THE BOOK--Playing The Percentages In Baseball

22 wins in 5 seasons, for an average of 4.4 per year?

I’m confident Tom Tango means Wins Above Replacement, not the old-fashioned pitcher wins statistic.

That’s Wins Above Replacement.  Assuming 1 decision per 9 IP; ZiPS has his average record about 12-8, while Oliver is about 15-8.  Pretty good, but not great.

Thanks guys.  I had no idea what Christian/1 was asking me about.  I was thinking… uh, yeah, so what’s your point.

It never dawned on me that we’d be talking about good-ole pitcher-assigned wins.

***

Oliver has him on par with Kershaw (in both ERA and IP), and best in MLB.  Oliver has Strasburg as a match in ERA (though way behind in IP).

All of them are ahead of Verlander, Felix, and Jered Weaver.  That’s some mighty big b-lls to have dangling out there.

What do the projections have Darvish’s K/9 and BB/9 at?  Sounds like the Oliver projections are too optimistic.  I would think the error bars on this type of thing would be pretty large, especially given the huge change in the run environment that they had in the Japanese league last year.  I think this would be a good opportunity for one of Tango’s Fan’s projections.  ie - Yu Darvish has the same chance of earning 25 WAR in the next four years as he does earning less than X WAR in the next four years… or something along those lines.
vr, Xei

Xei: excellent idea!  Basically, do the same thing I did with Strasburg last year or two ago.

I mean, we just go through being told the Strasburg is a once-in-a-generation pitcher, and he might end up #3 behind Kershaw and Darvish.

My guess is that CJ Wilson will be worth more wins over the next 5 years than Darvish. And for $30m - $40m less over that length of time.

Id put 1/4 odds that he has a 22 win season in the next 5 years.

Wow, that Oliver projection is ridiculously optimistic.  I wonder how confident Brian is in his NPB translations.

I’m hoping that Brian shows up, because I haven’t had an argument regarding forecasting in a long time.

As my poll shows, no one is acting like Yu = Kershaw, but that’s what Brian is showing.  That’s not to say he’s wrong, but it does show he’s sticking his b-lls out there for everyone to see.

thanks for answering my question, this article makes a lot more sense to me now

ZiPS’ 85th percentile projection (28 WAR) is much closer to the median (22 WAR) than the 15th percentile (6 WAR), making the MEAN ZiPS projection worse than 22 WAR.

I was working on an article about this subject when I found this thread.

I want to “show my work” and not just have people take my word for it that the numbers work, so I started with a list of 33 Japanese pitchers who have come to MLB from 1995-2011. I’ll show their last 3 years in Japan, their projection, and their first 3 years in MLB.

Last week I coded an adjustment for the new standardized ball in NPB in 2011. It cut a run off the league ERA and dropped HR’s by a third. The adjustment worsened Darvish’s projection by about a third of a run.

For a FIP estimate of a pitcher coming from NPB to MLB, increase HRs by 68%, increase BB by 11.6%, decrease SO by 9.6%.

Darvish’s raw stats for 2011 were 232 IP, 5 HR, 36 BB, 276 SO. Converted are 232/8/40/252. Scaled to 207 IP (the projection) it’s 207/7/36/225. Adding in data from the previous two years, which were not as good as 2011, and regressing, the final projection is 207/10/44/227, a FIP of 2.27, but calculating ERA from his projected wOBA allowed it’s 2.57.

What’s your league average ERA?

What’s your justification for constant IP year to year?

It would also be useful to look at the comments starting from post #106

http://www.insidethebook.com/ee/index.php/site/comments/strasburg_predictions/#106

I have no projection for Darvish, as I have no Japanese translations.

If he is projected to throw in the low 90’s in MLB (I read that), unless he has ridiculous command or extraordinary secondary pitches, I can’t imagine him being a Kershaw, Verlander, Halliday, Strasburg type of pitcher.

Also, while I hear that he has great mechanics, or at least few mechanical flaws, to me (I am nowhere near an expert on pitching mechanics) his arm motion looks like an elbow injury waiting to happen.

I want to “show my work” and not just have people take my word for it that the numbers work, so I started with a list of 33 Japanese pitchers who have come to MLB from 1995-2011. I’ll show their last 3 years in Japan, their projection, and their first 3 years in MLB.

Last week I coded an adjustment for the new standardized ball in NPB in 2011. It cut a run off the league ERA and dropped HR’s by a third. The adjustment worsened Darvish’s projection by about a third of a run.

For a FIP estimate of a pitcher coming from NPB to MLB, increase HRs by 68%, increase BB by 11.6%, decrease SO by 9.6%.

One thing I noticed when trying to forecast Darvish was that relief performance translated better than starter performance.  So for Darvish I only looked at how the following pitchers did.

Igawa, Kei
Irabu, Hideki
Ishii, Kazuhisa
Kawakami, Kenshin
Kuroda, Hiroki
Matsuzaka, Daisuke
Nomo, Hideo
Yoshii, Masato
Lewis, Colby
May, Darrell

I get an ERA of around 3.41 for Darvish as a Ranger.

Most Japanese relievers have done very well over here, but I hadn’t separated them from the starters. I will try that.

I did split players who started their careers here vs in NPB. For pitchers, small differences in BB & SO (guys new to MLB translate worse coming here), HR was somewhat more so.

Pitchers     BH    HR   BB     SO 
American  1.006 1.664 1.125 0.923
Japanese  1.001 1.842 1.181 0.871  

Maybe Japanese starters are more like relievers in Japan because of the longer time in between starts.  IOW, they can “air it out” all game because they have more time to recover as compared to American starters who have to pace themselves (as compared to relievers).

So we use around a run difference between starters and relievers here in the U.S.  Maybe in Japan, it is only around .5 run.  So the translations would have to be done differently for starters and relievers.

Is that possible?

MGL - yes, that’s possible. I just need to label each pitcher in the matched pairs as a starter or reliever, calculating and applying factors for each.

I queried MLB pitchers 2005-2011 (Gameday data),
gs >= 20
bb% < .070
so% > .220
gb% > .520

                      IP    ERA  GB%  BB%  SO%
Felix Hernandez 2010 249.2 2.27 0.56 .069 .232
Roy Halladay    2011 233.2 2.35 0.53 .033 .237
Adam Wainwright 2010 230.1 2.42 0.54 .060 .235
Roy Halladay    2010 250.2 2.44 0.53 .029 .221
Cole Hamels     2011 216.0 2.79 0.55 .050 .229
Chris Carpenter 2005 241.2 2.83 0.55 .054 .224

Yu Darvish                 2.57 0.53 .052 .275

An article at FanGraphs attributed Patrick Newman at npbtracker for providing a 2011 gb% of 0.57 for Darvish, with a league rate a little higher than MLB. He’s always had a very low hr%, so it’s a safe assumption he’s a ground ball pitcher.

Everyone on the list had a lower so% than I project Yu for, but even for between .220 to .240, combined with higher than average gb% and lower than average bb%, they all had ERAs below 3.00.

So yes, that’s the level of production I’m projecting from Darvish, and I would tend to think that a posting fee of over $50m indicates the Rangers feel the same way.

But none of them have a TRUE TALENT level of 2.57 ERA.

True -

2.76 Kershaw
2.82 Strasburg
2.90 Verlander
2.98 Halladay
3.06 Lee
3.06 J Johnson
3.17 Cain
3.18 Felix

I was doing a lot of tinkering with the Japanese pitcher projections last night, accelerating things that had been on my to do list. In the end, even though there was some movement in the components, I got the exact same ERA estimate.

The part I’m least sure of (for all the pitchers) is the BABIP estimate. I’d like to do more with gb/fb and quality of fielders. Darvish has always been very low, projection is about .280. That’s easier to maintain with a lot of flies than with a lot of grounders. The babip curve does start back down at the high end of gb%, but not as much as at the low end.

In my model, it’s still going to be damn hard to get an ERA projection for him over 3.00, I have been trying.

Two questions:

1. What is the league average ERA you are setting?

2. Are those park-based?  That is, Darvish is 2.57 in Texas for half his starts or in neutral for all his starts?

1. I’ll have to get out my calculator. Each component is scaled with log5 to the MLB average of the past few years.

2. For Darvish it’s park neutral. I’d have to figure out some manual override to put him in Texas.

Two more;

1. Did we every settle the question on whether MLB stats carry more weight than MLE’s, and if so, how much?

2. Is there a greater uncertainty level with the Darvish projection compared to the guys with 600 MLB IP over the last three years?

1. Part 1 is YES.  Part 2 is I don’t know.

2. Definitely.

If you frame your questions into “how much”, then we can talk.  In terms of yes/no, of course it has to be that there’s less uncertainty if you forecast Felix and Kershaw than Yu.

Heck, there’s less uncertainty if you forecast Felix for 2011 than Cliff Lee for 2011, based solely on the fact that Felix didn’t change teams.

To calculate MLE factors for each league, I do a matched pairs comparison. What I will do is after calculating the league factors, which should reduce the mean error to zero (all the low errors cancel out the high errors) go back and repeat the matched pairs with the league adjusted value to get the total (root mean square) error. These errors should get larger the further away from MLB the league is, and would be a measure of the uncertainty. The more uncertainty, the less weight that can be given to the translated stats, which is then replaced with more regression.

But also to #26, I also use the term MLE to refer to MLB stats that have been translated for park factors. There’s raw and there’s adjusted, and I use MLE generically for any adjustment. Which would go to Tango’s comment in 27 about Felix vs Lee.

MGL - here are translation factors broken by SP and RP

NPB to MLB, 2011 Ball  WtPA    HR   UBB    SO
American Starters     66697 1.520 1.089 0.981
Japanese Starters     18802 1.727 1.237 0.862
American Relievers    52943 1.876 1.127 0.918
Japanese Relievers    20098 2.019 1.210 0.914

These results suggest Japanese starting pitchers suffer the worst translation for BB and SO, but the differences between Japanese starters and relievers is not as great as the difference between all Japanese pitchers and all American pitchers.

Brian: I’d love to see the starter/relief thing I did for MLB applied to Japan.  That would be quite the informational piece.

You can email me if you want to discuss further…