World Throughbred Rankings

C

chroniclandlord

Senior Jockey
Joined
Oct 4, 2007
Messages
5,779
Location
Co. Armagh
Frankel rated 140

Following a historic recalibration,

Dancing Brave dropped to 138 from 141

Shergar dropped to 136

Alleged dropped to 134
 
Last edited:
The graphs at the end of the paper suggest, even after the ratings adjustments, that mean performance levels in the last 25 years are drifting ownwards, so there's no support for those who assume modern training and breeding techniques inevitably mean standards are rising.
 
Grey said:
The graphs at the end of the paper suggest, even after the ratings adjustments, that mean performance levels in the last 25 years are drifting ownwards, so there's no support for those who assume modern training and breeding techniques inevitably mean standards are rising.
Click to expand...

You wouldn't expect to see evidence of rising standards: all of these ratings systems assume that the entire population of rated horses will be distributed around a similar average rating every year, which hides any such overall improvement.
 
John Oxx, having trained Sinndar, Alamshar, Azamour and Sea The Stars would be probably the best judge of the amendments to ratings especially relating to the first three named.
Hopefully his opinion will be sought.
I agree with landlord as far as I can understand it bar the upward adjustment in 2004.
 
Gareth Flynn said:
You wouldn't expect to see evidence of rising standards: all of these ratings systems assume that the entire population of rated horses will be distributed around a similar average rating every year, which hides any such overall improvement.
Click to expand...

Fair enough, Gareth, but in that case wouldn't you expect to see horizontal lines rather downward-sloping ones?
 
No, because they're not sampling the entire population of rated horses each year, just the top ~1%, so you'll see fluctuations.

That's the flaw in the methodology used for the recalibration. Obviously they're restricted by the records they have available for them, but what they really needed to do was estimate the distribution of ratings for the entire population of rated horses (not just those rated 120+) for each year and then ensure that each year was pinned to the same average. New ratings for the outliers at the top of the tree would then follow naturally.
 
I don't really agree, Gareth.

You could have the same average, but a larger standard deviation of ratings of top horses.

I have always thought that impressive wins in Group 1 company are given more higher ratings now than they would have been before.
 
Gareth Flynn said:
No, because they're not sampling the entire population of rated horses each year, just the top ~1%, so you'll see fluctuations.

That's the flaw in the methodology used for the recalibration. Obviously they're restricted by the records they have available for them, but what they really needed to do was estimate the distribution of ratings for the entire population of rated horses (not just those rated 120+) for each year and then ensure that each year was pinned to the same average. New ratings for the outliers at the top of the tree would then follow naturally.
Click to expand...

Does every year need to be the same average?
 
In 1977 145 horses were awarded a rating of 120+, by 2004 this figure was just 19.

Can't help but find quotes from Swinburn on Shergar and Harwood about Dancing Brave frustrating when complaining of their new ratings when it's crystal clear that neither understands the subject matter.
 
Bar the Bull said:
I don't really agree, Gareth.

You could have the same average, but a larger standard deviation of ratings of top horses.
Click to expand...

If its assumed that the ratings are distributed normally, then we know that the 68 / 95 / 99.7 rule applies; i.e. that 99.7% of all ratings will be contained within 3 standard deviations of the average.

Take the current full set of official ratings for Flat Turf horses on the BHA website. There's 9252 ratings. The average is 66.6 with an SD of 17.9. So 99.7% of ratings should be contained within the range 66.6 - 3*17.9 and 66.6 + 3*17.9 which is between 13.0 and 120.2

As it happens, 9252 of the ratings are found in that range, which is 99.8% of the population.

But look at the range again - it only goes up to 120. After that we're basically talking about outliers, right? But these 120+ horses are precisely those which the ICs/WTRRs rate. Expecting the outliers to be neatly distributed doesn't make sense - that's why they're outliers!

But if you know how much they differ from the average, then you can normalise each year consistently. You're basically just moving the same normal curve sideways.
 
SlimChance said:
Does every year need to be the same average?
Click to expand...

In order to keep the ratings comparable year-on-year, yes. But you do have to take into account changes in the sport which might change the average - e.g. the introduction of more low-grade racing which introduces many more moderate horses into the rated population, thus dragging the average down.
 
Gareth Flynn said:
In order to keep the ratings comparable year-on-year, yes. But you do have to take into account changes in the sport which might change the average - e.g. the introduction of more low-grade racing which introduces many more moderate horses into the rated population, thus dragging the average down.
Click to expand...

In that case they've probably gone as far as they can?
 
Within the parameters they set themselves - i.e. purely looking at the distribution and averages of ratings in each year of the WTRRs/ICs, yes, I think they went as far as they could.

But they could have gone further. They never even considered actually digging into some of the races; they worked from the principle that the previous ratings were correct, it was just the overall level that was wrong.

But the biggest reason for the differences between the 80s and now are down to them changing the poundage they use. Prior to the early 90s they used a scale similar to Timeform. Now they use a more conservative scale. This doesn't just reduce the ratings - it compresses them. They haven't taken that into account. The ratings are not discrete measurable quantities, they're all inter-related and inter-dependent. If you change the maths of the relationship between them, you can't adjust them uniformly.

Here's an example:

Let's imagine the average rating should be 70. One year in the 80s a horse was rated 140. Now they go back and decide that that year, the level was 7lbs too high compared to what they would be now. So the 140 horse becomes 133, a 130 horse becomes 123 and a 120 horse becomes 113.

But what if the reason the 140 horse was rated that high in the first place because he gave the 130 horse a 10lb beating and the 120 horse a 20lb beating, and under today's poundage those same beatings would only be worth 7lbs and 14lbs? Well then the new ratings should be 133, 126 and 119. The horses have all been reduced but not uniformly - the new poundage has brought them closer together.
 
davidjohnson said:
In 1977 145 horses were awarded a rating of 120+, by 2004 this figure was just 19
Click to expand...

David, could you share Timeform's number of 120+ horses in 1977 and in 2004? (in England, Ireland and France)
I haven't got my "Racehorses of ..." collection with me right now, but I seem to remember Timeform's ratings were also "inflated" in early years of its publications, weren't they?
 
Haven't got the figs to hand and am out of the office until next week. Remind me, if someone hasn't come forward with them.
 
Gareth Flynn said:
If its assumed that the ratings are distributed normally, then we know that the 68 / 95 / 99.7 rule applies; i.e. that 99.7% of all ratings will be contained within 3 standard deviations of the average.

Take the current full set of official ratings for Flat Turf horses on the BHA website. There's 9252 ratings. The average is 66.6 with an SD of 17.9. So 99.7% of ratings should be contained within the range 66.6 - 3*17.9 and 66.6 + 3*17.9 which is between 13.0 and 120.2

As it happens, 9252 of the ratings are found in that range, which is 99.8% of the population.

But look at the range again - it only goes up to 120. After that we're basically talking about outliers, right? But these 120+ horses are precisely those which the ICs/WTRRs rate. Expecting the outliers to be neatly distributed doesn't make sense - that's why they're outliers!

But if you know how much they differ from the average, then you can normalise each year consistently. You're basically just moving the same normal curve sideways.
Click to expand...

I get that.

But here is the thing, I am not expecting outliers to be neatly distributed. Of course they are going to differ. However, I suspect that the standard deviation of the top-end horses (these outliers) has increased over the past 30 years.

Anecdotally, I feel far more "rub your eyes" performances are taken at face value by handicappers than was the case over a decade ago.
 
LBM said:
David, could you share Timeform's number of 120+ horses in 1977 and in 2004? (in England, Ireland and France)
I haven't got my "Racehorses of ..." collection with me right now, but I seem to remember Timeform's ratings were also "inflated" in early years of its publications, weren't they?
Click to expand...

List of ratings by order doesn't appear in the 77 issue like it does now with the Top 100. I'm not going through each horse individually to count them I'm afraid.
 
Back
Top