'Significant' Impact Value

[podonne]

For those of you who use impact value analysis, is there a certain level that you would consider 'significant' enough to use in a system.

I understand IVs to mean that 1.0 shows that the value has no impact, i.e. the distribution if winners in the sample roughly equals the distribution of runners. So I wonder how far above or below 1.0 the impact value would have to go to show a significant effect.

Or, if anyone is reluctant to go on record with an exact figure, a method for determining it would be appreciated. I don't think the standard statistical methods are applicable.

[Overlay]

Quirin detailed the procedures he used for determining a statistically significant deviation from the number of expected winners (either positive or negative) in Appendix A (pages 293-300) of Winning at the Races. I believe that the standard he used was any value that fell above 2.5 (for positive factors) or below -2.5 (for negative factors) for the equation:

(Actual Number of Winners (NW) - Expected Number of Winners (EW)) divided by the square root of (EW (1 - EW/Number of Horses (NH)).

I once asked Mike Nunamaker whether he had applied similar tests to the findings that he came up with in Modern Impact Values. He said that he had not, but that the large sample size upon which his findings were based performed the same function and provided the same assurance as the statistical testing that Quirin conducted on his smaller sample sizes.

I'm not sure that you can draw a hard-and-fast correlation between the results of those tests, and any particular level of impact value. I would say that the tests are the determinative factor. A particular handicapping variable may have a relatively high or low impact value, yet still not fall outside of that -2.5 to +2.5 spread.

The application issue may be that you are likely to be using multiple IVs for horses in any one race. The errors of taking say the cube root of three IVs multiplied together are more significant than any error for one IV but very difficult to quantify race by race.
If you are using multiple IVs, then using any not greater than say 1.20 (or 0.8 on negative side) do not add too much when you then take the roots of the multiples.

[Overlay]

Originally Posted by podonne

For those of you who use impact value analysis, is there a certain level that you would consider 'significant' enough to use in a system.

I understand IVs to mean that 1.0 shows that the value has no impact, i.e. the distribution if winners in the sample roughly equals the distribution of runners. So I wonder how far above or below 1.0 the impact value would have to go to show a significant effect.

Or, if anyone is reluctant to go on record with an exact figure, a method for determining it would be appreciated. I don't think the standard statistical methods are applicable.

To expand on my previous comment, the objectives that I try to achieve in selecting and using impact values are:

(1) to the greatest extent possible, choosing factors that qualify as independent variables/significant values (if they are single-point factors); or that have independent/significant values at the top and (preferably)/or bottom of their impact-value ranges, and that have a smooth progression of impact values from top to bottom (if they are a ranking-type factor);

(2) avoiding redundancy of variables within the major handicapping categories (speed, pace, condition, class, etc), or dependent relationships among the categories, that would distort the accuracy of blending them together;

(3) if deciding between two ranking factors that measure the same handicapping element, preferring the one with the greatest range or spread from top to bottom; and

(4) when dealing with single-point factors, concentrating on significantly positive variables rather than significantly negative ones, since the positive variables have demonstrated their ability to favorably influence the horse's winning chances, regardless of what negative aspects the horse may have contained in its record.

[ranchwest]

I'd rather know 1,000 positive indicators and have a general notion of how strong each was than to have 3 positive indicators that I knew to be exactly a very strong indicator but with no knowledge of the stratification of those indicators.

Many people try to look at data in an arbitrarily exact way. Most data is very stratified. Do you look at the totals or do you break it out by each combination of field size, number of horses in the race matching the criteria, type of race, age of the horses, distance, surface, track or circuit (including the quality of racing there), etc., etc., etc.

What about negatives? What is a "showstopper"? What is the relationship between showstoppers and positive factors?

What about combinations of positive factors? Are there ever combinations of positive factors that are negative?

The truth of the matter is that computers aren't really telling us much when we examine isolated factors such as horses carrying over 114 pounds. Many people who do not use computers assume that computers are incapable of making determinations beyond such a simple level. Generally, computer handicapping is probably still in its infancy, but has already come a long way from 20 years ago.