No, I do NOT think CPI is better than SLWTS.

I think it is simpler. I'm not sure of the accuracy yet.

Chris, good job and good discussion. I missed that article.

Of course, it all boils down to what metric we use for defense. For example, you have Tejada at +2 runs at that time and I have him at around +20. We know of course that Tejada is probably not a "true +20" but he may have saved 20 runs in 2004. Who knows. Defense can and does fluctuate like anything else, around a player's true talent.

I do have a problem with using context neutral metrics for MVP awards though. I hate to say it, but about the only thing that should be used are garbage stats, like runs and RBI, and even then, an "adjustment" should be made for games actually won by a player's team.

It is not "how well a player played" award. If a player has a high lwts or XR or VORP, but his good hitting happened to occur in meaningless situations, was he really "valuable?" Probably not. And even if a player amasses lots of runs and RBI, but again, those runs and RBI happened to occur in losing games, again, "Was he really valuable?"

mgl,
I agree that context neutral evaluations should be amended to consider clutch hitting (and whatnot). I think Tejada's effort a few years ago in the clutch is an example.

My "final" workup will depend on how well Tejada played in September. As he finished at the top (or near it) in ZR, I suspect he picked up quite a few runs (as well as others dropping, re-setting the "average").


I am working on that now. It'll be better, as I am improving the system.

mgl,

i think you make a really good point about valuable. a really good point. and that's why i think that guys who help their team win a title are considered more valuable than those who don't. it isn't the "highest win shares award."
so maybe we should look more at how a player does in the most important situations. (can i say clutch?)

So, if Jenkins (+25 in left) should be a centerfielder, can Beltre (+29 at third) play shortstop? I'm not asking rhetorically, I think it could happen.

What if the Red Sox sign Beltre to a long-term deal and ask him to play SS for a year while waiting for Hanley Ramirez? If Beltre was a true +20 player at third base, and we say that Beltre loses 5 runs b/c of the positional difference and another 5 runs are lost in the adjustment period (although Beltre did come up as a shortsop) we could still project him as a +10, or well above average, shortstop.

Belre seems like a much better choice than Varitek and probably a better choice than Beltran.

2005 ZiPS

Beltran (age 28) .278/.373/.527
Varitek (33) .274/.361/.452
Beltre (26) .296/.350/.541

I should also point out that while Beltran and Varitek are projected at hitter's parks, Beltre's projection includes playing his home games in LA.

Beltran should get credit for excellent baserunning but I'm guessing (for lack of SLWTS) that Beltre's defense in more valuable. He looks like he the best FA on the market and might even be underrated at this point.

I did a lot of analysis this winter on inter-year correlations of UZR (using the full 2000-3 spreadsheet). You definitely want to regress to the mean to get a best-guess projection for an upcoming year. This Year = .48 * [Last Year] + .25 * [Year -2]does the trick, although the average error (straight average, not RMS) is a whopping 10 runs. (Three years ago had a mild, significant (p = .03) correlation looked at in isolation, but added nothing to the multiple linear model at all.)

OTOH, those who are wondering whether the relationship of UZR to UZR of two previous seasons is significant can sleep tight. For the 130 sets of three player seasons I looked at, p < 0.0000000000005 (5 * 10^-13). For this multiple linear moidel, r = .6, r^2 = .36, which is not bad.

When lacking two years of previous data, just use .60 times the previous year. Correlation is decent sized (r = .53, r^2 = .28) and ridiculously significant over these 267 pairs of seasons (p = 2.2 * 10^-20).

There is more regression to the mean for bad years than for good ones (doubtlessly because of injury years); for a player -10 or more, the factor is .47, while it's .70 for a player +10 or more.

Eric MV wrote, "I did a lot of analysis this winter on inter-year correlations of UZR (using the full 2000-3 spreadsheet)."

Is that full 2000-3 spreadsheet we look at still fairly state-of-the-art, or has it been upgraded/corrected?

MGL wrote, "Edmonds was -24. Don't know what's up with him. He has been negative for several years now."

According the public UZR spreadsheet, here are Edmonds' years (UZR per 162 games):

'00 5
'01 -27
'02 6
'03 7

His total is given as 0.

(His arm was +1 or +2 except for a +9 in '01. Similar to what MGL reports about '04, in '01 he had both terrible range and exceptional arm action. Some sort of shallowness nexus?)

MGL, if you do make improvements to the recent past information (or get the chance to extend UZR info back another decade), I hope your arrangments will allow you to keep instructing us. I suppose it's a question of what's not too recent, or of 'fair use'...

Thanks for these highlights about '04!

I think the 00-03 public data was generated using the same (current) methodology as this year's data.

Eric M. Nice work on the data.

For you non-stats people, just because we have, as Eric says, a fairly nice year-to-year correlation, doesn't mean that UZR measures what we want it to measure...

Eric,
that is some interesting work . It's good to see that.

Well, using the handful of numbers mgl presented here (there are no NL RF names), for the NL defense, my numbers have an r^2 of 0.8+ for 1B, 3B, SS and 0.4 for the OF, and an awful 0.12 for 2B.

I have to use a larger sample, and will compare to UZR from 00-03.