probabilistic model of range charts [Archive]

ojo

03-28-05, 02:52 PM

not good. not good at all.

http://www.baseballmusings.com/cgi-bin/ChooseChartPlayer.py

shows how matsui, jeter, bernie and sheff are all below avg. badly.

at least womack is an upgrade, though.

PaulieIsAwesome

03-28-05, 07:50 PM

The numbers really don't seem great to me. First, some are just so non-intuitive (Kevin Millar as a great first baseman, ok, but as a great rightfielder also!?). Also, they go against some of the other metrics (UZR and FRAA.) Still, this is another tool, allowing us to move closer and closer to answering the question of defense.

ojo

03-28-05, 09:16 PM

maybe he's an excellent RF with damon in CF...

PaulieIsAwesome

03-29-05, 04:38 AM

I might buy that, looking closer. Pinto thinks Corey Patterson is a fantastic centerfielder, and that Sosa plays great defense. Not intuitive, but maybe Park Factors or something made them both to be pretty darn good. Now that I look at it, BP's RAA2 rated Sosa at +8 Runs/100 games last year, which is of course fantastic.

Of course, Pinto doesn't think Damon's all that great, so it doesn't really make any sense that Millar would be great as a function of Damon's greatness.

Prickly Pete

03-29-05, 08:50 AM

I'm wondering if there's a sample size issue here, because some of the findings are just off the wall. Perhaps because of the way he uses probabilities in his calculations, he needs more than one season's worth of data to give an accurate reading.

I am a firm believer that observation alone is a terrible method of judging defense, but some of Pinto's results are hard to reconcile with what I've seen.

PaulieIsAwesome

03-29-05, 09:00 AM

I'm wondering if there's a sample size issue here, because some of the findings are just off the wall. Perhaps because of the way he uses probabilities in his calculations, he needs more than one season's worth of data to give an accurate reading.

I am a firm believer that observation alone is a terrible method of judging defense, but some of Pinto's results are hard to reconcile with what I've seen.

Good points. I like the DRA/UZR regression at Hardball Times, http://www.hardballtimes.com/main/printarticle/defensive-regression-analysis/

I'd like to see some correlation analysis on these results with Pinto's.

Prickly Pete

03-30-05, 06:00 PM

Good points. I like the DRA/UZR regression at Hardball Times, http://www.hardballtimes.com/main/printarticle/defensive-regression-analysis/

I'd like to see some correlation analysis on these results with Pinto's.
Thanks for the link. I hadn't read that. I have some real concerns with regression-based models and happen to think UZR is the best available system so far, but that article answers some questions for me.

PaulieIsAwesome

03-31-05, 04:50 PM

Thanks for the link. I hadn't read that. I have some real concerns with regression-based models and happen to think UZR is the best available system so far, but that article answers some questions for me.

I also think that UZR is the best. I just can't find a good database for all the UZR numbers, and BP has a nice free search engine, so I use that, despite regression problems.

meng_mao

03-31-05, 11:25 PM

hey PaulieIsAwesome, what do you know about math?
Do you have a math degree from MIT or something?
Douche.

KissTheRings

04-01-05, 03:25 PM

hey PaulieIsAwesome, what do you know about math?
Do you have a math degree from MIT or something?
Douche.

nice first post.

PaulieIsAwesome

04-01-05, 04:25 PM

nice first post.

He's a friend of mine, he's not into baseball at all, he found out about this forum, and after some mutual insulting (he's on a Mixed Martial Arts forum) he trolled here, to make up for my trolling on his forum.

W.A. Cummings

04-03-05, 10:09 AM

I don't know what you guys are talking about, Millar's graphs have him sucking in the outfield. Look at those fly ball graphs. Wow. But as a first baseman his above average play doesn't really surprise me. He made some great strides this year.

nyybleachercreature

04-07-05, 02:28 PM

What are the axes in these graphs? Is the x-axis is location? Is so, how does he compile this data? Is the y axis a sort of player rating, 0 being average?

BobbyMurcerFan

04-11-05, 11:38 PM

Yeah, can we get a translation please? Thanks!

Prickly Pete

04-12-05, 01:00 PM

Yeah, can we get a translation please? Thanks!
The short answer is that he's developed a method for calculating "expected outs" and then compared players' actual performances (the number of outs they recorded) to what was expected of them.

The number of expected outs is calculated based on a variety of factors, including how hard a ball was hit, where it was hit, park factors and whether the ball was a fly ball, line drive or grounder (there are other factors).

As for the charts themselves, the horizontal axis represents the physical location on the field where plays are made. If you look at Jeter's chart, for example, it shows a range between 3B and CF -- the places where a SS would be called upon to make plays. The vertical axis appears to be based on his "probability of outs" calculation, but I admit I'm not totally clear on that.

The main difference between this system and UZR seems to be that UZR compares each player to an "average player" -- which itself is based on every player's rating. But Pinto's Probabilist Model system compares the player to an artificially calculated "expectation." I think UZR is better, but both have some strengths and weaknesses.

Hope that helps (and I hope I got that all right).

PaulieIsAwesome

04-12-05, 03:20 PM

The short answer is that he's developed a method for calculating "expected outs" and then compared players' actual performances (the number of outs they recorded) to what was expected of them.

The number of expected outs is calculated based on a variety of factors, including how hard a ball was hit, where it was hit, park factors and whether the ball was a fly ball, line drive or grounder (there are other factors).

As for the charts themselves, the horizontal axis represents the physical location on the field where plays are made. If you look at Jeter's chart, for example, it shows a range between 3B and CF -- the places where a SS would be called upon to make plays. The vertical axis appears to be based on his "probability of outs" calculation, but I admit I'm not totally clear on that.

The main difference between this system and UZR seems to be that UZR compares each player to an "average player" -- which itself is based on every player's rating. But Pinto's Probabilist Model system compares the player to an artificially calculated "expectation." I think UZR is better, but both have some strengths and weaknesses.

Hope that helps (and I hope I got that all right).

Yeah, I had been thinking about this. What if you had, let's say, 3 really awful fielders in an absolute sense? Then, wouldn't above average fielders look incredible by UZR, when in an absolute sense they are only sort of good?

Prickly Pete

04-12-05, 05:11 PM

Yeah, I had been thinking about this. What if you had, let's say, 3 really awful fielders in an absolute sense? Then, wouldn't above average fielders look incredible by UZR, when in an absolute sense they are only sort of good?
That's a good point, but I think it's only a problem if those 4 players are the only ones in the sample. I guess you have to hope that a larger pool of players smooths over that problem.

With Pinto's model, I guess I'm just not yet convinced that his calculation of "expected" outs adequately measures all the variables, especially park factors. I can see why there are park issues to deal with, but why should a ground ball down the first base line at Wrigley be harder to field than a ground ball down the first base line at Comerica?

I'm wary of comparing a player to an artificial "expectation." I'd rather compare players to the actual performance of other players. That's my two cents.

PaulieIsAwesome

04-12-05, 09:06 PM

That's a good point, but I think it's only a problem if those 4 players are the only ones in the sample. I guess you have to hope that a larger pool of players smooths over that problem.

With Pinto's model, I guess I'm just not yet convinced that his calculation of "expected" outs adequately measures all the variables, especially park factors. I can see why there are park issues to deal with, but why should a ground ball down the first base line at Wrigley be harder to field than a ground ball down the first base line at Comerica?

I'm wary of comparing a player to an artificial "expectation." I'd rather compare players to the actual performance of other players. That's my two cents.

Well, I was thinking even if you had 30 fielders, it would run into problems, especially if you compare to a period where there was only awful fielder.

For example, let's say in 2004 there were 4 terrible, terrible centerfielders. These fielders all, in an absolute sense, were about 40 runs worse than the all time MLB average for centerfielders.

Then the next year, 3 retire. Now there is one awful, awful, awful centerfielder. A bunch of guys who had been at 0 the previous year will go down to -5 or something, even if they have the same defensive performance.

I don't know though. It stinks that so much of defensive stats are propietary. One thing I love about offense, pitching and the internet, is that using retrosheet and B-R we can basically be general managers, and can tell who are the best performers.

W.A. Cummings

04-16-05, 07:02 PM

That's a good point, but I think it's only a problem if those 4 players are the only ones in the sample. I guess you have to hope that a larger pool of players smooths over that problem.

With Pinto's model, I guess I'm just not yet convinced that his calculation of "expected" outs adequately measures all the variables, especially park factors. I can see why there are park issues to deal with, but why should a ground ball down the first base line at Wrigley be harder to field than a ground ball down the first base line at Comerica?

I'm wary of comparing a player to an artificial "expectation." I'd rather compare players to the actual performance of other players. That's my two cents.

I assume the PFs wouldn't really be geared towards grounders, but more towards playing outfield, for example it is much much different to play left at Fenway than it is to play left at Yankee Stadium. But also, foul ground will come into play for 3rd Basemen, 1st Basemen, and especially Catchers

Prickly Pete

04-17-05, 09:32 AM

I assume the PFs wouldn't really be geared towards grounders, but more towards playing outfield, for example it is much much different to play left at Fenway than it is to play left at Yankee Stadium. But also, foul ground will come into play for 3rd Basemen, 1st Basemen, and especially Catchers
Read Pinto's blog entries for March 1 and March 28. He acknowledges some of the park factor issues.

http://www.baseballmusings.com/archives/cat_defense.php

Park factor, in general, is a sticky issue for defense measurement. LF and RF in Fenway Park, for example, should not be subject to an identical park adjustment.