adds this:
By [that] I mean adjusting for the fact that offense has been at higher levels in Cuba than in MLB in recent years. I had the Cuban league (Serie Nacional) with wOBAs of .358, .360 and .336 in 2011-2013 for instance (although these are estimates because I don’t have HBP and PA).
In summary, there have been two questions with which Cross has had to contend: how to estimate the quality of Abreu’s competition in Cuba and also how to adjust Abreu’s raw stats relative to league average in Cuba. Difficult, that.
The first iteration of Abreu’s slash-line noted above (.292/.381/.554) was the product of estimating the Cuban league as equivalent to High-A, thought without any sort of adjustment for Abreu’s raw numbers. The second iteration (.279/.364/.518) is a product of regarding the Cuban league as more of just a Class-A league, still without any sort of adjustment to Abreu’s raw numbers relative to Cuban league averages. The final, current projection for Abreu (.272/.356/.541) was produced by classifying the Cuban league as actually somewhere between Class-A and High-A — and this time including an adjustment of Abreu’s raw figures relative to league average there.
If Abreu’s current Steamer projection is less optimistic than in that first iteration, it’s because of that further league adjustment, it would appear. That said, all three iterations paint a mostly similar portrait — of a hitter with above-average power and decent plate discipline, and a likely candidate to produce wins at an above-average rate in Chicago this summer.
Carson Cistulli has published a book of aphorisms called Spirited Ejaculations of a New Enthusiast.
A bit off topic, but will the ZiPS and/or Marcel projections be posted on the Projections page?
So…Alexander Guerrero is definitely hitting .250/.309/.418 next season?
I’ve read that telling people about a statistical model that you once built and have since revised is a bit like introducing people to your ex-wife. I don’t have any ex-wives so I can’t speak to this directly. That said, I think think Steamer can make this new Abreu projection stick. Another triumph of hope over experience.
“A recent change in ball size”? Did that really happen?
Yes, possibly because they don’t test for PEDs in Cuba.
I’ll show myself out.
Think you meant 40 teams for 17 parks.
No, that’s correct. Read Eno’s piece, they apparently have a bunch of games that get played at really weird places.
they also only have him down for 398 PA’s
Another complication is that sometimes for visa reasons Cuban players miss a year before moving from playing Cuban baseball to playing in the United States. That introduces another source of variance so to get an unbiased projection we’d have to make not just a single year age adjustment but an additional year(s) plus some adjustment for not playing competitively for a year. But with such a small sample, that’s really not possible without overfitting the data.
If we were to do a 95% confidence interval on the projection for Abreu, it would be something like a 750 OPS to 950 OPS. Which is not a very informative prior. Basically, all that we can say is that he almost certainly has the talent to be on a MLB roster, but it’s not clear whether he’ll be a below-average hitter for a 1B/DH or an All Star.
WAR is a black-box subjective stat remember. Hence why different sites obtain different values. Some sites will have a guy +3 WAR on defense, while this site might be -2 wins. All stats and projections you can’t verify the formula yourself are only marginally valuable at best (similar to ESPNs Total Quarterback rating)
True, but offensive WAR, for the most part, has been standardized across different sites. So for a DH, the WAR values will be pretty much identical.
Just because it’s extremely difficult to calculate on your own doesn’t mean it’s a black-box. While I’m not aware of publicly posted full documentation on any of the versions of WAR, this document gives a pretty detailed run down (http://www.baseball-reference.com/about/war_explained_comparison.shtml).
There are a few somewhat subjective decisions that have gone into their development (which defensive stat to include, and generally what makes up that defensive stat, what replacement level to use, etc.), but as Big Daddy says the offensive component is pretty standardized.
…whereas Fangraphs and other sites just tweak defensive WAR a few times a year to ensure that Trout’s defensive WAR is improved (remember when Miggy’s defense was statistically rated better than his for a good portion of 2013…)
Worst post of the century.