FANALYTICS: Tweaking the delicate balance

The models and strategies we've developed here at are driven by the statistical environment in which our game lives. As this environment changes, we have to adjust.

Last year, Ray Murphy wrote about the changes we made to the filters for the LIMA Plan. This month, I have been looking at some of the other models we rely on here.

Mayberry Method

For the individual category scores to work, the talent levels they represent need to be spread across a bell curve, more or less. It doesn't work if everyone scores a 5 in a skill level, or the score skews low towards zero. For instance, these are how the xERA scores are spread:

xERA  Score  % players
====  =====  =========
9999    0       9%	
4.80    1      12%	
4.40    2      23%	
4.00    3      30%	
3.60    4      16%	
3.20    5       9%	

Although these skew slightly to the high side, you can see that we have maintained a fairly balanced bell curve here. We can't say the same with the K/9 category in these high-strikeout days:

 K/9  Score  % players
====  =====  =========
 0.0    0       6%	
 5.0    1      13%	
 6.0    2      22%	
 7.0    3      23%	
 8.0    4      18%	
 9.0    5      17%	

This distribution skews very heavily toward the high end, yielding nearly three times as many 5s as 0s, and nearly twice as many 4/5s as 0/1s.

So we needed to adjust the K/9 thresholds at each level. The new table for K/9 will be:

 K/9  Score  % players
====  =====  =========
 0.0    0      10%	
 5.4    1      14%	
 6.4    2      24%	
 7.4    3      23%	
 8.4    4      17%	
 9.4    5      12%	

This distribution still skews a bit high, but by about the same variance as the xERA table. More important, this is a necessary improvement over the current distribution.

Reliability Grades

Tweaking the Reliability Grades is a bit more problematic. Here, there is far less of an expectation of a normal distribution of events. For instance, for the Health grade, we can't assume that the number of Grade A healthy players would be roughly equivalent to the number of Grade F disabled list denizens. These grades are driven purely by the number of major league disabled list days accumulated over the past three years (with more recent stays weighted heavier).

On the batting side, the distribution of grades makes sense to me. But the distribution on the pitching side seems far out of whack:

Grade   % players
=====   =========
  A        60%
  B        12%
  C         7%
  D         6%
  F        15%

A reverse bell curve actually works in this instance. You have a large pool of relatively healthy players, a large pool of perpetually injury-prone players and then a smattering of everyone else in between. But these percentages still seem way too skewed, especially given the vast number of DL stays being accumulated these days.

There is a large group of minor leaguers in our database -- these players are automatically assigned an A grade—but they skew these results by only 9%. That still leaves the A-graders accounting for more than half of all pitchers:

HEALTH (without minor leaguers)
Grade   % players
=====   =========
  A        51%
  B        15%
  C         9%
  D         7%
  F        18%

Still, I could buy into this if we were evaluating health over one single season. Yes, about half of all pitchers get hurt each year. But these grades cover the equivalent of three seasons. Those A-graders should be far less.

I would have expected the distribution to be somewhat more evenly spread. But even the Consistency table is a reality that I can buy into:

Grade   % players
=====   =========
  A        32%
  B        23%
  C        10%
  D         6%
  F        29%

I can accept that a vast majority of players (71%) display an improvement in consistency, leaving a large pool of F-level pitchers with highly volatile performance, year to year. Actually, those Fs also include all the players with less than two full years experience (and thus, have not established a trackable level of consistency), including the minor leaguers in our database. I won't be touching these.

The Experience grade is at first glance confounding:

Grade   % players
=====   =========
  A        16%
  B         9%
  C        21%
  D        30%
  F        24%

But it makes sense when you consider that the player pool is, in fact, getting younger. The A-graders are your long-termers who have been around forever. I might have expected a few more Bs but that might be a short-term anomaly. The Cs and Ds represent the huge recent influx of 20somethings. And like the Consistency grades above, the Fs are once again weighed down by the minor leaguers in our database. I won't be touching these either.

But the Health grades, those need some attention.

I've adjusted the DL-days break points and now we're looking at this:

HEALTH (without minor leaguers)
Grade   % players
=====   =========
  A        40%
  B        21%
  C        10%
  D         6%
  F        23%

A lot of this is just working the numbers enough to pass the smell test, and frankly, the above is barely on the line for me. With these break points, a pitcher drops to a B for anything more than one 15-day DL stint over three years. Previously, it took the equivalent of two full 15-day DL stints to merit a B. Not sure I agree with that but let's roll with it for now.

Note that all these changes will be incorporated into the 2014 Baseball Forecaster this winter, and into our features here next year.

September Leagues

Registrations for a free team (or two, or five) in the September one-month leagues will open on Tuesday, August 27. Deadline is Sunday, September 1. Head over to for complete details.


Click here to subscribe

  For more information about the terms used in this article, see our Glossary Primer.