Calculating impact values for post positions

[mountainman]

A simple question to which I SHOULD know the answer occured last night. Should the iv of wider posts, let's say the 9- post in this case, be calculated only from races containing 9 or more starters? Or instead calculated by including fields of all sizes before doing the math ( % of 9-post winners divided by % of 9-post starters) ? The iv derived will be different in each case. But should it be?? Or have i missed something obvious and applied faulty math?

[Dave Schwartz]

If you use the Quirin approach to IV's (which was faulty), then you must look at it as field-size specific.

However, if you simply use an adjusted version of the formula, you can use all the races at each distance.

Obviously, the files need to be distance specific.

Thus, you wind up with (say) SA 5.5f, SA6f, SA6.5f, etc.

The formula needs to change from:

Pct of Winners / Pct of Starters

to

Winners / Expected Winners

Where:
Expected Winners is computed as the sum of 1/field for each horse.

[cj]

If you use the Quirin approach to IV's (which was faulty), then you must look at it as field-size specific.

However, if you simply use an adjusted version of the formula, you can use all the races at each distance.

Obviously, the files need to be distance specific.

Thus, you wind up with (say) SA 5.5f, SA6f, SA6.5f, etc.

The formula needs to change from:

Pct of Winners / Pct of Starters

to

Winners / Expected Winners

Where:
Expected Winners is computed as the sum of 1/field for each horse.

This is exactly right. Here is an example mountainman for 6f races at Mountaineer:

FS Ra Pt Ws ExpW IV
1 0 1 393 325.8 1.21
2 0 2 365 325.8 1.12
3 2 3 343 325.8 1.05
4 3 4 306 325.1 0.94
5 51 5 295 324.3 0.91
6 177 6 260 314.1 0.83
7 372 7 265 284.6 0.93
8 490 8 245 231.5 1.06
9 638 9 157 170.3 0.92
10 991 10 97 99.4 0.98


A did omit the 11 and 12 field sizes and posts since there were only 3 races. For your example, the 9 post, you calculate expected wins by adding 1/9 for each race with a field size of 9, and 1/10 for each race with a field size of 10. For the 8 post, you add 1/8 for each race with 8 horses.

[mountainman]

If you use the Quirin approach to IV's (which was faulty), then you must look at it as field-size specific.

However, if you simply use an adjusted version of the formula, you can use all the races at each distance.

Obviously, the files need to be distance specific.

Thus, you wind up with (say) SA 5.5f, SA6f, SA6.5f, etc.

The formula needs to change from:

Pct of Winners / Pct of Starters

to

Winners / Expected Winners

Where:
Expected Winners is computed as the sum of 1/field for each horse.

Tx very much, Dave. I've used quirin's method many times before last night's belated realization that it's seriously flawed-and undershoots correct iv values for wider posts. My profile is, of course, distance specific and encompasses only mnr (some surprising tendencies have been revealed).

One more question: As to the correct formula-winners/ expected winners- Won't i get the same results by tweaking quirin's formula to include
only fields that contain the specific post position in question? In other words, since mnr has 10 starters per- race max, can't i calulate the correct iv on post 8 by restricting the data to fields containing 8, 9 or 10 starters?

[mountainman]

This is exactly right. Here is an example mountainman for 6f races at Mountaineer:

FS Ra Pt Ws ExpW IV
1 0 1 393 325.8 1.21
2 0 2 365 325.8 1.12
3 2 3 343 325.8 1.05
4 3 4 306 325.1 0.94
5 51 5 295 324.3 0.91
6 177 6 260 314.1 0.83
7 372 7 265 284.6 0.93
8 490 8 245 231.5 1.06
9 638 9 157 170.3 0.92
10 991 10 97 99.4 0.98


A did omit the 11 and 12 field sizes and posts since there were only 3 races. For your example, the 9 post, you calculate expected wins by adding 1/9 for each race with a field size of 9, and 1/10 for each race with a field size of 10. For the 8 post, you add 1/8 for each race with 8 horses.

Tx, CJ. What passes for an epiphany with me occured last night when i realized the old quirin formula was flawed. I think i figured a more circuitous, but likewise accurate, way to get the same results as the correct equation you so kindly offered. The profile i'm constructing also encompasses track condition, odds and running style. Having just completed the laborious process of entering data from 2 years worth of charts, i'm eager now to analyze it.

[TrifectaMike]

If you use the Quirin approach to IV's (which was faulty), then you must look at it as field-size specific.

However, if you simply use an adjusted version of the formula, you can use all the races at each distance.

Obviously, the files need to be distance specific.

Thus, you wind up with (say) SA 5.5f, SA6f, SA6.5f, etc.

The formula needs to change from:

Pct of Winners / Pct of Starters

to

Winners / Expected Winners

Where:
Expected Winners is computed as the sum of 1/field for each horse.


The proper way to measure this is by the use of a statistic called the G-Stat. The G-Stat is used to compare Actual Results to Expected Results with the Log Likelihood Ratio.

The G-Stat is given by 2 * (sum-over-i(Oi * ln(Oi / Ei)) where Oi is an observed value and Ei is the corresponding expected value and ln is the log to base e.

The closer the G-Stat is to 0, the closer the actual results are to the expected results; you can look up specific probabilities if necessary.

Mike (Dr Beav)

[cj]

The proper way to measure this is by the use of a statistic called the G-Stat. The G-Stat is used to compare Actual Results to Expected Results with the Log Likelihood Ratio.

The G-Stat is given by 2 * (sum-over-i(Oi * ln(Oi / Ei)) where Oi is an observed value and Ei is the corresponding expected value and ln is the log to base e.

The closer the G-Stat is to 0, the closer the actual results are to the expected results; you can look up specific probabilities if necessary.

Mike (Dr Beav)

I think I'll stick with the simple version, but good luck to you.

[Robert Goren]

The G-statistic is used biology research. It is generally is not considered a good test for small samples. I am not going to into long discussion about the reasons why but "Fisher" is probably a better test here for goodness of fit. That being said, from a practical stand point CJ is sarcastic quip is probably right. The fancier you get in your tests the more you lose sight of goal, making money at this game and the more you get involved with just math of it. I know some of you will find this hard to believe, but statistical math can be a seductive mistress .
You can quit yawning now and get back to picking winners.

[Capper Al]

If you use the Quirin approach to IV's (which was faulty), then you must look at it as field-size specific.

However, if you simply use an adjusted version of the formula, you can use all the races at each distance.

Obviously, the files need to be distance specific.

Thus, you wind up with (say) SA 5.5f, SA6f, SA6.5f, etc.

The formula needs to change from:

Pct of Winners / Pct of Starters

to

Winners / Expected Winners

Where:
Expected Winners is computed as the sum of 1/field for each horse.

Dave,

How does this differ from Quirin?

Thanks

[Dave Schwartz]

Quirin uses percent of winners / percentage of starters.