The Book on Defense

[brett]

This is a book I'd like to see. A total summary of the defensive side of baseball, historically, by position, and how each of the major defensive metrics work, just everything about defense.

[leewileyfan]

I am trying to compile exactly what you are after [Part 1]. However, it is in-depth 1901-Present for each positon [pitchers excluded, so far]. As for the second part, comps with other metrics ... I am not going there. As I see it, a metric should stand on its own and be done with it. The reader can then read or peruse to whatever depth he wants to pursue - and can then decide for himself which presentation, format, explanation and credibility of results makes the most sense to him/her.

[abolishthedh]

LWF-- Does this mean you won't have any sabremetric measures at all? I can understand that, because it is still open to debate how useful these would be. However, it would be imperative that readers/users be able to compute their own range factors and other modern metrics. This would help some folks, if not most.

[brett]

I am trying to compile exactly what you are after [Part 1]. However, it is in-depth 1901-Present for each positon [pitchers excluded, so far]. As for the second part, comps with other metrics ... I am not going there. As I see it, a metric should stand on its own and be done with it. The reader can then read or peruse to whatever depth he wants to pursue - and can then decide for himself which presentation, format, explanation and credibility of results makes the most sense to him/her.

It is a tough "sell" to get people interested in a book focusing on a single metric by someone who is not already established. There are different statistical factors that people have equated to defensive performance and these may change over time. The best shot at a book on defense is one that does everything. For example some have posited that a team that gives up very few triples compared to expectations is a sign of having a first and third baseman with excellent range. This alone could be a chapter-"do triples correlate to corner infield defense." Another is that doubles allowed is perhaps the best estimator of outfielders range. A metric does not have to include these to still have the ability to investigate these questions-in fact the metric can test these hypotheses-ie 10 third basemen have the best range, and their teams allowed low ratios of triples versus doubles.

[leewileyfan]

Thanks to abolish and brett.

As far as sabermetrics [per se] are concerned. I consider the entire metric a sabermetric exercise. I believe this is a true qualification in that it MEASURES defense against a model[s] of comparison. In a nutshell, it takes this approach:

1. It recognizes each position as unique, in function, challenges, range requirements, game involvement, opportunities and penalties for failure.
2. It responds to changes in how the game is played, generation to generation: deadball, small ball, live ball, power ball ... along with changes in the degree that batters put balls in play.
3. It signifies this sensitivity to change via TEMPLATES, which are the net of +/- data inputs for each position, each player. As RATINGS rise up to and beyond the prescribed template, there is evidence of a game change.
4. The TEMPLATES are flexible, changing as position demands change.
5. The arithmetic [I do not say "mathematics"] is complex ONLY because of adjustments made because of game changes.
6. The core of the metric is the assignment of degrees of difficulty for A, PO, DP, E, PB, and WP for each position, along with position based negatives, the cost potential for misplays. I intentionally restrict data inputs to these basic elements because: [I believe this addresses brett's question on, say, extra base occurrences at 3B]:

a. My desired perspective and focus is on historic context [1901-Present]. I want both to appreciate players from different generations of play, and to be able to compare their performance across the generation. The starting point hold to the adage "Less is more."

b. I firmly believe that the more direct and simple the input elements are, the clearer and well-defined the results will be. Credibility is earned, I believe, more from the background study one puts in knowing the game at its different stages, and putting oneself in the position of players from different decades [role playing] on actual diamonds.

Before I prepared my first spreadsheet [1901] I had spent weeks making notes on diamond layouts, dimensions, and ball park graphics at the turn of the century ... the uniforms, equipment and conditions under which the players performed. Only then did I assign the DEGREE OF DIFFICULTY to each play demanded at a specific position.

As far as the challenge in getting published, I am well aware - having had a top NYC agent interested [but non-productive] over 18 months. I guess I'm in a unique situation, in that my actuarial sands have to be running out [let's say the back stretch rather than the home stretch]; I am happily retired; I'd LIKE to see something come of the effort to publish; but I won't be destroyed by non-publication. I am convinced the metric is credible and stable. I'd like to share it. Meanwhile, the ultimate reward, for me, is finalization of the polished work.

I hope you guys will raise some player, position and related question in the Evolving MLB Defense History thread, where I may have shot myself in the foot by adding 2B to the title. It's up for all positions and all players [1901-Present].

[abolishthedh]

Here is a summary of some of the things I have wondered about over the past 20+ years, or so.
1) How could you measure the effect on defense of pitchers who take too much time between pitches? Would you have to rely upon game times posted at the bottom of the box score? Would you have to compile your own data of how much time pitchers spend on the mound, or is the time spent on the mound already available?
2) How well have teams focused on defense during spring training? Could this be measured in a case study basis by focusing on a manager (such as Buck Showalter) who might have kept track of his defensive drills during spring training? Is there another way of measuring this?
3) How does roster turnover affect defense, for better or worse? Would this be measured in the aggregate, or by case study? Would it be possible to measure several trades all at once, or would you be forced to study key trades which did or did not work out well?
4) What trends might be evident on any common metric, over several decades? For example, to what extent did increased glove size improve defense?
5) What effects might be evident over temperature extremes? Baseball is unique in the sense that temperature might vary over the course of a six month season more than in other sports. Do pulled muscles occur in these extremes more often, and what other effects are evident?
6) Are there such things as defensive slumps? How would you measure that when most managers will opt for a defensive replacement or trade before allowing a player to work it out?

Those are statistical investigations, which could be augmented by old-school observations of coaches, managers, players, retired officials and players, and media. The augmenting of stats with old school insights would lend a greater audience and probably add legs to how well the book sells over time.

[leewileyfan]

Abolishdh: You raise some interesting points, only a few of which my research addresses in any direct way.

1. Pitcher game delay between pitches, I would suspect, if exaggerated, would be wearing on defensive players wanting to "get on with it." I probably wouldn't invest much time in this myself, because of the following:

a. There is a Rule Book history that touches on this, countered by the inherent KA-CHING factor of the cash register for time in which to generate commercial revenues.

b. If the delay wears on team morale [or opposing team patience] managers, coaches and players are not shy about registering complaints loud and clear. I kind of doubt that a very aggravating situation would last long enough to really hurt a team before some corrective action is taken [like a trade].

However, game times, as you noted, could be measured alongside a basic defensive metric like team defensive efficiency ratings in B-R.

2. As for spring training, I believe managers, trainers and coaches must, of necessity, expect that some players will report is some kind of condition less than they might have wanted. This, I believe, would be the minority in this competitive, dog-eat-dog atmosphere. Some valued "assets" will be pampered along, even as savvy coaches and trainers will be keeping an eye peeled for potential replacements. I can't address this directly at all, except to say that trainers become very critical in this part of the year; and good scouts and coaches are invaluable. Bottom line, for me, good stretching and limbering exercises, mixed with wind sprints, arm "un-kinking," and timing in the batter's cage.

3. Roster turnover and trades: This brings to mind, my 2004 BOSOX, SS, and Pokey Reese. If there is a potential imminent turnover problem on the horizon at any point at a critical position, being certain your bench [or you prospects] are well drilled and conditioned at all times ... ready to step in when needed. I would think that player psychology is a big factor here. I would imagine that some wise baseball mind had a session with Pokey Reese, advising him, "We don't expect you to try to be Nomar. We want you to play your best Pokey Reese "D" and hold our infield together."
[What a value he was that year!]

4. Here I can go. The whole metric is game evolution oriented, with TEMPLATES and ratings co-dependent upon games changes, including equipment.

5. As to temp extremes, I believe that Items 2 and 5 and perhaps 6 could be combined into a thesis paper on "situational conditioning, adaptation, and warming up ... honing defensive skills and extending careers." [Might make quite a paper].

6. I believe that my metric results show an absolute tendency to defensive slumps [or valleys] in most careers. Managers too quick to yank a valuable player in such a slump are probably shooting themselves in the foot if they don't consider full value return against full value lost by yanking a player. If you bench a .255 hitter who creates 75 batting runs a year - but usually in above average defensively with a replacement who shines with the glove but is .235 [60 runs] with the bat ... you are throwing the baby out with the bathwater.

[brett]

Here is a summary of some of the things I have wondered about over the past 20+ years, or so.
1) How could you measure the effect on defense of pitchers who take too much time between pitches?

I saw a study of college hitters that showed that when the pitcher threw a pitch within 10 seconds versus after more than 15 seconds the batter was 33% LESS likely to identify the pitch properly as a breaking ball versus a fast ball-well actually that a breaker after a fast ball was harder to identify, and vice versa.

[Jackaroo Dave]

In offensive statistics, the big breakthrough was linking individual production to team wins via linear weights or runs created and pythagorean wins. I would like to see a similar link for defense. Maybe it's out there already. People don't hesitate to talk about defensive runs. But I don't hear anything about the linkage. Perhaps I'm not looking in the right places.

[leewileyfan]

The linear link for defense MAY be inherent at the heart of my metric. I'll leave that for you to decide.

I have built the entire metric and its evaluations around a model for each position. Then I assign a DEGREE OF DIFFICULTY to the package of ball-in-play events for that position. I segregate Assists from Put-Outs in the IF, seeing PO as more routine events of lesser value points than Assists, which can be uniquely challenging at every IF position.

I started with the playing atmosphere of 1901: dead ball era; base-to-base strategy; bunts as a crucial skill [and defense against the bunt]; base coverage rotation on balls in play. Having constructed weighted inputs for each position, I then determined the negative [damage] caused by misplays at each position. This is KEY.

As I calculate each player/position, the raw product [net of all input elements] is different fro every other position. The RATING is derived through a conversion formula that presents ALL position ratings in a familiar format [three or four decimal places, resembling the good old fielding percentage].

When I convert RATINGS into DEFENSE RUNS, it is the negative values for errors that makes the conversion possible.

That, I believe, may be the linearity you are looking for.

[brett]

The linear link for defense MAY be inherent at the heart of my metric. I'll leave that for you to decide.

I have built the entire metric and its evaluations around a model for each position. Then I assign a DEGREE OF DIFFICULTY to the package of ball-in-play events for that position. I segregate Assists from Put-Outs in the IF, seeing PO as more routine events of lesser value points than Assists, which can be uniquely challenging at every IF position.

I started with the playing atmosphere of 1901: dead ball era; base-to-base strategy; bunts as a crucial skill [and defense against the bunt]; base coverage rotation on balls in play. Having constructed weighted inputs for each position, I then determined the negative [damage] caused by misplays at each position. This is KEY.

As I calculate each player/position, the raw product [net of all input elements] is different fro every other position. The RATING is derived through a conversion formula that presents ALL position ratings in a familiar format [three or four decimal places, resembling the good old fielding percentage].

When I convert RATINGS into DEFENSE RUNS, it is the negative values for errors that makes the conversion possible.

That, I believe, may be the linearity you are looking for.

You have written that you use errors, putouts, assists, double plays WP and passed balls exclusively, but do you also make an estimate of innings played on defense? If a shortstop has 0 errors, 50 putouts, 100 assists and 20 double plays in 30 games, that might be quite good. If he does it in 100 games it would not be. How do you figure out expected values for a given player?

[brett]

In offensive statistics, the big breakthrough was linking individual production to team wins via linear weights or runs created and pythagorean wins. I would like to see a similar link for defense. Maybe it's out there already. People don't hesitate to talk about defensive runs. But I don't hear anything about the linkage. Perhaps I'm not looking in the right places.

Matt Souders PCA system turned all runs into their effect on pythagorian winning percentage effect that they would have if that player were placed on an otherwise average team.

one oddity though with sabermetrics is that we tend to rate batters on the sum of their batting events: singles, doubles, triples etc, and not at RBIS or runs scored, but sabermetrics tends to look at runs allowed by a pitcher (and modifies that total based on other factors like defense). The only logical reason for this is that pitchers and defenses tend to allow runs NOT LINEARLY and RANDOMLY dependent on the hitting events of the batters faced. A good pitcher DOES actually tend to pitch better situationally-with runners on base for example.

[leewileyfan]

You have written that you use errors, putouts, assists, double plays WP and passed balls exclusively, but do you also make an estimate of innings played on defense? If a shortstop has 0 errors, 50 putouts, 100 assists and 20 double plays in 30 games, that might be quite good. If he does it in 100 games it would not be. How do you figure out expected values for a given player?

Starting with 1901, it was trial and error, beginning with a fair idea of positional distribution of responsibility, opportunity and challenge. From ther on, it was feedback-base, with each modification game in playing conditions [ball liveliness AND new ball turnover, glove design, playing surface] impacting ratings and prompting expectation model change. The essential input values are little changed; but the TEMPLATES of expectation are flexible.

I ran you shortstop for three varying innings played: 50 PO; 100 A; 100 DP; 0 E

500 innings = .965 rating, +14.96 DR > Player X at .925

450 innings = .992 rating = + 25.06 DR

670 innings = .904 rating = - 7.85 DR

NOTE: The fact that he commits NO errors does not yield a 1.000 [or higher rating]. The TEMPLATE [1.000], which can be exceeded by outstanding performance, is set at a level that requires all defensive elements to add to its standard.

[Bothrops Atrox]

It is a tough "sell" to get people interested in a book focusing on a single metric by someone who is not already established. .

That is pretty much what Michale Humphreys did with Wizardry. Even though he was somewhat known in the sabermetric community and those who have bought the book were probably somewhat familiar with him already.

[leewileyfan]

You have written that you use errors, putouts, assists, double plays WP and passed balls exclusively, but do you also make an estimate of innings played on defense? If a shortstop has 0 errors, 50 putouts, 100 assists and 20 double plays in 30 games, that might be quite good. If he does it in 100 games it would not be. How do you figure out expected values for a given player?

Brett, I've just posted an evaluation of top 1B defenders [1910-Present] which may show adaptability of the metric across generations. I posted it on the Evolving MLB Defense thread I started several weeks ago. Feedback welcome.