Perils of projecting Flat ratings to hurdles

HOW good will a Flat horse prove when sent over hurdles? So many good hurdlers have been recruits from the Flat that a method for converting ratings between the two codes is helpful and here I will demonstrate a simple method.

This is useful to gauge a horse’s potential over hurdles, though considerable variation is found at the individual level. We will discuss the reasons which may be behind this.

While this rule of thumb is easy to work with, it does not work well across the spectrum of Flat talent. It turns out there are diminishing returns at work: the better a Flat horse, the less well it tends to translate established merit between codes. We can make allowance for this by using a statistical model, an easy chart of conversion between official ratings of the two codes.

One obvious talking point is whether the relationship between Flat and hurdle ratings has changed.

The data

The BHA offers a searchable database of official ratings at britishhorseracing.com/regulation/ official-ratings/ratings-database. This was accessed on October 1, and 328 horses found with both a Flat and hurdle rating from the 11,523 horses listed. A summary of the data for the population of dual-purpose horses can be found in Table 1.

The first row of the table shows that BHA hurdle ratings are about 37 points higher than BHA Flat ratings for horses with a rating under both codes.

The maximum BHA Flat rating in the data is 101, belonging to the nowretired Midnights Legacy. The fiveyear-old was a fluent winner over hurdles at Fontwell in September off a mark of just 124 – just 23 points higher than his Flat rating.

It may also be interesting to examine the distribution of ratings under the two codes. A statistic that provides a measure of this is standard deviation, the bottom row of the table. Among these 328 horses, their BHA hurdle ratings are more spread out because the standard deviation is higher. This can be seen graphically via a density plot in Figure 2. Here, the green distribution which represents hurdle ratings has a wider base and a lower peak than the orange one, which represents Flat ratings. The inference is that the challenge of running over hurdles causes more variation in expressed talent.

Embracing uncertainty in making projections

This difference of 37 points between the BHA Flat and BHA hurdle ratings of dual-purpose horses only describes the average difference observed in the data. What would be useful to know is how Flat ratings predict what a horse will do over hurdles.

Flat and hurdle races obviously have commonalities. But they also have major differences that introduce uncertainty in translating ability shown under one code to the other. The data shows that the better a Flat horse, the more likely it is that its talent is Flat-specific. So, when good

Flat horses go over hurdles, we need to be more cautious in making projections than when moderate Flat horses make the same conversion. Among the second group, hurdling is a bit more likely to bring out stamina or jumping skill that had not been highlighted on the level.

A statistician can allow for this effect via a statistical model. To see how, let us plot the ratings for the dual-purpose horses on a scatter graph in Figure 3.

Each point on the graph represents an individual horse. The horizontal x coordinate is its BHA Flat rating and the vertical y coordinate its BHA hurdle rating. The green line is that of ‘best fit’ because it has the property of coming closest to the cloud of points of any it is possible to draw.

Table 4 is an easy-to-use chart to demonstrate the more accurate method of converting ratings from one code to another.

To use Table 4 with a BHA Flat rating, first find the bracket the rating belongs to which is in the third column, then make the conversion described in the fourth column. For example, if a horse has a BHA Flat rating of 68, this is the ‘60s’ bracket, so the third row dictated that the equivalent BHA hurdle rating is 103, which is arrived at by adding 35 to the rating of 68.

If you run your eye down the final column, the significant effect of diminishing returns becomes clearer.

Flat horses rated 100+ just do not translate to the kind of hurdler you would expect, on average, because it becomes rare to find horses with high-class Flat ability who have the requisite qualities of a top hurdler – jumping, stamina, toughness. The data points to better hurdlers coming more from other backgrounds.

Reasons to be cautious over these projections

Whether you use the simple method or the more complicated one, there are two reasons to be cautious when projecting hurdling talent. These also happen to highlight important precepts in the science of statistics.

Firstly, go back to Figure 3. When we drew that line of best fit, it was chosen to minimise the mean vertical distance from all points. This distance is the ‘standard error’ of the model and is a similar concept to the standard deviation of a set of data. In this case, it is relatively large at 13.94, which warns us to expect considerable inaccuracy when we try to project a BHA hurdle rating from Flat form. In fact, one-third of the time our projection will be at least 14lb out!

Secondly, we need always to think of how the data was created because this can introduce selection biases. Here, it has an order in that a horse tends to receive a BHA Flat rating before it qualifies for a BHA hurdle rating, so the set of horses with ratings under both codes does not include Flat horses who were not persisted with long enough to receive a published hurdle rating. Usually, this was because they did not prove to have the aptitude for hurdling. The upshot of this is that making a projection about a horse’s BHA hurdle rating from its Flat rating is likely to prove too optimistic, because this does not factor in the chance that it has little aptitude and is not persisted with.

The signal versus the noise

The statistical truth here is that our model is too simple. There are so many factors which determine how well BHA Flat ratings translate to BHA hurdle ratings that the standard error of the model has too large a magnitude and overwhelms the granularity of any predictions.

If I like a horse in a juvenile hurdle because it was 7lb better than another on the Flat, and you could counter by observing that its rival is known to have schooled well, is a big, strong type only growing into its frame and has a ton of stamina, the confidence in my prediction is severely eroded because we know there is so much that my model does not explain.

Does Flat ability translate less well to hurdles nowadays?

Thirty years ago, smart horses from the Flat seemed to dominate when sent over hurdles. This was typified by the results of the Champion Hurdle at Cheltenham in the 1990s. Kribensis, Royal Gait, Alderbrook, Collier Bay, Make A Stand and, most famously, Istabraq had all made a name for themselves at various levels on the Flat – just as their breeders had principally intended.

As the years have rolled on, however, jumps racing has increasingly sourced its stars from elsewhere. Since 2014, the

Champion Hurdle has fallen to

Jezki, Faugheen, Annie Power,

Buveur D’Air, Espoir D’Allen,

Epatante and Honeysuckle (twice).

All these had a background in points, bumpers or the French non-thoroughbred AQPS domain.

Of course, not all the Champion Hurdle winners of the 1990s were smart Flat horses. Make A Stand ended his Flat career rated just 73 before his transfer to Martin Pipe and his electric jumping saw him rapidly ascend the hurdling ranks. Even the mighty Istabraq was rated just 87 when he departed John Gosden’s yard for Ballydoyle.

While the arena of hurdles racing starts to bring out the best in jumpsbred horses, or those specifically prepared or purpose-bred, Flat horses must adapt their talent to a different challenge and their better traits simply do not play so well, if at all. A good Flat horse who happens to have the right stuff for jumps will always prove a potent force, but the chances are there are fewer of these around and they are committed to jumping in smaller numbers because resale values for racing abroad have swollen.

In conclusion, a simple rule to convert ratings from Flat to hurdles is to add 35. The difference between the scales is 37 but caution is advised because of selection bias in the data. However, there is so much more to the situation than a single-number rule can provide. In the words of Ben Goldacre’s masterpiece on statistics: I Think You’ll Find It’s a Bit More Complicated Than That.