Originally Posted by woof43

Hi,
The following method is relatively crude, compared to what is possible. (... think of how you can use this technique not only to correct Calls, BUT ALSO BY Class) But I would also continue to think about the ramifications of changing the data domain.

First set yourself up a table of OBSERVED historical probabilities for getting any given call value. Express them as PERFORMANCE PERCENTILES. So in my example, you end up with eight percentile floor cutoffs, like (a made-up example):

Call 1 = 82.2%
Call 2 = 71.1%
Call 3 = 50.7%
Call 4 = 41.3%
Call 5 = 30.8%
Call 6 = 21.7%
Call 7 = 09.5%
Call 8 = 00.0%

Now, for any given formline where you have a horse and he scored a Call of 3, you know he performed AT THE 50.7 PERCENTILE LEVEL OR ABOVE. So throw away his original Call, and substitute his PERCENTILE score. Do that for all of a horses lines. Add up the percentile scores, and divide by the number of lines you used, and you get the horses AVERAGE PERCENTILE SCORE, which you use as a CORRECTED AVERAGE Call

To project the 800 Call on TODAY'S race, simply get the percentile rating for all horses, put them in order, and give the highest horse a projected break of 1. Next highest a projected break of 2, and so on, just using a simple RANKING of the PERCENTILE AVERAGES.

Important Note: To get a more accurate rating for each horse, don't take the FLOOR percentile for that call, but take the mid-point between the floor and the ceiling. For example, if a Horse has a Formline with a call of 3, don't use 50.7 for his score, instead use ((50.7+71.1)/2) which is the midpoint of the percentile step.

The "trick" here is that you are NOT getting a correction for each formline. Instead, you are getting a corrected OVERALL average for the Horse by changing the domain of your data from CALLS to PERCENTILE PERFORMANCE RATINGS.

And finally, and OBVIOUSLY, this general technique is pretty valuable when applied to other problems of this type commonly found in the handicapping world.