Derby/Preakness BSF & PF's

[somerfrost]

Here are a couple posts by Dr Roman defending his PF's for Derby and Preakness as opposed to BSF's...he maintains the BSF are too high, as I posted earlier, he gave Curlin and SS a PF of -52 for the Preakness which translates into a BSF of 103. Interesting data:

"My guess is that Beyer used a projection to arrive at the Derby BSF and that the projection was based on Street Sense's BC Juvenile since the surface and trip were similar. What he didn't take into account, and never does, is the relationship between speed figures at grossly different distances which may have nothing to do with each other since they certainly depend on a horse's distance affinity profile. You can see the effect of distance on speed figures at my web site where I have posted the PF vs distance profiles of individual graded stakes winners. Those interested can link to the data from the Main Menu or directly at
PF vs Distance Profiles of GSWs


Virtually every regression analysis I have done over the years comparing BSFs with PFs affords an R-squared value exceeding 0.8, and these studies have been continually updated since 2000. Although individual comparisons may vary widely, the overall correlation is outstanding. Therefore we can have reasonable confidence that converting a PF to a BSF or vice versa is acceptable for analysis purposes. As an example, if you average the PFs and BSFs for all performances by all starters in the 2001 Derby up to and including the Derby itself you find average PF -27.7 and average BSF 94.3. From the comparison chart at my web site you will see that PF -28 correlates exactly with BSF 94. You can repeat this with any comparative data sets and you'll get a similar result. As I noted, the 2007 Derby prep races also were very close in terms of PFs and BSFs.

I don't think punishment or benefit of individual Derby/Preakness winners for performances in the prep races they weren't involved in is an issue either because the point of the study was to show the probability of an entire crop at a given pre-Derby performance level producing a certain higher level of performance in the classics. The data clearly show that the probability of this crop producing better than average classic performers is virtually nil. According to Beyer this crop has achieved something possibly never before seen in the history of racing.

Also, the probability of four horses making profound forward moves in the transition from prep distances to classic distances is equally improbable. The general scenario is that one or two at most may step up while the others generally regress. In many years none of the competitors reach classic standards, not even the winners.

Concerning improvement by lengths rather than %, the figures are (based on the correlation of this year's Derby/Preakness BSFs with PFs):

2007, 7.7
2006, 3.5
2005, 1.6
2004, 4.0
2003, 2.1
2002, 2.7
2001, 2.9
2000, 0.5
1999, -2.8

If you use the actual PFs for 2007 the improvement is 2.3 lengths.

Here's is an expanded study that has a bearing on the Derby/Preakness BSFs.

I've looked at the BSFs for every current graded TC prep race at a mile or more on dirt going back to 1992 except for the Sham Stakes for which I don't have the data available. That comes to 16 races over a period of 16 years. Then I took the average annual BSF of those 16 preps and compared them to the average of the Derby and Preakness BSFs for those years. From this I calculated the annual BSF difference between the classics and the preps and converted the result to a difference in equivalent lengths at a mile and a quarter using Beyer's own beaten lengths adjustments.

Between 1992 and 2006 the average improvement in lengths between the preps and the two classics was +6.17 lengths with a standard deviation of 1.91. The difference between the preps and the classics in 2007 is 10.44 lengths. The probability of such a difference occurring was calculated using the online Cumulative Normal Distribution Calculator found at stattrek.com/Tables/normal.aspx and the result was 1.3 out of 100 or once every 77 years.

If you limit the study by going back only to 1999 you get essentially the same result I got using PFs. In this case the BSF average was +6.50 lengths with a standard deviation of 1.35. This probability is 2 out of 1000 or once every 500 years.

For arguments sake we can consider a result between the two. IOW, the probability that a crop of 3yos with a prep race BSF average of 95.6 producing Derby/Preakness winners with an average BSF of 110.5 is between 1 in 77 and 1 in 500. "

[Dunbar]

Originally Posted by somerfrost

Here are a couple posts by Dr Roman defending his PF's for Derby and Preakness as opposed to BSF's...he maintains the BSF are too high, as I posted earlier, he gave Curlin and SS a PF of -52 for the Preakness which translates into a BSF of 103. Interesting data:

"My guess is that Beyer used a projection to arrive at the Derby BSF and that the projection was based on Street Sense's BC Juvenile since the surface and trip were similar. What he didn't take into account, and never does, is the relationship between speed figures at grossly different distances which may have nothing to do with each other since they certainly depend on a horse's distance affinity profile. You can see the effect of distance on speed figures at my web site where I have posted the PF vs distance profiles of individual graded stakes winners. Those interested can link to the data from the Main Menu or directly at
PF vs Distance Profiles of GSWs


Virtually every regression analysis I have done over the years comparing BSFs with PFs affords an R-squared value exceeding 0.8, and these studies have been continually updated since 2000. Although individual comparisons may vary widely, the overall correlation is outstanding. Therefore we can have reasonable confidence that converting a PF to a BSF or vice versa is acceptable for analysis purposes. As an example, if you average the PFs and BSFs for all performances by all starters in the 2001 Derby up to and including the Derby itself you find average PF -27.7 and average BSF 94.3. From the comparison chart at my web site you will see that PF -28 correlates exactly with BSF 94. You can repeat this with any comparative data sets and you'll get a similar result. As I noted, the 2007 Derby prep races also were very close in terms of PFs and BSFs.

I don't think punishment or benefit of individual Derby/Preakness winners for performances in the prep races they weren't involved in is an issue either because the point of the study was to show the probability of an entire crop at a given pre-Derby performance level producing a certain higher level of performance in the classics. The data clearly show that the probability of this crop producing better than average classic performers is virtually nil. According to Beyer this crop has achieved something possibly never before seen in the history of racing.

Also, the probability of four horses making profound forward moves in the transition from prep distances to classic distances is equally improbable. The general scenario is that one or two at most may step up while the others generally regress. In many years none of the competitors reach classic standards, not even the winners.

Concerning improvement by lengths rather than %, the figures are (based on the correlation of this year's Derby/Preakness BSFs with PFs):

2007, 7.7
2006, 3.5
2005, 1.6
2004, 4.0
2003, 2.1
2002, 2.7
2001, 2.9
2000, 0.5
1999, -2.8

If you use the actual PFs for 2007 the improvement is 2.3 lengths.

Here's is an expanded study that has a bearing on the Derby/Preakness BSFs.

I've looked at the BSFs for every current graded TC prep race at a mile or more on dirt going back to 1992 except for the Sham Stakes for which I don't have the data available. That comes to 16 races over a period of 16 years. Then I took the average annual BSF of those 16 preps and compared them to the average of the Derby and Preakness BSFs for those years. From this I calculated the annual BSF difference between the classics and the preps and converted the result to a difference in equivalent lengths at a mile and a quarter using Beyer's own beaten lengths adjustments.

Between 1992 and 2006 the average improvement in lengths between the preps and the two classics was +6.17 lengths with a standard deviation of 1.91. The difference between the preps and the classics in 2007 is 10.44 lengths. The probability of such a difference occurring was calculated using the online Cumulative Normal Distribution Calculator found at stattrek.com/Tables/normal.aspx and the result was 1.3 out of 100 or once every 77 years.

If you limit the study by going back only to 1999 you get essentially the same result I got using PFs. In this case the BSF average was +6.50 lengths with a standard deviation of 1.35. This probability is 2 out of 1000 or once every 500 years.

For arguments sake we can consider a result between the two. IOW, the probability that a crop of 3yos with a prep race BSF average of 95.6 producing Derby/Preakness winners with an average BSF of 110.5 is between 1 in 77 and 1 in 500. "

(emphasis added)

Why "limit the study by going back only to 1999"? Maybe I'm missing something, but why look at a subset of the 1992-2007 data? Is Roman picking a subset that looks worse than the full data?

I may be missing a valid reason.

--Dunbar

[somerfrost]

Originally Posted by Dunbar

[/b]

(emphasis added)

Why "limit the study by going back only to 1999"? Maybe I'm missing something, but why look at a subset of the 1992-2007 data? Is Roman picking a subset that looks worse than the full data?

I may be missing a valid reason.

--Dunbar

The reason is that Dr Roman didn't start computing PF's until 1999 so analysis prior to that would be impossible.

[The Indomitable DrugS]

The Preakness card was one of the toughest days I've ever encountered to make figures for.

I was consistantly wrong on how fast races should go in timewise--based on how fast the track was playing--throughout the card that day.

I have a sneaking suspicion that the track maintance crew and the weather played a key role.

[somerfrost]

Originally Posted by The Indomitable DrugS

The Preakness card was one of the toughest days I've ever encountered to make figures for.

I was consistantly wrong on how fast races should go in timewise--based on how fast the track was playing--throughout the card that day.

I have a sneaking suspicion that the track maintance crew and the weather played a key role.

Yeah, I believe so...when I first saw Dr Roman's PF (-52) I was shocked at how much he devalued the race, Beyer's BSF seemed much more in line but Roman has statistical data to back up his number...it does seem unlikely that this group could earn a BSF so high, anyway...it really doesn't matter when looking at 3yo races but could be of interest when they start to face older.

[The Indomitable DrugS]

The Beyer figures certainly don't have the track playing the same speed throughout the card either.