The only thing that wasn’t going so well was my attempt at speed handicapping Quarter Horses.  These races made up anywhere from a third, to an entire day’s race card at the Southwestern tracks, but I could not, even with a mountain of tables and charts, make any type of Beyer-like speed figures work.

The really bothersome thing with Quarter Horse figures was that class-par-times—the foundation of all Beyer-like figures—were atrociously inconsistent.  To make matters worse, the times themselves were in decimal seconds (e.g. “17.83”) which for some reason, seemed to make inconsistencies and errors just scream off the page.  I had no problems doing the necessary fudging of figures in “fifths” to create Thoroughbred par times, projected times, etc., for the same tracks, but I guess decimals looked “scientific,” so it didn’t feel right to monkey with them.

As described last week, the Beyers are based on class-par-times.  Horses’ actual times are compared, in an elaborate scheme, to what times should be for horses of the same value class.

I was trying to do something similar for Quarter Horses, but class-par-times simply were not working.  Except in the grossest sense, there seemed to be no correlation between “class” and what time a group of horses would run.

The excuse of having a stand-out horse skew the time in any particular race usually wasn’t there—entire fields were crossing the finish line in “blanket finishes,” with no more than a length or two between 12 finishers spread across the entire width of the track.

$2,500 claimers at Ruidoso Downs would run 440 yards faster than the $5,000 claimers did, 20 minutes later—same surface, no wind—no explanation.  Allowance horses might, or might not, run slower than both.  $8,000 claimers were the most consistent in this mess, and the fastest horses on the course, with the sole exception of Grade I fields.  If you tried to build class-pars by averaging the times of a season of races, they might look like this:
 

Grade 1	21.1
$8000 Claimers	21.2
Allowance > $15 K Purse	21.3
Mdn Spc Wt	21.4
Etc.	Etc.

At Ruidoso in particular, with enormous Grade I purses and teeny-weeny purses for the prep races leading up to them, just figuring out a hierarchy of classes, with the prep races mixed thoroughly into more normal races, was a much bigger trick than doing the same for Thoroughbreds at say, Santa Anita.

For Thoroughbreds, Andy Beyer says, “If the sample is large enough, the par figure for any class will be greater than that of the class below it.  The $6,500 claimers will run faster than the $5,000 claimers, who in turn run faster than the $4,000 claimers,” (Beyer On Speed, p. 25).

The key words are:  “If the sample is large enough.”  But I would have to add:  “And, maybe not then.”

You can often learn something about the “normal” by looking at the extreme.  In this case: Quarter Horses.

Quarter Horse times are much more accurately recorded than Thoroughbreds.  There is no “flying start." Quarter Horses run as fast as they can from the start to the finish line. Thoroughbreds do not.

Since Quarter Horses run as fast as they can, regardless of class, you would expect the class-pars to stair-step even more neatly than Thoroughbreds, which must rate through at least part of a distance.  No Quarter Horse straightaway dash has ever been won by setting a slow pace to throw off the competition.  However, their times do not stair-step neatly.  Quarter Horse class times are just as erratic as Thoroughbreds. 

Since Thoroughbreds do rate, it is possible to have even a Grade I race run in splits that would make a $10,000 claimer blush.  Am I lying?

This is where I have to repeat (for me, if not for you), that I am, unabashedly, Andy Beyer’s biggest fan.  As I go through this process in the next few columns, it does not diminish that fact in any way.  Andrew Beyer is The Man. 

The rub is that it is rare to find a sample that is large enough at any track for any breed, across all classes and across all distances.  The last part becomes “key” also:  across all classes and distances, because the goal is to have speed figures that translate across distances.

By mid-season at many tracks, you may be developing a good stock of 6 furlong dirt race data.  You would probably have a smaller stock of 6.5f dirt races, and darned few 4f or 7.5f races.  But remember, the first step in creating this type of speed figure is to develop a hierarchy of race classes at a particular track.

The class hierarchy is specific to the track.  Maybe on some circuits the classes will mostly match, but the hierarchy at Belmont is going to be significantly different from say, Retama.  It should also be specific to the distance (which I have never seen anyone do), because certain classes of horses run certain distances far more often, while other classes don’t run them at all.  For example, having projected times for Grade I horses at 4.5f at Fonner Park is probably a waste of time and imagination, yet neatly stair-stepping pars will be projected for almost as bizarre non-happenings.

The first little quandary enters when you set up the hierarchy of classes itself.  Here are two hypothetical hierarchies, one for a major track, one for a minor track:
 

Major Track	Minor Track
Grades Stakes	Ungraded Stakes
Ungraded Stakes	5K Clm
100K Clm	NW1
NW3	4K Clm
75K Clm	3.2K Clm
60K Clm	Maiden Spc. Wt.
NW2	2.5K Clm
50K Clm	2K Clm
NW1	Mdn Clm 5K
35K Clm	Mdn Clm 4K
Maiden Spc. Wt.	Mdn Clm 2.5K
25K Clm
18K Clm
14K Clm
Mdn Clm 50K
Mdn Clm 35K
8K Clm

 

How do you set up this ranking?  Obviously, $75,000 claimers should run faster than $60,000 claimers—but do they?  Say the distance in question is 6 furlongs.  You have three races on record this year from 60K claimers and they ran 1:09:3, 1:08:1, and 1:08:4.  You have two races by 75K claimers and they ran 1:09:3 and 1:08:4.  Are 75K claimers as a class faster then 65K claimers?  Of should they be lumped as a class of “75K – 60K claimers”—or is the sample too small to come to any valid conclusion?  This is one of the classic, large/small sample size problems in horse racing.

If you go to a big sample, rolling in several years' worth of data, and take an average of all 6f races run by 75K claimers, according to theory, the average time should be slightly faster than 60K claimers.  However, since 75K and 60K claimers run routinely, day-in-and-day-out within the same range of times, do you have a valid distinction?

How about a less frequently run distance?  How many years of data would you have to accumulate to set up a basic class distinction between Maiden Special Weights running 6.5 furlongs at the minor track above, to be sure it is correct to place them between the 2.5K and 3.2K claimers?

I’d suggest that you would never be able to prove that some of these class placements are valid.  There is too much variation in times run within any of these fine-grained divisions of classes at infrequently run distances to even mathematically prove that the classes exist as clear divisions of speed.

Yet this first step, of defining class divisions, is the foundation of most modern speed figures.  Next week, next step.