Modeling

[57viking]

From last nights discussion and the data set , I calculated the hit rate for the exacta and trifecta using the top 5 RX3 horses. I did not eliminate maiden races. How to read the table.
The 3 characters are for WIN Place Show.

Y = Horse was in the Top 5 for that position
N = Horse was not in Top 5 for that position
NA = either First time starter , European Import without data , or data missing . I need to do more work to split this out .

Examples - YYY would mean the trifecta was hit
YYN would mean the exacta was hit but not the trifecta
YNAY would mean Win and Show were hit , but there was an issue with the place horse data. I can cut the data in many ways if there is interest.

[57viking]

I did further work on the dataset. Thanks to whoever put it together.

First I corrected the data issues from above. There were about 125 in total. The largest issue was when a race had two different race ids. ( Example MVR 2/11 Races 2-9) I used the race ids to group lines into race to model the exacta and trifectas. This was about 60% of the errors. The second largest issue was the coupled entries which I saw Ted corrected going forward.

Ted , the other issue I saw was for dead heats. If there was a dead heat for 3rd, both horses showed as finishing 4th. ( example 4/9 TAM Race 8 or 3/5 PRX Race 8 )

Using the updated dataset, I reran the exacta and trifecta numbers for Rx3 Top along with some cuts of the data.

[Bill Lyster]

Great work 57Viking, when I did the first 300-350 sprint races I came up with 80.5% exacta and I know I made mistakes - eye strain, etc.

Great, thanks for being a valuable asset.

Bill

[Bill Lyster]

If you look at the data presented you will see that I stopped bracketing or iterating the data when I hit a high 90% value of CBL at the various calls. There were a couple of 95.5% all the way up to almost 98%. I asked the db for max values for most of the work, just to see how far back the worst winner, place or show horse was. This is not a justification and if someone wants to live by the one who was farthest back, so be it.

On the other end of this study, which has not been pursued yet, is what are the cutoff values for 0-0 horses that do not finish as a 0 horse or 0-0-0 horses and their %Vdc values. Just eyeballing some of the data and from memory a 0-0 horse that ends up in the Final column with a value over 3.5 (I think Tim G posted this last year some time.) is hard to keep as the winner, and a triple zero horse would have a % Vdc value whose limits I have not also explored.

For instance, when I did dirt and turf routes the farthest back on dirt was 15.8-10.8-12.8 at each of the three calls and for turf was 20-11-10.8. The chart I put up for dirt was 10-8-6 and for turf was 11-8-5. But even these numbers are a little misleading because I asked for the max value, INDEPENDENTLY, of each variable. That does not mean in the farthest back case that the horse who was 15.8 at F1 was also 10.8 at SC, nor does it mean that a horse that was in the 96.9% range was 10 back at F1 and 8 back at SC. In most cases the horses furthest back need to make up more ground in the ensuing fractions to be a win candidate. These are stand alone values not linked to any other, its just a statistic. And the relationship of how much gain is needed has not been explored either.

When a horse loses ground between calls, we have not investigated what other existing RDSS factors might tell us the same thing. Like TT or HID from the primary screen. Its another way and could be confirmed by other factors already in the program.

Depending on the race, the dirt horse that was 15.8 back at F1 could have been #1 Vdc, or RX3 or any other variable. But it also depends on the mix of running styles of the rest of the field. If you found a reason to delete the original 0.0 horse at F1 the corresponding F1 value for the winner would be reduced by the difference between the original F1 lead and the 2nd one.

Some times it’s a small fractional number but it can be several if the initial F1 leader had a really high %med. Which others would excuse because of their track model. No problem, it works for them. This is just another way to interpret the data. Also the 15.8 horse would not necessarily be the #1 LPR in that days race.


54235

54236