The most intriguing statistic from Modern Impact Values

[Bill Cullen]

The most intriguing statistic from Modern Impact Values in my opinion was the finding that horses running in a maiden dirt route today that have had exactly two workouts since there last race win over 15% of their races with an impact value of 1.44 and a positive ROI of 6%. The sample of horses showing that characteristic numbered 446. I would have tossed this as an anomaly if it wasn't for the fact that horses showing 3 works since their last race and entered today in a maiden dirt route show over 13% winners with an impact value of 1.25 and a higher-than-expected minus ROI of 14%.

Any thoughts, anyone?

Thanks,

Bill Cullen

[formula_2002]

Bill, with respect to horse racing, impact values mean diddley when attempting to forecast races at a profit.

I use All-Ways Bris data file and have well over 250,000 horses in my data base.

All-Ways appears to do an excellet job using impact values to find winners AFTER, the races have been ran, but then so does anyone.

I started a thread earlier today regaing this very subject and I will add to it with emperical data as I can.

Joe M

.

[Derek2U]

What a lame statistic. It has NO meaning in horse racing **
as far as PREDICTION goes. It's a good descriptive stat but that's
it. IMO.

[Bill Cullen]

Originally posted by Derek2U
What a lame statistic. It has NO meaning in horse racing **
as far as PREDICTION goes. It's a good descriptive stat but that's
it. IMO.

I'm curious whether anyone else has obtained similar results that Nunamaker found or whether it's just an aberration.

Bill Cullen

[formula_2002]

example. Belmont. Impact value results determined using races prior to May,10,2002

433 dirt sprints 5% profit

333 dirt sprints after the above date, 9% loss


240 dirt routs -11% loss compared to 194 dirt routs for a 2% profit.


.

[Bill Cullen]

Originally posted by formula_2002
example. Belmont. Impact value results determined using races prior to May,10,2002

433 dirt sprints 5% profit

333 dirt sprints after the above date, 9% loss


240 dirt routs -11% loss compared to 194 dirt routs for a 2% profit.


.

There was more suggestive evidence than just the single statistic: their were 5 discrete categories which were 0 workouts, 1 workout, 2 workouts, 3 workouts, 4 or more workouts. The impact values increased twice, peaked, then decreased twice. The ROIs increased twice, peaked, then decreased twice in an exact corresponding pattern.

That's prima facie evidence of a frequency distribution not entirely ordered by chance. It might well be chance that caused all those things, but until there's additional data to reject or further strengthen the hypothesis, it's still an open issue in my book.

Thanks for your comments.

Bill Cullen

[formula_2002]

Originally posted by Bill Cullen
, it's still an open issue in my book.

Thanks for your comments.

Bill Cullen

Bill, my book has beem closed for some time. I'm making a killing with my data...by not betting.

Good Luck

Joe M

[NoDayJob]

Originally posted by formula_2002
Bill, with respect to horse racing, impact values mean diddley when attempting to forecast races at a profit.

I use All-Ways Bris data file and have well over 250,000 horses in my data base.

All-Ways appears to do an excellet job using impact values to find winners AFTER, the races have been ran, but then so does anyone.

I started a thread earlier today regaing this very subject and I will add to it with emperical data as I can.

Joe M

.

Have you done any of your impact values and other seaches in conjunction with purse level ranges?

NDJ

[Bill Cullen]

Originally posted by NoDayJob
Have you done any of your impact values and other seaches in conjunction with purse level ranges?

NDJ

No

[ranchwest]

Originally posted by formula_2002
Bill, with respect to horse racing, impact values mean diddley when attempting to forecast races at a profit.

I use All-Ways Bris data file and have well over 250,000 horses in my data base.

All-Ways appears to do an excellet job using impact values to find winners AFTER, the races have been ran, but then so does anyone.

I started a thread earlier today regaing this very subject and I will add to it with emperical data as I can.

Joe M

.

You make an incorrect assumption.

Impact values can be useful or not.

If there is truly a correlation, a high impact value with a positive ROI means that winnings will be consistent.

The problem comes about when data is thrown together willy-nilly without determining correlation.

[GameTheory]

Also, impact values measure performance of a factor in terms of winning/losing, not making money. A factor with an impact value below 1.0 can have a positive ROI, and vice-versa. The ROI can change while the impact value stays the same. Impact values are about forecasting who will win, not what the odds will be...

[formula_2002]

Originally posted by NoDayJob
Have you done any of your impact values and other seaches in conjunction with purse level ranges?

NDJ

NDJ, I have performed the analysis for all of the over 80 factors provided in the All_ways data file., including purse values...

I have found it impossible to consistently beat the mass of public opinion and the track take-out.

[BIG HIT]

RPM sold a system like that it came with a whole meet record.
Wish could remenber who he was anyway he should a profit.I think avg price of winner $9.60.And the last wk had to be with in 14 days of today's race i think

[formula_2002]

Originally posted by ranchwest
You make an incorrect assumption.


If there is truly a correlation, a high impact value with a positive ROI means that winnings will be consistent.

The problem comes about when data is thrown together willy-nilly without determining correlation.

The All-ways factors are highly correlated to the odds and, by definition, the win percentage. The top ranked factors win more then the second ranked.... etc., and the average odds for the top ranked are shorter then the second ranked..etc.

In the example I posed I used a combination of but three highly correlated factors. Bris Speed, Bris Class and Bris Power


Joe M

[Bill Cullen]

Originally posted by BIG HIT
RPM sold a system like that it came with a whole meet record.
Wish could remenber who he was anyway he should a profit.I think avg price of winner $9.60.And the last wk had to be with in 14 days of today's race i think

That's very interesting, Now I really wish someone could test this out.

Thanks for your input.

Bill

[Bill Cullen]

Originally posted by Bill Cullen
That's very interesting, Now I really wish someone could test this out.

Thanks for your input.

Bill

I forgot to add that, if I hadn't been aware that Mike Nunamaker research uncovered the statistic and someone tried to sell me a sytem based on the single factor of a horse having two works sionce its last race and today's race was a maiden dirt route, I'd be extremely sceptical.

Bill Cullen

[thelyingthief]

i recently found an IV that draws a bullseye around bob baffert's two year old starters. if bob or any of his stable hands are seen or are rumored to have been seen plunging their RIGHT index finger in their nose prior to, or shortly after, a race on the third day of a new moon, then today's first time starter in a sprint, so long as baffert's horse is a gelding, was a gelding, or wishes he was a gelding, will definitely win, place or show, or maybe beat some other horse almost 100% of the time. the i.v. was 5.97! the roi, well, that was a delightful -5.97.

i have lately improved dramatically on this roi by correlating said nose picking with any race wherein all the entries will be running on three legs or less except, of course, for bob's; here i have found an iv that approaches a positively infinite asymptote, and the profit is stunning. unfortuntely, one must wait a very long time for this particular scenario to develop. i hope you'll all wait with me for this grand wagering oppurtunity to arise.

any time i see an iv that specifies something akin to "2 works only" specified above, i immediately think of bob, his stable hands, and the use of index fingers for plunging purposes.

thanks for your support.

[InsideThePylons-MW]

:D

[ranchwest]

Originally posted by formula_2002
The All-ways factors are highly correlated to the odds and, by definition, the win percentage. The top ranked factors win more then the second ranked.... etc., and the average odds for the top ranked are shorter then the second ranked..etc.

In the example I posed I used a combination of but three highly correlated factors. Bris Speed, Bris Class and Bris Power


Joe M

Had you established that there was an ROI greater than 1?

[MikeH]

Ranchwest -

I used Allways for 3+ years and, yes, there were positive ROIs occasionally. Some were due to the inclusion of a very large longshot; for those that weren't due to the longshot, the ROI would disappear quickly.

The problem that I had with Allways: it was difficult to look at the underlying data to see if there was a longshot present. However, studying ANY ROI with Impact values will give you a very good idea of what factors the Public is using, and which ones they are overlooking.

For example, there is a very strong assumption that the DelMar turf course is unfavorable to early speed. Over the past several years, EP types have been profitable, but they are hard to find. Reason: the trainers read the DRF also and don't enter E & EP types. One year that I looked at, there were only 10 good E/EP types entered all meet on the grass.

[ranchwest]

In my opinion, it is imperative that you have the win pct, avg win payoff and the ROI, especially if you can't see the underlying data. The IV along with those would help, too.

The average win of all races is somewhere around $12 on win wagers, I believe. If you can hit for anywhere near a reasonable percentage with that high of payoff, you need to live close to the bank.

If you get below $5 average payoff you'll have a heck of a time hitting a high enough percentage to turn a profit. That low of payoff requires an extremely high percentage. If you're down in that zone, you'll have days when you hit a high percentage and still lose.

You have to find your comfort zone.

[Tom]

Originally posted by formula_2002
Bill, my book has beem closed for some time. I'm making a killing with my data...by not betting.

Good Luck

Joe M

Your book or your mind? :confused: :D

[hurrikane]

Touche' :D

[Charlie Judge]

There is a nice article by Gordon Pine at:

http://www.netcapper.com/TrackTractsArchive/TT010223.htm

which looks at IV compared with the odds to determine if the iv is valid or not.

I personally do not use iv in my calculations, but I use a procedure similar to Gordon's idea. in constructing trainer stats.

For example: Trainer Bill Mott off a 30 day layoff. The average odds of these horses was 3.11/1. This translates to a win probability of .199 corrected for take.
Mott's actual win% after a 30 day layoff is 21% then the freshening can be looked at as positive, with a 1.1% higher than expected win probability.

Reporting a stat in isolation does not help you understand if the angle is beneficial. If Bill Mott's horses were to have an overall win % of 21%, then the trainer angle ' off a 30 day layoff' win 21% doesn't help you. I agree with Gordon that it is much more important to use as a standard of comparison the win% of similar horses rather than random expectation alone as is prescribed in IV theory.

Sample size is important in these calculations as well. The chi-square statistic helps you determine if an observed difference can be attributed to random variation or if it is statistically significant.

The Chi square statistic looks at the squared value of (observed wins - expected wins) divided by the expected wins. This number is then compared to a table and if the chi square stat is higher than the table value, the statistic is significant.

A recent tabulation of Bill Mott's stats for the off the layoff category shows 369 starts with 78 wins at an average odds of 3.11 . Converting the win odds to probabilities and correcting for track take, we get .199 win probability based on the odds.

Assuming the crown is a perfect oddsline maker, the expected wins are 369 starts * .199 = 73.6 or rounding up 74 wins.
The actual wins are 78.

The chi square is

( observed-expected)^2/expected or

(78-74)^2/74 = .216


Going to the tables, the chi square for 95% probability is .00393, so the observed number of wins is statistically significantly higher than expected

So, if you are going to valid stats on any measure in horseracing, you must have a valid base for comparison and a large enough sample size to be sure you are seeing a true effect, not one which could be explained by random variation.

The discussion of the chi-square statistic may be found in any standard statistics text.



CAJ

[formula_2002]

caj.. And what happened when you tested this paradigm on the out of sample races?

[Charlie Judge]

There is a major error in the above post... This is what happens when you post a complicated message at 1:30 am.

I entered the data from the chi squared table looking for 95% probability of success, or 5% probability of error as .00393. It turns out I entered the table backwards and gave the chi square for 5% probability of success.

The actual chi-square statistic for 95% probability of success with one degree of freedom is 3.841.

Recalculating, the observed-expected difference would need to be 16 in order to reach statistical significance at the 95% level (5% chance of error)

Since the observed-expected difference is only 4 the effect of a layoff on Mott's horses is not significant.

The reminder of the above post is correct.

The point I was trying to make is that when calculating any factor's influence in racing,you need to look at the number of starts, the expected wins for the group as a defined group , and the difference between observed and expected number of wins based on the chi square stat before you can claim that the factor is important.


I will stop posting when I am half asleep in the future.

CAJ

[formula_2002]

Just for the record let me repeat my often posted comment.
Any analysis must be made on the smallest incremental odds range.

The smaller the range the more accurate the analysis.

Analysis based on average odds, as many of us have mentioned here , is all but meaningless.

[ranchwest]

Originally posted by formula_2002
Just for the record let me repeat my often posted comment.
Any analysis must be made on the smallest incremental odds range.

The smaller the range the more accurate the analysis.

Analysis based on average odds, as many of us have mentioned here , is all but meaningless.

What is your basis for this statement?

[NoDayJob]

Originally posted by thelyingthief


any time i see an iv that specifies something akin to "2 works only" specified above, i immediately think of bob, his stable hands, and the use of index fingers for plunging purposes.

thanks for your support.

Ah, then I see you have discovered that IV means IMPROBABLE VALUE, too?

NDJ

[formula_2002]

Originally posted by ranchwest
What is your basis for this statement?

Perhaps one of the reports on my web page would help explain my basis. Look at the report "Odds Ranges"

http://globalwinningpicks.homestead.com/GLOBALWINNINGPICKSX.html

[Secretariat]

Has anyone come up with a measurement that units IV with ROI?

I know Gordon Pine was working on an AE factor? I was wondering if multiplying IV times ROI might merit an interesting statistical measurement.

For example:

An IV of 0.9 times an ROI of 1.1 = .99
An IV of 1.2 times an ROI of 1.0 = 1.2

I know a guy who multiplies the win% times ROI and says it provides a good overall perceptive of the strength of uniting the factors. A lot of good brains on this board with a lot of databases. I just thought it would save me some testing.

[ranchwest]

Originally posted by formula_2002
Perhaps one of the reports on my web page would help explain my basis. Look at the report "Odds Ranges"

http://globalwinningpicks.homestead.com/GLOBALWINNINGPICKSX.html

This tells me nothing new.

[formula_2002]

Originally posted by ranchwest
This tells me nothing new.

then you know more then you think;)

[ranchwest]

Originally posted by formula_2002
then you know more then you think;)

Ah, your responses are very consistent.

[formula_2002]

Ranch, the average win % for the out of sample system plays was
28% and a dollar loss of 8%.

Now look at each odds line. You will find hugh differences in the win % and p/l% in the incremental odds lines when compared to the average.

As one of the All-State commercilsa goes.." I thought that ment something to you"...

[ranchwest]

Originally posted by formula_2002
Ranch, the average win % for the out of sample system plays was
28% and a dollar loss of 8%.

Now look at each odds line. You will find hugh differences in the win % and p/l% in the incremental odds lines when compared to the average.

As one of the All-State commercilsa goes.." I thought that ment something to you"...

You have the wrong insurance company, but since I'm about to become a Farmers agent, we won't go there. lol

I still have no idea of how you suggest linking odds ranges to other handicapping considerations or why.

[formula_2002]

Originally posted by ranchwest

I still have no idea of how you suggest linking odds ranges to other handicapping considerations or why.

You want to know the odds because its part of the edge formula.

edge= (win% X odds)-(1-win%)

the more accurate the odds range, the more accurate the edge calculation.

Joe M

[ranchwest]

Originally posted by formula_2002
You want to know the odds because its part of the edge formula.

edge= (win% X odds)-(1-win%)

the more accurate the odds range, the more accurate the edge calculation.

Joe M

So, your earlier statement was that the only way to handicap was to examine the smallest possible odds range.

Does this mean that if you have extremely strong indications that a horse will win and its odds are above the miniscule optimal odds range, then you should not wager on that horse?

[formula_2002]

Originally posted by ranchwest
[B]So, your earlier statement was that the only way to handicap was to examine the smallest possible odds range.
[b]

The only way to ANALIZE the RESULTS of handicapping is to examine the smallest possible odds range.

[ranchwest]

Originally posted by formula_2002
The only way to ANALIZE the RESULTS of handicapping is to examine the smallest possible odds range.

And that accomplishes....

[Bill Cullen]

I think it speaks volumes when some folks race to prove a system or factor wrong, but then speak strongly against a factor or system without offering any empirical proof.

Is silence to be construed as golden in such situations?

Bill Cullen

[formula_2002]

Originally posted by ranchwest
And that accomplishes....

Well for a start , I would avoid playing any system pick that went off at less then 7-1.

(system pick as mentiion on my web page.."odds range")http://globalwinningpicks.homestead.com/GLOBALWINNINGPICKSX.html


A REPORT ON A THREE YEAR STUDY, COVERING OVER 170,000 HORESES.

[ranchwest]

Originally posted by formula_2002
Well for a start , I would avoid playing any system pick that went off at less then 7-1.

(system pick as mentiion on my web page.."odds range")http://globalwinningpicks.homestead.com/GLOBALWINNINGPICKSX.html


A REPORT ON A THREE YEAR STUDY, COVERING OVER 170,000 HORESES.

So, the odds range that needs to be included in any computer analysis is greater than 7-1? So, let me see if I understand. Everything else is 7-1 or under, so it is the other odds range.

So, your smallest possible odds range turns out to be over or under 7-1. That's about what I figured to begin with, just wanted confirmation.

[formula_2002]

Originally posted by ranchwest
So, your smallest possible odds range turns out to be over or under 7-1. That's about what I figured to begin with, just wanted confirmation.


Ranch,,, too much work for me to get you on my "page"..Sorry.

Good Luck

Joe M

[ranchwest]

Originally posted by formula_2002
Ranch,,, too much work for me to get you on my "page"..Sorry.

Good Luck

Joe M

That, too, is what I thought.

[Bill Cullen]

Some folks took a lot of pot shots but no one disproved Nunamaker's basic finding that started this thread.

Bill C