Steven Roman - Dosage Index, Leaving the game

[dilanesp]

Originally Posted by HalvOnHorseracing

Quote me correctly. I said, "as a handicapping criteria, the results of using dosage couldn't have been random because it worked for 60 straight years."

I don't even know what "due to variance" means in relation to the observation, although I didn't say it so I don't have to.

You've only got two choices. Either it is the greatest coincidence racing ever saw, or it was statistically not a random result. You see, in statistics, random means not predictable, not having a recognizable pattern. How do you argue that the same result over that period of time is random?

My sense is that the personal attack (I can't possibly know what I'm talking about because one statement proves I am mathematically illiterate) has become de rigueur at PA. It is certainly easier than coming up with a winning argument. However, if you want to argue the point based on mathematics and statistics, give it your best shot, although you might want to stick to whatever it is you are good at.

It wasn't a big coincidence. Given what the standard deviation of Derby results must be, "60 in a row" backfitting the data is really nothing.

[HalvOnHorseracing]

Originally Posted by pandy

Another theory that has to be considered, is pedigree a legitimate handicapping factor? I used to bet horses that I thought were well bred for turf, if the odds were right, the first time they tried turf. And plenty of them did improve and win. But, most of them didn't win and many didn't improve. It turned out, many of them simply weren't that fast on any surface. The bottom line is, as a handicapping factor, pedigree isn't that reliable.

Intuitively pedigree as a handicapping factor makes sense, but it is still one of many. We all know parents pass down characteristics to their children. On the human side I'd reference the Gronkowski or McCaffery families. The kids didn't get their athletic talents accidentally.

Certain sires seem to have more than their share of precocious progeny. When you see a horse with first time starters winning at 22%, that has to be an important factor despite the fact if you bet the sire blindly you'd lose 78% of the time. I'm fine with the factor being directional.

I referee high school basketball. There are some kids who look like they have their full adult bodies at 17, others who look closer to 15. Same with horses. Some mature earlier than others, and you can often trace that to genetics.

As for turf horses, I've written that pedigree is important - once. What I've more often found is that horses poorly bred for the turf who start first time on the turf and flop may offer value when moved to the dirt. I also think fillies trained by Chad Brown with good turf breeding and pretty much automatic plays first time out. Still, I'd mildly disagree that pedigree isn't important. First time on the turf I'll use pedigree to divide the field into contender/non-contender and then use other handicapping factors to separate the contenders.

What I actually think is more important for turf horses is stride type (simplistically, turf horses pick their feet up slightly more than dirt horses that glide), shoe size (larger feet are better on a turf horse), and turn of speed late. Even so, those characteristics can be passed down from ancestors.

[dilanesp]

One thing to remember is it's entirely possible to do a study of whether Dosage correlates with performance im the Derby. You can even correlate it to last Beyer post time odds, or some other metric. (Note, no study that just looks at winners is valid.)

The fact that nobody has ever produced such a study is VERY telling.

[thaskalos]

Originally Posted by HalvOnHorseracing

Intuitively pedigree as a handicapping factor makes sense, but it is still one of many. We all know parents pass down characteristics to their children. On the human side I'd reference the Gronkowski or McCaffery families. The kids didn't get their athletic talents accidentally.

Certain sires seem to have more than their share of precocious progeny. When you see a horse with first time starters winning at 22%, that has to be an important factor despite the fact if you bet the sire blindly you'd lose 78% of the time. I'm fine with the factor being directional.

I referee high school basketball. There are some kids who look like they have their full adult bodies at 17, others who look closer to 15. Same with horses. Some mature earlier than others, and you can often trace that to genetics.

As for turf horses, I've written that pedigree is important - once. What I've more often found is that horses poorly bred for the turf who start first time on the turf and flop may offer value when moved to the dirt. I also think fillies trained by Chad Brown with good turf breeding and pretty much automatic plays first time out. Still, I'd mildly disagree that pedigree isn't important. First time on the turf I'll use pedigree to divide the field into contender/non-contender and then use other handicapping factors to separate the contenders.

What I actually think is more important for turf horses is stride type (simplistically, turf horses pick their feet up slightly more than dirt horses that glide), shoe size (larger feet are better on a turf horse), and turn of speed late. Even so, those characteristics can be passed down from ancestors.

The esteemed author Mark Cramer RAVED about the effectiveness of the Tomlinson Turf ratings, in his first Kinky Handicapping book. Now...it appears to me that the horse with the LOWEST Tomlinson Turf rating has about the same success probability as the horse with the highest one.

Handicapping "theories" have a short life-span...IMO.

[pandy]

I know I've mentioned this before, but some years ago there was a guy, I can't remember his name, who had a blog where every day he gave out longshot picks in maiden races, based solely on dosage. The criteria was simple, I believe it was 3:50 or higher was a play in sprints as long as the horse was either a first time starter or had only raced once or twice, and anything under 3:50 was a play in routes, same criteria.

He was in the black for while. I bet a few of the horses and hit one that paid $150 in a sprint race. The horse was a first time starter in a 4.5 furlong sprint and he had a dosage rating over 7.00.

[HalvOnHorseracing]

Originally Posted by dilanesp

It wasn't a big coincidence. Given what the standard deviation of Derby results must be, "60 in a row" backfitting the data is really nothing.

For your backfitting theory to hold any water you would have to show that the original methodology used to calculate DI was adjusted to fit the historical data. It wasn't. The same methodology was applied to all data.

If your argument is that there was a conspiracy to select chefs-de-race that would produce results to fit the DI cutoff, again you are simply wrong.

You can offer no evidence that data were manipulated to produce a specific outcome. Examining historical data is done all the time in science to bolster a theory.

As for your "standard deviation of the Derby results," are you sure you know what you are talking about? Without going into the math, first you have to calculate the mean of the data set, which for the sake of argument we'll say is the mean of all Derby winners since a respective date. At that point you can calculate the standard deviation, which would represent the distance from the mean of a respective winner.

The worth of the standard deviation calculation in this case is exactly....zero. The measure of success is not the standard deviation from the mean, but a simple DI value below 4.00. If the mean of the winners was, say, 2.50, perhaps that could be tested to see if the farther from 2.50 a horse is, the higher the probability of losing. Perhaps that might merit a standard deviation calculation, as in "horses within 1 standard deviation of the mean won at a higher rate than horses outside that measure," but that is all irrelevant at this point.

That's twice you've used terms I have to believe you are not experientially familiar with. I'd advise you to give up the ghost here and stick to things where you have expertise.

[HalvOnHorseracing]

Originally Posted by thaskalos

The esteemed author Mark Cramer RAVED about the effectiveness of the Tomlinson Turf ratings, in his first Kinky Handicapping book. Now...it appears to me that the horse with the LOWEST Tomlinson Turf rating has about the same success probability as the horse with the highest one.

Handicapping "theories" have a short life-span...IMO.

I think the issue on Tomlinson would be to calculate how far below the mean of first time turf winners the number had some sort of predictive value. Is a horse with a 382 inherently better than a horse with a 330? It might be that once a horse gets to a certain number, other factors become more relevant. Like I said, the horse with the 330 is more likely to get my money if it is a filly trained by Chad Brown.

The point, and it is the important point, is that IMO handicappers beat mechanical players in the long run.

[thaskalos]

Originally Posted by HalvOnHorseracing

I think the issue on Tomlinson would be to calculate how far below the mean of first time turf winners the number had some sort of predictive value. Is a horse with a 382 inherently better than a horse with a 330? It might be that once a horse gets to a certain number, other factors become more relevant. Like I said, the horse with the 330 is more likely to get my money if it is a filly trained by Chad Brown.

The point, and it is the important point, is that IMO handicappers beat mechanical players in the long run.

I agree. That's why my opinion is that the handicapping authors are committing an INJUSTICE when they try to "systematize" the handicapping process in a simplistic way.

As Einstein famously stated: "Everything must be made as simple as possible...but not simpler."

[Fager Fan]

Dosage, True Nicks, e-nicks, etc, they all use historical data to give a ranking, and then change their ranking when new data supports their prior ranking incorrect. None consider fully the mare (if they consider her, it's her sire they're considering). This is why they're interesting I suppose but hardly something someone should use in handicapping or mating decisions.

[HalvOnHorseracing]

Originally Posted by dilanesp

One thing to remember is it's entirely possible to do a study of whether Dosage correlates with performance im the Derby. You can even correlate it to last Beyer post time odds, or some other metric. (Note, no study that just looks at winners is valid.)

The fact that nobody has ever produced such a study is VERY telling.

Actually, Gramm and Ziemba (the famous Dr. Z) did a study called The Dosage Breeding Theory for Horse Racing Predictions.

https://pdfs.semanticscholar.org/863...365e26c370.pdf

I'm sure you'll have some reason this work isn't valid either.

So if I have a hypothesis that says the winner of the Kentucky Derby will be a horse with a DI under 4.00, and I look at every winner for 60 years, and I find that it does, that study is not valid? Because that was the hypothesis of Roman.

Fascinating. Just fascinating.

[HalvOnHorseracing]

Originally Posted by Fager Fan

Dosage, True Nicks, e-nicks, etc, they all use historical data to give a ranking, and then change their ranking when new data supports their prior ranking incorrect. None consider fully the mare (if they consider her, it's her sire they're considering). This is why they're interesting I suppose but hardly something someone should use in handicapping or mating decisions.

No more insightful than saying you should discard Beyers numbers because they don't inherently account for pace. The value of dosage was directional, not infallibly predictive. And from that perspective it did have value. Like every tool, it only has value if you know how to use it.

[PaceAdvantage]

Posts have been deleted and edited, and a penalty has been assessed to Fager Fan.

I told you guys I wasn't screwing around anymore. You want to act like kids, you get treated as such.

Make it civil in the horse racing section please.

[dilanesp]

Originally Posted by HalvOnHorseracing

Actually, Gramm and Ziemba (the famous Dr. Z) did a study called The Dosage Breeding Theory for Horse Racing Predictions.

https://pdfs.semanticscholar.org/863...365e26c370.pdf

I'm sure you'll have some reason this work isn't valid either.

So if I have a hypothesis that says the winner of the Kentucky Derby will be a horse with a DI under 4.00, and I look at every winner for 60 years, and I find that it does, that study is not valid? Because that was the hypothesis of Roman.

Fascinating. Just fascinating.

Halv:

It's not about winners alone.

Take something that does have validity, Beyer speed figures. Not only do they predict their share of winners, but the second place finisher is likely to have a higher Beyer than the 3rd, the 3rd higher than the 4th, etc.

That's what a handicapping tool looks like when it has statistical validity.

In contrast, tons of "systems" produce lots of "winners" over small samples.

The study that dosage proponents never did was to show that Dosage is predictive of better performance, i.e., that the horses with lower dosage tended to finish better (even if they didn't win) than equivalent horses with higher dosage. It would be pretty easy to construct that study, and since it would be studying all Derby horses, not just winners,,the sample size would be far higher. And yet they never did it. As I said, that speaks volumes about what Dosage really was.

(I should add, I don't buy that there's something special about the Derby. An even better study would include all 1 1/4 mile stakes races. The fact that Dosage never worked particularly well in 1 1/4 mile stakes outside the Derby was also telling.)

[PaceAdvantage]

Originally Posted by dilanesp

The study that dosage proponents never did was to show that Dosage is predictive of better performance, i.e., that the horses with lower dosage tended to finish better (even if they didn't win) than equivalent horses with higher dosage. It would be pretty easy to construct that study, and since it would be studying all Derby horses, not just winners,,the sample size would be far higher. And yet they never did it. As I said, that speaks volumes about what Dosage really was.

I say dilanesp has an interesting point here..but then again, who am I?

[DeltaLover]

Originally Posted by HalvOnHorseracing

No more insightful than saying you should discard Beyers numbers because they don't inherently account for pace. The value of dosage was directional, not infallibly predictive. And from that perspective it did have value. Like every tool, it only has value if you know how to use it.

I have a feeling I'd get accused of arrogance and being a know-it-all by you if I said on a clear day the sky is blue.

Of course, I just consider the source.

I am not sure that I understand how you differentiate between "directional" and "infallibly" predictions. A theory that predicts a "direction" can be infallible or not so the one does not oppose the other.

A prediction is something that can be empirically invalidated and proven wrong. This is exactly what has happened with the Dosage theory. We can say that it possibly started as a scientific proposal, which was eventually degenerated to a pseudo-science since after it was proven wrong, its adherents continued to patch it with several layers of tautologies trying to fit reality by the addition of ad-hoc restrictions.