Gold Updates: Cosmetics and PRB

As well as providing bundles of top class thought-provoking editorial during this interminable lockdown, we've also been beavering away on generating some new bells and whistles on our racecards. Actually, we've been mostly cosmetically enhancing our existing features. Let's start with those...

Blue is the new grey

First up, you'll see a lot more blue about the place and a lot less grey.

The card tab now looks like this:

 

Full Form, with its collapsible blocks, now looks like this:

In the above example, for a geegeez.co.uk syndicate horse, I've collapsed the Race Form and Race Entries blocks.

 

Perhaps the biggest change is to Instant Expert where we've inverted the colour blocks. So, where previously the outlines and numbers were in the colour (green, amber, red), now the block is that colour with the number font in white. It looks like this:

Your first 30 days for just £1

 

 

Similar cosmetic amendments have been made to the result, pace, odds and draw tabs, which leads me nicely on to...

 

New Draw Metric

We've introduced a new metric on the Draw Analyser and in the draw tab, called Percentage of Rivals Beaten, or PRB. I've explained more about it in this post, which I very much recommend you read if you haven't already.

The value of PRB over, say, win or place percent is that every runner in every race receives a performance value, with only the last placed horse getting 0. So, for example, in a six horse race, there would be a winner, one additional placed horse (as well as the winner), and four unplaced horses.

In the win percentages, that race would produce a breakdown of 100/0/0/0/0/0 (100% win for the winner, 0% win for the rest of the field).

Place percentages would have 100/100/0/0/0/0 (two placed horses, four unplaced '0' horses).

But the third horse has performed better than the fourth, fifth and sixth horses; and the winner has performed better than all of its rivals. PRB aims to more accurately place a value against finishing position. So the percentage of rivals the winner beats will always be 100%, and the PRB of the last placed horse will always be 0%, but in between there will be a sliding scale. In this six-horse race example, the second horse has beaten 80% of its rivals (four out of five rivals), and the fourth placed horse has beaten two home, which is 40% of rivals.

In a fair draw each stall, or group of stalls, would see a PRB score of 50%, or 0.5. And many stalls are within one or two percentage points of that. If a draw location has a PRB of 55%+ (0.55+) it is probably favoured; the converse is also true: if a stall has a PRB of 45% or less it may be somewhat unfavoured. Here's how it looks on the draw tab:

The table columns to the right hand side list PRB and PRB2. In this case we can see that high is favoured to a small degree and low commensurately unfavoured.

PRB2 is simply the PRB score multiplied by itself. What this does is accentuate the percentages: in practical terms it rewards those finishing closer to the winner than those finishing further down the field, recognising that horses may not be ridden out for the best possible placing if that placing is going to be eighth of 20, whereas they virtually always will if that placing is third of 20. There is more on how that works in the horse racing metrics post.

When looking at individual draws, I've introduced a metric called PRB3. Similar to IV3, it takes a rolling three-stall average PRB of the stall in question and its immediate neighbours. So, for example, the PRB3 of stall six would be the average PRB of stalls five, six and seven. It is, in exactly the same way as IV3, a means of smoothing the curve and making sense of draw data distribution. Here it is in action:

 

PRB has lots of potential applications in horseracing datasets, and we've started our adoption in the draw space. It will be especially useful when, as in the examples above, there is not a lot to go on in terms of runs, wins and places. There is still not a great deal in the PRB dataset but, by scoring every horse in each race in the sample, there is more data depth in which to fish.

That's all for this update. Very soon we'll be able to get stuck back into one of our favourite pastimes: messing around with racing data! And Geegeez Gold will have it well covered.

Matt

2 replies
  1. shorts65
    shorts65 says:

    Matt

    I know over on Proform they sometimes use PRB squared – any thoughts on the usefulness of this as a metric?

    Reply
    • Matt Bisogno
      Matt Bisogno says:

      Hi John,

      The main advantage, as outlined in this post – https://www.geegeez.co.uk/horse-racing-metrics-a-e-iv-prb/ – is that it weights the ‘reward’ for horses finishing nearer to first.

      For example the PRB vs PRB^2 (another way of writing PRB squared) figures for a six runner race would be:

      1st – 1.00 1.00
      2nd – 0.80 0.64
      3rd – 0.60 0.36
      4th – 0.40 0.16
      5th – 0.20 0.04
      6th – 0.00 0.00

      As you can see, PRB is a linear (i.e. even) distribution, whereas PRB^2 is exponential.

      In layman’s terms, I think PRB is fine on bigger datasets, where things tend to naturally balance out; but when there are less races in the sample, the additional ‘penalty’ PRB^2 bestows on those finishing further away from first helps to distinguish.

      So I’d say it is useful when there are less data to review. A good PRB^2 might be anything above 0.3 and a poor one anything below 0.2.

      Hope that helps,
      Matt

      Reply

Leave a Reply

Want to join the discussion?
Feel free to contribute!

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.