Why do we still use Impact Values?

[podonne]

Greetings,

A thought ocurred this evening when reading some articles that kept bringing up Impact Values. I'm curious why Impact Values are still so prevalent in this game instead of using Pearson's chi-squared test?

It takes the same basic input variables (albiet defined slightly differently), a number of winners and losers not selected by some factor, and a number of winners and losers selected by some factor, but the result is a well proven stastically relavent measure of probability. Very easy to calculate too.

Impact Values seem like a very crude approximation, and they do not account for the change in overall sample size between the total number of races and the total number of entries selected by the factor.

Just curious. Seems like one of those things that should have died out. Pearson was writing at the end of the 19th century.

Regards,
Podonne

[Robert Goren]

Impact values or index numbers are widely used outside of horse racing. Impact Values can be add and/or multiplied together and often are. Whether this is a good idea or not is another topic. Chi Square numbers numbers do not lend themselves to that kind of manipulation. There was a theory that when computer first became widely available that if you could find enough impact values and manipulated them properly, you could predict the outcome of a race profitably. Books were written on it. Computer programs sold. Stories were told of people who did it. There are still people trying to piece together them like sort of large puzzle. There maybe some people who have done it, I just have never meet one, although I know several who have tried.

[podonne]

Originally Posted by Robert Goren

Impact values or index numbers are widely used outside of horse racing. Impact Values can be add and/or multiplied together and often are. Whether this is a good idea or not is another topic. Chi Square numbers numbers do not lend themselves to that kind of manipulation. There was a theory that when computer first became widely available that if you could find enough impact values and manipulated them properly, you could predict the outcome of a race profitably. Books were written on it. Computer programs sold. Stories were told of people who did it. There are still people trying to piece together them like sort of large puzzle. There maybe some people who have done it, I just have never meet one, although I know several who have tried.

Sorry if this seems like jumping on your response Robert, not my intention, and more of a rant, and much respect for you from your previous posts, but I gotta say I hear those arguments a lot and I don't agree.

I've looked for other disciplines using it, scientific papers, research, and nothing. There is no Wikipedia entry for Impact Value. My statistics textbook and the three data mining books on my shelf don't have the term in their indexes. Google searches for the term "impact value" only come up with handicapping websites, plus a soil sample methodology that isn't the same and a discredited method of estimating the relevance of scientific papers (called "impact factor" and again, calculated differently).

Unless other disciplines use a different name for it, or if IVs started as economic index numbers, trying to show the magnitude of the change from one point to the other based on the previous point being 100 (or 1.0, I guess?) Just trying to think where I might be misunderstanding.

Oh, and I would be cautions of anyone doing math with Impact Values. You can only add them together if the factors are 100% independent, and I don't ever see anyone testing this. I've found that very few factors in this game are truly independent.

[BillW]

Originally Posted by podonne

more of a rant,

This game is mostly about seeing what the public doesn't see and using figs/numbers etc. that the public doesn't use. Rather than rant about the public not using that technique, why not use it yourself and enjoy the benefits of a unique approach to analysis?

[podonne]

Originally Posted by BillW

This game is mostly about seeing what the public doesn't see and using figs/numbers etc. that the public doesn't use. Rather than rant about the public not using that technique, why not use it yourself and enjoy the benefits of a unique approach to analysis?

Well said, Bill. :-) Perhaps hoping that some aspiring horseplayer like I was years ago will pick up this thread and know there is a better way. I think I've just gotten to the point in my own personal evolution where I can defend a principled stance against them (and suggest an alternative, instead of just ranting that they suck).

Perhaps to discredit those making money by selling "high impact value factors" to the unsuspecting. Money spent on touts is money not being spent in pools. We should all want more money being spent in pools. :-) Imagine how much more handle would rise if we weren't paying so much for data and past performances?

[Dave Schwartz]

I think impact values have their place. If you understand Bayesian analysis, you can easily understand that IVs are kind of like the layman's version. Not the same; not as powerful; but in the same direction.

[Red Knave]

Originally Posted by podonne

I'm curious why Impact Values are still so prevalent in this game instead of using Pearson's chi-squared test?

I think Impact Values are simpler to understand. The result of an Impact Value calculation is basically a positive or negative percentage.
Perhaps the better solution is to use both. Calculate an Impact Value and then determine what confidence level we have in it.
As you point out there are few, if any, variables in horse racing that are independent so calculated values, impact or chi-square, are suspect because of this.

[Robert Goren]

Originally Posted by podonne

Sorry if this seems like jumping on your response Robert, not my intention, and more of a rant, and much respect for you from your previous posts, but I gotta say I hear those arguments a lot and I don't agree.

I've looked for other disciplines using it, scientific papers, research, and nothing. There is no Wikipedia entry for Impact Value. My statistics textbook and the three data mining books on my shelf don't have the term in their indexes. Google searches for the term "impact value" only come up with handicapping websites, plus a soil sample methodology that isn't the same and a discredited method of estimating the relevance of scientific papers (called "impact factor" and again, calculated differently).

Unless other disciplines use a different name for it, or if IVs started as economic index numbers, trying to show the magnitude of the change from one point to the other based on the previous point being 100 (or 1.0, I guess?) Just trying to think where I might be misunderstanding.

Oh, and I would be cautions of anyone doing math with Impact Values. You can only add them together if the factors are 100% independent, and I don't ever see anyone testing this. I've found that very few factors in this game are truly independent.

Impact values are nothing more than a handicapping term for index numbers. They figured the same way and are the same thing. Index numbers are pretty commonly used a seasonal adjustments for such things as retail sales and if used properly they can be a useful tool. I am not sold on their use in handicapping.

[davew]

whenever something good comes out, it gets used by many - in some gambling such as stocks and commodities, big crowd and momentum is good - not so with parimutual wagering.

I have heard of stories that the top speed horses would frequently go for double digit wins 50 years ago. Since Beyers came out, the only ones really making money from them are the ones selling the figures.

You should develop a Pearson's chi-squared test methodology that works, keep it to yourself and your computer programer. Run a bot and travel the world spending the money you make.

[Robert Goren]

The problem I have with Chi-Squared is that it will tell if something is relevant or not, but not how relevant it is. I want to how relevant something is if I am going to bet my money because of it.

[podonne]

Originally Posted by Red Knave

Perhaps the better solution is to use both. Calculate an Impact Value and then determine what confidence level we have in it.

Originally Posted by Robert Goren

The problem I have with Chi-Squared is that it will tell if something is relevant or not, but not how relevant it is. I want to how relevant something is if I am going to bet my money because of it.

Good points, both. I wouldn't say that a chi-sq test would be a end-all magic bullet system, (IVs aren't either) but using Impact Values to determine relevance is misleading and that the chi-sq test, or really any test that applies to 2x2 contingency tables, are far better for that specific goal.

I usually see IV listed with ROI, so in that case IV is being used as a relevance measure and paired with ROI to determine whether to invest in it. It might hint at relevance or be similar to concepts used in Bayesian math (as Dave points out), but it doesn't directly measure relevance, it's just a ratio of two percentages without concern for sample sizes.

[fmolf]

whether using iv's or chi squared #'s one still has to figure out when and where they are relevant and which ones fit in which type of race.This then sounds a lot like handicapping withany other method?...no?

[Overlay]

Quirin presented the chi-squared test as a means of determining whether results (which he expressed as impact values) that seemed to be positive or negative were a fluctuation without any particular statistical significance that might not continue into the future, or an independent trend or factor that was not interacting with other variables to produce the outcomes that it was showing, and that could more reliably be counted on to maintain its positive or negative character in subsequent samples. (Of course, Quirin used comparatively small sample sizes.)

I asked Mike Nunamaker at the time that he published Modern Impact Values whether he had subjected his relatively larger data sample to similar testing, and he indicated that the confidence level provided by the increased sample size that he used performed the same function with regard to the durability of the findings and the identification of independent variables as subjecting a smaller sample to chi-squared testing.

[TrifectaMike]

Originally Posted by podonne

Good points, both. I wouldn't say that a chi-sq test would be a end-all magic bullet system, (IVs aren't either) but using Impact Values to determine relevance is misleading and that the chi-sq test, or really any test that applies to 2x2 contingency tables, are far better for that specific goal.

I usually see IV listed with ROI, so in that case IV is being used as a relevance measure and paired with ROI to determine whether to invest in it. It might hint at relevance or be similar to concepts used in Bayesian math (as Dave points out), but it doesn't directly measure relevance, it's just a ratio of two percentages without concern for sample sizes.

Podonne,

You might be interested in this thread

http://www.paceadvantage.com/forum/s...=square+method

It came to an premature close, but you might find some of the answers you are looking for at the end of the thread.

Mike (Dr Beav)

[podonne]

Originally Posted by Overlay

I asked Mike Nunamaker at the time that he published Modern Impact Values whether he had subjected his relatively larger data sample to similar testing, and he indicated that the confidence level provided by the increased sample size that he used performed the same function with regard to the durability of the findings and the identification of independent variables as subjecting a smaller sample to chi-squared testing.

With all due respect to Mr. Nunamaker, the increased sample size is not automatically a panecea. The issue is with the difference in sample sizes between the selections and the entire sample of races. "I've got a lot of data, so I don't need significance testing" is not true.