2-3 Wide on both Turns Equals What??

[Silver Charm]

1-2 points??

I see a lot of good horses with bad posts. And some not so good with insides draws. If its 2 points its worth a shot on some Inside drawn live ones at good odds!!

[TreadHead]

Regardless of the answer to this, it is important to note that the inside at SA for sprinting distances is ABSURDLY negative right now.

At 6F with a sample size of 41 races this meet, posts 1-3 have an impact value of 0.42 (!) and the rail an impact value of 0.18 (!!!).  Conversely, posts 8+ have an impact value of 1.86 (!!!), according to BRIS.  The other sprint distances arent quite as bad, but this is still something to pay attention to and not just blindly look for ground savings.  

Unless I see something drastically different develop in the next couple days, A horse like Justin Phillip is an easy toss.

[Silver Charm]

Thanks for the data. No one intends to go about doing anything blindly. How many BC Winners have come from say Post 10 and out. Paseana rolled from the 13 and Silic was broke from the 12. North East Bound almost got there from the 14 and then there was Pluck. But its not easy so maybe inside middle is a better place to be looking. Look the outside can get there, I had 3 of those I mentioned above. But you need some value on the figures......

[kmart4503]

Can you explain impact value? I\"m not familiar. Thanks.

[TreadHead]

Nothing fancy about it, it is a statistical measure of influence on data, not specific to horse racing.  An impact value of 1.00 implies that the factor you are looking at is not impacting the data in any meaningful way.  Lower than 1.00 it is having a negative impact and higher than 1.00 it is having a positive impact.

IV numbers this extreme are rare in racing over such a large sample size.

[kmart4503]

Thanks

[richiebee]

I wanted to address the notion of \"impact value\" of a given post position just as a general reminder that utilizing statistics sometimes involves negotiating the proverbial \"slippery slope\".

When this thread originally appeared pre - BC, I looked into the notion of the \"impact value\" of a given post position, and wound up reading some interesting back and forth on the Pace Advantage discussion site. Apparently the formula by which \"impact value\" is calculated does not take into account post time odds. By my estimation post time odds should be taken into account in determining whether a given post, over the long term, is favorable or f-kt.

Lets take a hypothetical year of racing at the hypothetical racetrack Winorlose Downs. During that hypothetical year, 100 hypothetical races were run at the distance of 7 furlongs. To really simplify this, lets say that all 100 races had 10 starters.

In those 100 hypothetical races, post 1 produced 2 winners and post 2 produced 5 winners. The other 8 post positions produced between 8 and 20 winners apiece. At first glance, and possibly using the \"impact value\" formula, it might be fair to say that posts 1 and 2 are \"f-kt\".

Now lets say that further analysis reveals that due to the total randomness of the post position draw, the average post time odds of the runners in posts 1 and 2 for this sample was in excess of 35/1. Are posts 1 and 2 really disadvantaged, or are the runners randomly assigned to these posts just slower (much slower) than their competition? The case for the inside two posts being \"f-kt\" is certainly stronger if the off odds of runners from those posts is 10/1 rather than 35/1, no?

I am not discounting the use of statistics, but the use of statistics always seems to call for further statistical review. Certainly, I have formed the opinion over the years that the inside post in chute races on Belmont\'s main track seems to point horses somewhere other than the winners circle and needs to be overcome by horse and rider (see Sweet Reason in this years Frizette).

[FrankD.]

Richie,

Great post and a great example is the thoro pattern; I\'ve seen guys say I\'m betting this horse because he is 40% to pair & 32 % to run a new top. How many horses in the sample pattern were chalk or 40/1, how many were in the slop and not bred for it, disadvantaged posts, track bias etc....

I\'ve seen like many here the hard core \"Math or Science Only guys\" go broke in this game time after time. This crazy game is far from an exact science and we as T-generates have the biggest edge in it IMHO. So we get to make a few bucks on a consistent basis and keep on aggravating ourselves until we get as grumpy as MIFF !!!!

Frank D.

[TreadHead]

Richie your point is well taken, a purely mathematical approach takes all the fun out of handicapping analysis in my book, but I do think that when an impact value becomes this strong that people who aren\'t at least considering it as one of the factors in their quiver are doing themselves a tremendous disservice.

In the hypothetical you gave, do you understand the astronomical odds involved that over a 100 horse sample, the average odds for any one post position would be that high?  You could run 100s of simulations and it would be virtually impossible to have average odds of 35-1 or even 20-1 over a sample size of 100 races.

Sample size is everything when figuring out how strongly to consider a data point.  When that many races are involved and a figure that noticeable develops, it is time to take notice.  I\'m sure it was not lost on you that the sprint results in the BC exactas came in 9-12 and 11-9, and while that is anecdotal, it perfectly fits the narrative and the data.

I completely agree with you that if you are evaluating short-term data, that the odds of the horses involved are important.  When the sample size starts approaching the 50-100 range, the average odds of any post should start to level out making this irrelevant to a win impact value IMO.

[Mathcapper]

Tread, I agree with your assertion that extremes in impact values like the ones you noted in the recent SA sprints should be taken into consideration, but I also agree with Richie and Frank that impact values in their standard form are flawed.
 
Why BRIS still presents them this way I’m not entirely sure, but I have an idea. The concept of impact values was first introduced by a horseplayer/mathematician named Fred Davis back in 1974 in a pamphlet called “Percentages and Probabilities.” It was later popularized by Bill Quirin in his classic “Winning at the Races,” and everyone’s been using it in Quirin’s form ever since.

The problem, like Richie says, is that it doesn’t incorporate the effect of odds. Now Quirin’s no dummy (he’s a math & computer science professor), so he must’ve had a reason for not mentioning it in the book. I’m guessing it’s for just the reason you note – that over large sample sizes, the effect of odds should balance out. But that’s a luxury us horseplayers seldom have. The vast majority of the time, we are dealing with very small sample sizes, so it’s essential to include the effect of odds to avoid biased results that can lead to erroneous conclusions.

The best description I’ve seen on the kind of “odds-based” impact values Richie is talking about was written by Gordon Pine. It can be found in the link below (disregard the second-to-last paragraph – he calculated average odds incorrectly there). Essentially, rather than dividing the % of winners with a given characteristic by the % of starters with a given characteristic, you instead divide the actual win % of horses with a given characteristic by the expected win % of that group based on their odds:

Odds-Based Impact Value = Actual Win% / Expected Win%

I wish data suppliers would start publishing IV using this more informative metric, because it’s pretty useless in it’s current form when dealing with small sample sizes.


Rocky

Impact Value of Questionable Value

P.S. Anyone interested should also check out Gordon’s TTs Archive on the same NetCapper site – he’s written some great stuff, particularly on subjects related to wagering strategy.

[TGJB]

Or... Instead of looking at win %, we could look at what figures they ran compared to their own tops...

[TreadHead]

Do you have time to do such a study for the recent SA meet on 6F races?  Would be very interested in the results

MC, I think we are saying the same thing.  When sample sizes are small, the data should be taken less seriously, and in racing you are almost always dealing with small RELEVANT sample sizes for a given statistic.  For example, knowing Bill Mott\'s win % on turf over 100 races isn\'t of much use when you have to go back 18 months to get to 100 races.

But IMO, when you start approaching 50 races within a specific meet (a tighter timeframe) this is a large enough and compact enough data set to begin balancing out the odds and start to take notice of such a dramatic figure developing.  If it\'s only 10 or 15 races over the course of two weeks, completely agree with your points.

[Breakage]

The better trip on the majority of non-turf courses in North America.

[richiebee]

TGJB Wrote:
-------------------------------------------------------
> Or... Instead of looking at win %, we could look
> at what figures they ran compared to their own
> tops...

This would be interesting, but would there not be an inherent presumption that inside posted horses would be more likely to save ground, and thus be more likely to come close to or equal previous tops?


I have always thought that there would be a lot of interesting statistics which could be derived from looking at the correlation between \"run based\" and \"figure based\" statistics, or even just refining figure based stats. Pre BC, Covel posted that over the years he has learned that not all runners who look like they are set to bounce actually bounce.

This lead me to reflect on Will Take Charge, who after running his career best (negative 0² in the Jim Dandy as the runner up to Palace Malice), looked to me like a prime bounce candidate in the Travers. I felt very strongly about this, as strong of a handicapping conviction as I had all year, that Will Take Charge would bounce like a hybrid combination of a superball, a Pennsy Pinkie and a Spaldeen.

What was behind this costly decision, not to use WTC, who I liked in the Jim Dandy at a price, in the Travers (where WTC was also a decent price)?

First -- WTC\'s Jim Dandy was not just a \"top\", it was significantly faster than his previous career best, a 2² he had run at Oaklawn. I think this is what some T-generates might call a \"big new top\".

Second -- Spacing. The Travers was exactly 4 weeks after the Jim Dandy.

Third -- DW Lukas was I think 1/40 at the Spa at the time he saddled WTC in the Travers, another factor in my thinking that there was a decent chance WTC would  regress in the Travers.

Someone pointed out that WTC\'s Thoro Pattern going into the Travers predicted a pair or another top (he ended up pairing the Jim Dandy negative 0²). The quandary created by Thoro Pattern analysis is this: as far as I can tell, Thoro Pattern does not distinguish between a \"big new top\" (ie WTC improving from 2² to negative 0²) and a mere \"top\" (lets say improving from 2² to 1). Further ciphering is necessary, which is of course what makes the game interesting.

In revisiting my epically wrong prediction of WTC\'s fortunes in the Travers, and prompted by Covel\'s bounce/no bounce pre - BC post, I began to think of a possible statistical inquiry which might be posed to the TG database, one which is figure based, so as not to run further afoul of the TG powers that be, one which might help further determine whether a runner will progress or regress.

That inquiry would involve taking all runners who posted new tops in their last race (\"big new\" or otherwise) and separating them into two groups: (1) runners who posted new tops in a winning effort and (2) runners who posted new tops in a non- winning effort. Arguably, runners from group (1) expended more energy than their group (2) counterparts. Almost certainly, in most overnight races in the US, the group (1) runners will be penalized weight - wise after their winning effort.

So which runner is a bigger bounce candidate: the runner who ran a new top in a winning effort, or a runner who ran a new top but did not win? Would the inquiry produce a clear cut determination? Would a statistical inquiry such as this beget further statistical inquiries?

Probably an inquiry for a time in the future, when, for an exorbitant sum of money (and well worth it) the Thoro - folks will make their database interactive and T-generates can sit at their nook in Living Room Downs and hurl such questions at the database. The DRF folks are asleep at the wheel and probably would not timely object to this interactive database being called \"Brownulator\".

Ah the future. Cue Donald Fagen--

    ...the future looks bright
    on that train all graphite and glitter
    undersea by rail
    ninety minutes from New York to Paris...
   
    a just machine to make big decisions
    programmed by fellows with compassion and vision
    we\'ll be clean when their work is done
    we\'ll be eternally free and eternally young

    what a beautiful world this will be...

[TGJB]

Re your first paragraph-- ground saved or lost would be factored into both back figures and the figures they\'re measured against and so would be of no consequence.