Elimination methods for HTR4MSA

[Ken Massa]

I like to work on elimination methods in my testing. Here is a good one to try with HTR4MSA, and you can use the general idea for other data.

If working with data on class/speed figs, A/P or other combination rankings - I have found a solid method of elimination with Fr1. Simple enough - eliminate the bottom 25% of the field as ranked in Fr1. This can be done in the db.

For example, suppose I'm testing ACL in male, non-maiden dirt races where it works well. A field of 8 is tested and the top ranked ACL horse is also ranked -7th in Fr1. That horse ranks in the bottom 25% of the field (7th of 8) in Fr1 and is eliminated from the test. The results may improve considerably by tossing these Fr1 duds. Or test the eliminated horses by themselves to see how poorly they do.

To get the low 25% of Fr1 ranked horses from HTR4MSA - you need to use two fields: nFLD (field size), and rFr1 (fraction 1 rank). Use the following query or VB code in the db to find the bottom 25% of Fr1 rankings =

rFR1 / nFLD < .75 ---> query to find the top 75% ranked only
rFr1 / nFLD >= .75 ---> query to find the bottom 25% rank Fr1

VB code to test #1 ACL combined with a top 75% FR1 (eliminates the bottom 25%) rank =
where x = horse being tested

IF rACL(x) = 1 then
IF rFr1(x) / nFLD(x) < .75 then
TESTITEM(x) = 1
ENDIF
ENDIF

If "its all greek to you" then maybe Donnie or MikeDee can help with the concept in Access queries. I'm not an Access user so forgive me if this looks complicated - its not.

This same formula could be used on any ranked factor to find the top (or bottom) ranked items in terms of percentage. The percentage could be spread wider to 33% if desired. We tend to test our eliminations based on the rankings themselves, such as Fr1 > 7. That works fine on large fields, but you won't eliminate anyone from a field of 7 by using that.

C90 and other speed rating methods are particularly sensitive to the Fr1 %rank when tested. My tests show C90 horses with strong Fr1 win at a far higher % and have much better ROI than those that have weak Fr1 ranks.

[MikeDee]

I figure someone will ask so I'll just post something now to get you started. To query this, open a query that you are using for your ROI and/or testing. Go to a empty column and put your cursor in the field box. In this box type:

Test:rf1/nfld
note; if you are using a brand new query to do this save it first before you put the above statement in. This is so access knows where to get rf1 and nfld from. Otherwise it will ask you
what they are.

Next put your cursor in the criteria row in this column and type >=.75

Next go to the rACL field and put in the number 1

When you run this query you will get all of the runners that are >=.75 of rF1 and have a rACL of 1.

After you run this you will notice that Access gives you a long decimal value for test. you can pretty this up in the design mode by putting your cursor in any of the criteria boxes in the test column and right clicking. This will open a pop up menu. Click on properities and click on format.

You can change this to a fixed number or a percentage. Now go to decimal places and change it to 0 or any number of places you would like to see in the query output.

The vba code approach would require that you put a field in your table called testitem. You could then put a 1 in this field if the test is positive. There are a couple of ways to populate this field. The easiest way is with a update query and you can also do it with a VBA module.
If you are interested in going this way let me know and i'll post instructions.

I have attached and example, this is sorta a test as I don't know if this will come through as an attachment.

Nope attachment didn't work, probably to big

[fred4now]

Great info Ken and Mike. Just to clarify if you type >=.75 this will give you bottom 25% and <=.75 will give you top 75%. Yes?

[Ken Massa]

Thanks for the nuts and bolts Mike!

Fred4now:
I see how that could look confusing. > .75 etc

The rFr1 is a ranking from 1-9. The field size ranges typically from 5-12

Suppose a field size of 10.
The formula rFR1 / nFLD would equate to the following for each ranking as follows

rFr1
1= .10
2= .20
3= .30
7= .70
8= .80
9= .90
10 (defaults to 9) = .90

By using the query rFr1 / nFLd >= .75 would uncover those ranked 8 and 9 (10) in this group of horses.

The contrary rFr1 / nFLD < .75 would find those ranked 1-7 as their total is less than .75

[MikeDee]

Did a little test on Kens' fr1 elimination method. I set the fr1 to <=.75 as suggested. (Did it right this time - thanks F4N). looked at the follwing tracks for the last 7 days. and found that rVel2 had a positive ROI.

this is for rVEL =2 rFR1 <=.75, all Dirt or Wet, all races:
Trk nWin nBet Win% aveMut amtBet amtWon P/L ROI
AQU 3 28 11% $4 $56 $12 ($44) -79%
BEU 7 34 21% $9 $68 $61 ($7) -10%
CD 7 41 17% $15 $82 $106 $24 29%
CRC 9 42 21% $5 $84 $43 ($41) -49%
CT 7 40 17% $12 $80 $82 $2 3%
DED 2 13 15% $53 $26 $106 $80 308%
FE 3 17 18% $10 $34 $29 ($5) -15%
FL 6 32 19% $11 $64 $64 $0 0%
FP 8 26 31% $15 $52 $120 $68 131%
GG 6 33 18% $5 $66 $30 ($36) -55%
HAW 5 28 18% $13 $56 $63 $7 13%
HOL 6 24 25% $8 $48 $46 ($2) -4%
HOO 10 42 24% $8 $84 $80 ($4) -5%
HOU 5 20 25% $8 $40 $39 ($1) -3%
LRL 5 32 16% $14 $64 $69 $5 8%
MNR 7 32 22% $10 $64 $70 $6 9%
PEN 4 27 15% $13 $54 $50 ($4) -7%
PHA 6 40 15% $14 $80 $82 $2 3%
RP 13 30 43% $6 $60 $82 $22 37%
SUF 9 31 29% $8 $62 $68 $6 10%
TDN 4 26 15% $13 $52 $52 $0 0%
TUP 9 38 24% $7 $76 $66 ($10) -13%
WO 8 39 21% $8 $78 $62 ($16) -21%
149 715 21% $1,430 $1,482 $52 4%


Since FTS'ers have a big impact on any pace related ranking I removed any race with 1 or more FTS and re-ran

Trk nWin nBet Win% aveMut amtBet amtWon P/L ROI
AQU 3 22 14% $4 $44 $12 ($32) -73%
BEU 7 30 23% $9 $60 $61 $1 2%
CD 7 25 28% $15 $50 $106 $56 112%
CRC 8 29 28% $5 $58 $39 ($19) -33%
CT 7 35 20% $12 $70 $82 $12 17%
DED 1 12 8% $2 $24 $2 ($22) -92%
FE 3 16 19% $10 $32 $29 ($3) -9%
FL 6 31 19% $11 $62 $64 $2 3%
FP 7 21 33% $15 $42 $103 $61 145%
GG 5 25 20% $5 $50 $27 ($23) -46%
HAW 5 20 25% $13 $40 $63 $23 57%
HOL 5 14 36% $7 $28 $37 $9 32%
HOO 8 34 24% $8 $68 $65 ($3) -4%
HOU 5 17 29% $8 $34 $39 $5 15%
LRL 4 26 15% $16 $52 $63 $11 21%
MNR 7 28 25% $10 $56 $70 $14 25%
PEN 4 23 17% $13 $46 $50 $4 9%
PHA 5 38 13% $14 $76 $72 ($4) -5%
RP 9 18 50% $7 $36 $62 $26 72%
SUF 8 27 30% $8 $54 $62 $8 15%
TDN 4 22 18% $13 $44 $52 $8 18%
TUP 9 32 28% $7 $64 $66 $2 3%
WO 7 31 23% $8 $62 $58 ($4) -6%
134 576 23% $1,152 $1,284 $132 11%

[Donnie]

Mike,
Would be interested to see the same numbers only split by route (mile and longer) and sprints (less than mile). And Ernie wants to see them split by male/female.
Just interested to see where it lands?
Thanks! (if you've got the time that is!)

[MikeDee]

The way I update my db I only have all the tracks for the past week on Mondays. Here are the tracks I am following where I have the data:

sprint
Trk nWin nBet Win% aMut aBet aWon aP/L ROI
AQU 2 18 11% $5 $36 $9 ($27) -75%
CD 5 28 18% $19 $56 $95 $39 70%
CRC 4 28 14% $5 $56 $18 ($38) -68%
DED 2 11 18% $53 $22 $106 $84 382%
FE 1 8 13% $8 $16 $8 ($8) -50%
HOL 6 21 29% $8 $42 $46 $4 10%
HOO 5 28 18% $9 $56 $43 ($13) -23%
HOU 2 10 20% $7 $20 $13 ($7) -35%
MNR 2 20 10% $13 $40 $25 ($15) -38%
SUF 6 21 29% $9 $42 $52 $10 24%
WO 4 21 19% $9 $42 $35 ($7) -17%
39 214 18% $428 $450 $22 5%
route
Trk nWin nBet Win% aMut aBet aWon aP/L ROI
AQU 1 10 10% $3 $20 $3 ($17) -85%
CD 2 13 15% $6 $26 $11 ($15) -58%
CRC 5 14 36% $5 $28 $25 ($3) -11%
DED 0 2 0% $0 $4 $0 ($4) -100%
FE 2 9 22% $11 $18 $21 $3 17%
HOL 0 3 0% $0 $6 $0 ($6) -100%
HOO 5 14 36% $7 $28 $37 $9 32%
HOU 3 10 30% $9 $20 $26 $6 30%
MNR 5 12 42% $9 $24 $45 $21 88%
SUF 3 10 30% $5 $20 $16 ($4) -20%
WO 4 18 22% $7 $36 $26 ($10) -28%
30 115 26% $230 $210 ($20) -9%

male
Trk nWin nBet Win% aMut aBet aWon aP/L ROI
AQU 1 19 5% $3 $38 $3 ($35) -92%
CD 4 24 17% $7 $48 $28 ($20) -42%
CRC 6 22 27% $3 $44 $18 ($26) -59%
DED 2 8 25% $53 $16 $106 $90 563%
FE 1 9 11% $7 $18 $7 ($11) -61%
HOL 5 13 38% $8 $26 $40 $14 54%
HOO 5 20 25% $8 $40 $40 $0 0%
HOU 4 8 50% $8 $16 $33 $17 106%
MNR 6 20 30% $9 $40 $55 $15 38%
SUF 6 19 32% $9 $38 $53 $15 39%
WO 5 24 21% $8 $48 $40 ($8) -17%
45 186 24% $372 $423 $51 14%

female
Trk nWin nBet Win% aMut aBet aWon aP/L ROI
AQU 2 9 22% $5 $18 $9 ($9) -50%
CD 3 17 18% $26 $34 $78 $44 129%
CRC 3 20 15% $8 $40 $25 ($15) -38%
DED 0 5 0% $0 $10 $0 ($10) -100%
FE 2 8 25% $11 $16 $22 $6 38%
HOL 1 11 9% $6 $22 $6 ($16) -73%
HOO 5 22 23% $8 $44 $40 ($4) -9%
HOU 1 12 8% $6 $24 $6 ($18) -75%
MNR 1 12 8% $15 $24 $15 ($9) -38%
SUF 3 12 25% $5 $24 $15 ($9) -38%
WO 3 15 20% $7 $30 $22 ($8) -27%
24 143 17% $286 $238 ($48) -17%

[Frank N]

Originally posted by fred4now
Great info Ken and Mike. Just to clarify if you type >=.75 this will give you bottom 25% and <=.75 will give you top 75%. Yes?

Fred: No. Using >=0.75 and <=0.75, will include the horse exactly ranked at 0.75 (rank 3 in a 4 horse field, rank 6 in an eight horse field, and rank 9 in a 12 horse field) on both queries.

Because you're using a percentage that uses a ranking, in which the lower numbers represent better ranks, you want to do it this way:

rF1/nFLD:

Top 75%: <=0.75
Bottom 25%: >0.75

For example, in an eight horse field, the first would return ranks 1 through 6 (since 6/8 is exactly 0.75) and the second would return ranks 7 and 8.

Of note are two special instances. Consider this example:

Top 50%: <=0.50
Bottom 50%: >0.50

This will return half of the field in races with an even number of horses. However, it will return one less horse in fields with an odd number of horses. For example, in a field of 9, this will return the top 4 horses, since 5/9 is 0.56 (rounded).


The other instance is making either percentage too small. For example:

Top 10%: <=0.10
Bottom 10%: >0.10

Note, that this will not return any horse for the top category in fields smaller in size than 10. For example, in a 7 horse field, the top ranked horse would be 1/7, or 0.143 (rounded), which is not smaller than the 0.10 criteria. 0.21 is probably the smallest usable criteria to still return at least one horse, since it will do so in fields as small as 5 horses. Note that using exactly 0.20 would not return a bottom rated horse for the smaller category in a field of 5.