Posts

New Metric on Geegeez Gold: PRB

We're constantly striving to improve Geegeez Gold, our flagship racecards and form tool service. After a few quieter months - lots going on in the background - we're about to inject a new metric into Gold.

We've deliberately kept it away from the more commonly used numbers, simply because if you don't want to engage with this, we don't want it in your way. At the same time, I very much believe you should take heed of the new number and that's why I've put this post together.

So, what is the new number? Well, it's not exactly brand new as we already display Percentage of Rivals Beaten (PRB) within our draw content. But we're now extending it calculation and display to trainer, jockey and sire data. Here is some more information on Percentage of Rivals Beaten...

What is PRB?

Percentage of Rivals Beaten (PRB) is a calculation based on a horse's finishing position in relation to field size. It makes key distinctions between a horse finishing, say, third in a five-horse race (PRB 50%, two rivals beaten, beaten by two rivals) and finishing third in an eleven-horse race (PRB 80%, eight rivals beaten, beaten by two rivals).

For a collection of results - for example, a trainer's record over the last year - we take an average of all the individual PRB scores.

On geegeez.co.uk, we express PRB as a number between 0 and 1. So, in the examples above, 50% is 0.5 and 80% is 0.8.

What is convenient about PRB is that a par score is always 50% of rivals beaten, or 0.5. This means that a trainer with a one-year PRB of 0.55, 55% of rivals beaten, is doing very well; conversely, a trainer with 0.45 as his PRB is under-performing in finishing position terms.

It is always important to remember that finishing position is not the only number in town and, as with all numbers, it should be used sensibly and in concert with other metrics.

Why is PRB useful?

Your first 30 days for just £1

PRB is useful because it helps to make small datasets bigger. In racing we are almost always hamstrung by small datasets, relative to what general statistics would consider so at any rate. And when we then try to discern knowledge from the data by looking only at wins we ignore seven-eighths of the information we have (assuming an average field size of eight, one winner, seven losers).

If we had 1,000,000 wins to consider, that wouldn't be much of an issue. But we don't. We have much smaller groups of wins and runs with which to work.

Historically I've used place percentages to enlarge the positive to negative comparison: using our eight-runner average, we now have three 'wins' (placed horses) for five losses (unplaced horses). That's much better but still lacking in nuance.

PRB awards 'score' to every runner except tail end Charlie in every race (ignoring non-completions which are dealt with separately - an explanation of how we've accounted for them will appear in the user guide as it will add little value here). This has some challenges of its own; for instance, a horse that went hard from the front and is still battling for third place will be ridden right to the line, whereas that same horse may be eased off if/when four others have already passed it: it has given its running already and there is little be gained from finishing fifth or ninth.

Such issues are accommodated up to a point by squaring the PRB figure, and you can see how that manages the curve in the post linked to at the bottom of this one if you're that way inclined.

The crux is this: PRB is useful because it helps us understand the totality of performance of a dataset rather than just a fraction (win or place, for instance).

How should I use PRB?

PRB has utility in isolation because every score can be compared to 0.5 to understand whether the thing being measured - trainer, jockey or sire performance in our case - is better or worse than what might be expected.

But, of course, we should expect that, for example, Paul Nicholls will have a far higher one-year National Hunt handicap PRB figure than Jimmy Moffatt. He does, 0.62 vs 0.5 at time of writing. But knowing that is unlikely to add to our bottom line; at least not in or of itself.

As it happens, both have been profitable to follow blindly in handicaps in the past twelve months: Nicholls has an A/E of 1.05 (and an SP win profit of +16.70) while Moffatt has 1.26 / +21.50.

If anything, Nicholls' figures are more impressive, for all that Moffatt's may be more sustainable.

What PRB tells us is the amount of merit in unplaced runs. It should be used to support understanding of an entity, rather than as an end in itself. And it is especially helpful in rendering the inference of small samples sizes slightly less of an act of folly.

Where does PRB live?

Regular Gold users will know that PRB - and its close relatives, PRB^2 and PRB3 - have been happily adding value to our draw content for some time.

And now (next week), PRB appears within trainer, jockey and sire data on the racecards and in reports. It is on the far right, out of trouble for those not (yet) interested in its utility.

On reports, it can be found in the same rightmost column location:

Use it or don't use it, but I'd suggest you make yourself aware, as a minimum, of what Percentage of Rivals Beaten is; and when it might pay to keep it in mind.

You can read more about all of our key metrics - A/E, IV and PRB - in this post.

Matt

p.s. more new features coming soon!