Defensive metrics question

[keystone]

I am still in the beginning stages of learning stat analysis, and I find that there are several different metrics for defense. I am currently checking UZR at FanGraphs, because it's a free site and UZR seems to be a respected stat. However, there is the Baseball Prospectus site (I don't know what statistic it uses for defense), as well as Bill James' site with Runs Saved.

So, my question is this: Is UZR the best single statistic to use to measure defense, or would it be worth the money to learn what is being offered on one of those other sites?

[leewileyfan]

Hi, keystone:

I would suggest that you pursue this interest in defense with an open mind, and always keeping your personal options open. Defense is a fascinating and challenging portion of the game and a little bit like Jello to nail down.

I have developed my own metric which is as yet unpublished. I have never been sold on UZR at all, nor have I been overly impressed by Dewan's books published as defense "Bibles." I was early on attracted to Pete Palmer's linear weights [LWTS] as applied to defense [Chapter 4] and quantified [Chapter 11] in his classic "The Hidden Game of Baseball." Then, too, Baseball-Prospectus approached defense with the PECOTA system, which I found to have some merit as well.

My own approach [with an historical focus, 1901-Present] is to consider each defensive position during a particular generation of play. For example, 1901-1920 or thereabouts, the 3B and 1B are more challenged with "small ball" [bunts, run and hit] than players after the live ball was introduced, in stages, 1911 and 1921. After 1921, the DP really exploded; so 2B and SS roles took on added challenges. Then, along the way, playing surfaces [grass seeds and strains; artificial surfaces; hybrid surfaces] changed batted ball dynamics, as did stadium design and enclosed arenas.

My personal approach is both complex and ultra-simple. If you are interested in arriving at your own value set ... some steps I took:

1. Assign degrees of difficulty to the types of plays unique to each position.

2. Weigh the degrees by athleticism demands of plays [assists more varied and challenging than put outs - involving range and arm elements among infielders especially] and assign input standards to each position.

3. Segregate put-outs for catchers, so as to eliminate pitcher K's. The catcher gets a PO for each; but if he can't record those. he doesn't belong in MLB. Catcher tell-tale gifts lie in the assists and DPs and the caught stealing rates.

4. Recognize that put-outs are the outfielders' bread and butter ... with some being routine and others being web gems. Factor in DPs and assists as well. Assists do not necessarily suggest a lame arm. Often they indicate the reverse.

If one constructs a defense metric in this fashion, then I would argue that a massive display of comparative defense history can be displayed across 112 years of play since 1901. One can rate the players and then convert the ratings into defense runs over a season or a career.

Also, depending upon the effort into constructing the metric and its sensitivity to changing inputs [as the game itself evolves], I suggest that all the evaluator then needs are the most commonly published stats in the box score: A, PO, DP, E, PB. Where the K's come in has to do with opportunity equivalence. I then end up with a dual evaluation:

a- defense rating and defense runs based on raw input data;
b- adjusted defense rating and defense runs, adjusted for the defensive player's team pitching staff strikeout recorded.

Say a Detroit 2B has a rating of .955 which might equate to +5.2 Defense Runs above average. If his pitchers struck out only 900 batters during the season against a League average of 1,100, then he has potentially [equivalence] had a number of opportunities more than opponents [900-1,100] = -200. That's a team number; so if 2B get 14% of fielding opportunities, then his opportunities are 200*.14 = 28. If each K is worth .11. then he would adjusted down by 28*.11 or 3.08 DR. Then too, there's playing time that has to be factored in.

Good luck. It's interesting stuff. If you can get a copy of "The Hidden Game of Baseball" by John Thorn and Pete Palmer I'd recommend it as a key starting point.

Some have discounted it over the years; but I am cynical enough to believe the harshest of these critics has his own axes to grind.

[Matthew C.]

Of the "mainstream" systems, there is a lot of debate if UZR is better than Plus/Minus or not. Michael Humphries came up with a published system in his book Wizardry that has some avid followers. Like Total Zone, it tries to capture defense prior to the batted-ball data era. And of course you still have FRAA and Defensive win shares, which despite some followers, are not supported by sabermetricians as much as UZR, etc. And then you have non-published systems, like the one our friend above is supporting (and our own SabrMatt's PCA), that all have their merits.

If you find out which system is really the best, let the world know. I don't know how much "money" you want to spend, but I would definitely spend the time to look into every system you possibly can.

[leewileyfan]

Of the "mainstream" systems, there is a lot of debate if UZR is better than Plus/Minus or not. Michael Humphries came up with a published system in his book Wizardry that has some avid followers.

Matthew C:

Thanks for the heads-up on Michael Humphreys being published. Several years ago I saw his metric approach outlined in a baseball forum and up for discussion. I believed he was on a track that I could appreciate. He was using a different approach from my own; but he made sense in what he was looking for. While our evaluations were not always in total agreement, I saw certain consistency in his work that I also found in the PECOTA approach to defense.

I lost track of him and often wondered how his work was progressing. Glad to see he was published. [Maybe there's hope for me - if I dust it off and make the attempt].

I am putting in an order tomorrow.

[keystone]

I second that. Should be an interesting read.

However, the drawback would be that you can't look up a current player on-line.

And I don't do math.

Besides the book, which website would you spend money on for defensive stats (if either):

Baseball Prospectus or
Bill James Online

Currently, I'm using Fangraphs. But I don't think they' posted some of their stats for this year yet?

[leewileyfan]

keystone:

I am only taking a stab at this; but I am thinking of taking up my own book effort again ... and if I do decide to pay any premium for up-to-date accessible stats that are meaningful for me, I'll go with Baseball-Prospectus. From my viewpoint, I want the stats. I can interpret them and apply them to my own metric.

[keystone]

I do hope you can gather the muster to start on your book again, leewileyfan. I think you have some really interesting ideas. I'd buy a copy!

Is Baseballll Prospectus appropriate for beginners like me?

[leewileyfan]

keystone:

I believe B-P is appropriate for anyone starting out or deeper into the stats. The B-P site once had some detailed stats, defense ratings player histories in innings played [rather than [vague] "games," all free. Much of that good defense data seems now to be available only for paid users. I am deciding that there's value in the site; so I can't go too far wrong by investing for that data I need.

Plenty of room to explore and grow with B-P. Of course, Baseball-Reference has a good deal of free data as long as you know how to browse their site to maximum benefit.

Thanks for the encouraging words.