Home » Racing & Sport » Making Sense: Ability Rating (AR) Adjustment

Making Sense: Ability Rating (AR) Adjustment

Good housekeeping and getting in line with UK ratings

Ability Ratings (AR) in our publications have been universally adjusted this week, to allow comparisons between crops over an extended period of years.

The adjustment also provides an opportunity to bring Ability Ratings (AR) in line with English rating of Raceform/Racing Post (RPR).

So now an English horse rated (for example) RPR 99 will be more or less the same as one of our horses rated AR 99.

Estimate

Ability Ratings (AR) show how good a horse is, or isn’t.

The ratings follow a normal distribution, graphically represented in a bell curve for a population of horses of the same sex. AR shows by how much a horse’s AR differs from that of the average horse of the same sex in that population.

AR is expressed in a number representing weight (pounds in our case, as well as in UK).

The daily calculations of AR are based on results of races, in terms of weight carried in the race and beaten lengths at the finish.

In addition, there is a weight-for-age (WFA) component, to compensate for immaturity of younger horses.

WFA is expressed in a number representing a weight allowance (pounds), depending on the age of a horse, the distance of a race, and the time of year.

WFA is not a real number, but an approximation of how the average horse matures over time.

This is shown in the WFA table. Different racing bodies use variations of this table.

WFA has an influence on the overall level of ratings in a population.

The most widely used WFA table (internationally) causes an automatic rise in the overall level of ratings over time.

We experimented with WFA during the Nineties over an extended period of time.

By adjusting the numbers, we found it was possible to slow the increase and even reverse it.

Our final (adjusted) WFA table, for our own use only, was close to our objective of keeping the ratings at the same level from year to year.

It had an effect of reducing AR by about half a pound per year, which on predicting outcome of races had little effect in the short term.

However, half a pound per year translates into 5lb in a decade, and 10lb in two decades, from the Nineties to today.

This explains why the overall level of our yearly ratings shows a marked drop compared to the Nineties.

Because little of the influence of the drop was felt in the short term, we did not make corrections over the 20 year period of time – but there is reason to want to do so now.

The adjustment is important in the calculation of our Power Of Ratings, the stallion statistic which rates sires using their offspring’s AR, going back in time.

Power Of Ratings can be found on the main page of our website, as well as on individual stallion pages on the website, and in the yearly Sires publication.

Power Of Ratings

The universal adjustment gives an opportunity, too, to align our AR with international ratings, notably the RPR of Racing Post/Raceform in UK (the level of which runs slightly below Timeform’s TFR).

To that end we calculated the overall level of the RPR (by sex) and use that as a base to adjust our AR to the same level, crop by crop.

Harry's Charm

Each crop, between 1996 to 2015, consists of some 1400 rated horses of each sex to be adjusted individually.

This is done by making the median lifetime-best AR of all horses with a rating (ie AR > 0) the same from year to year.

A fun side benefit turned out to be the ability to put a number on what it takes to earn black type.

For winners of black type races in the period between 1996 and now we calculated mean and median AR for males and females – somewhat surprisingly mean and medium turned out to be exactly the same.

Dancer's Daughter

For males the average AR to win Gr1 is 108, Gr2 104, Gr3 100 and Listed 97.

For females the average AR is 102 (Gr1), 99 (Gr2), 96 (Gr3), 92 (L).

The difference in AR between the sexes of about 5lbs is the same as for all AR ratings in the total population. It seems a 5 pound sex allowance makes sense!

Have Your Say

Comments Policy
The Sporting Post encourages allcomers to feel free to have their say in the spirit of enlightening the topic, the participants and the originator of the thread. However, if it is deemed to be either offensive, insulting, personal, false or possibly unsubstantiated, the Sporting Post shall, on it's own assessment, alter or remove comments.

7 comments on “Making Sense: Ability Rating (AR) Adjustment

  1. Jay August says:

    Interesting! Rebalancing (Correlating) based on the median rating over successive years and decades has always been my method for comparing one years ratings to another, whether I have used merit-type ratings or other methods (speed/class).

    The part which is never easily calculated is determining what potential improvement or decline there has been in the population from say the 1970’s to the 2010s. One can speculate on that quite a lot. I have attempted only two methods; adjustment by foal crop size and median times from decade to decade, but it remains a somewhat grey area.

    I see that you now have an average median rating of around 72, which I think is up from 66. Is that correct? How exactly did you correlate to the RPR ratings; by median or other method?

    Also what percentage or number of the foal population do you rank yearly? Is it the 1400 figure you mention? I notice further that your adjustment has shifted the right tail to 129-130. Which horses have achieved that rating? And lastly what percentage of horses occupy one standard deviation from your new mean?

    1. karel says:

      Jay, answers to a few questions.
      We rank the total population on career-best ratings which are >0, split by sex (!). These number around 1400 each per crop, so a little under 3000 in total per crop.
      The median AR is 76 (male) and 71 (female), the mean 73 (male) and 68 (female), both with standard deviations of 19.
      Of the males, 74% falls within 1 SDev (94% 2 Sdev, 99% 3 SDev), as do the females: 74% (95%, 99%).

      Median AR was correlated with median RPR (Raceform/Racing Post in UK) from the most recent years, at 76 for males. I have a database of Raceform results from 1998 to date, and went through the lot, but only took into account performances recorded in UK & Ireland. Rather strikingly, the difference between male and female was markedly higher with RPR (and also BHA) than with us (AR) – perhaps there is less integration of male/female in UK, making their difference less reliable? Also, unlike with us, male performers outnumber female ones, in yearly crops racing in UK/IRE of around 4000. Anyway, a few fixed pounds extra sex-difference do not materially influence the method from crop to crop.

      There are few horses in the right tail, predictably the highest rated are sprinters.
      129 is for Captain Of All; 127 Talktothestars; 125 Golden Loom, What A Winter; 124 Harry’s Charm.
      Of these Golden Loom has the most repeat 125’s – some of the others had one-off performances at the level.
      The best milers come in at 123 – Jet Master, Variety Club, Futura.
      As an aside, I note Variety Club has a career-high RPR 124.
      Most interesting for me is that the highest rated half-dozen from each crop have very similar ratings. Being Champion remains relative!

  2. Roderick Mattheyse says:

    I wonder if this will render my easy quick analysis of MR [official weight ] vs AR useless now

    1. karel says:

      Shouldn’t affect the comparison, other than that AR now is on a higher level.
      It may take a little time to get accustomed to that, but in a month or so you’ll be familiar again.
      Relatively nothing has changed.

  3. Jay August says:

    Karel, thanks for those replies. A foal crop of roughly 3000 makes sense. I had read that previously as 1400 but that is for one sex only.

    Two observations;

    1) I’ve noticed that SA horse populations produce approximately 72% clustered in the first SD which is somewhat off the more normal 66-68%. I’ve checked the BHA, Irish and NZ databases (because they are freely available online and can be checked by anyone) and they show a more normal ~67%. I notice that your ARs, the NHA MRs and mine all show the 70+ number. Even Dubai with its much smaller (and better class) population has 66% within SD1 . Any thoughts on why we seemingly have more average horses in SA? When I first looked into this I thought that the flaws (restrictions on efficient handicapping) in the MR system were influencing this result but as both your and my database show the same trend there must be some dynamic at play.

    2) I have noticed over the years, starting way back in the late 80’s, that your best ratings are for sprinters, which is not what one finds in Europe. Any idea why your numbers always seem to favour sprinters so much? Even though I use a much more lenient pounds per lengths than others at the sprint distances, my top ratings fall more in to line with the European trend where middle distance performers rate highest. Note that I do no stats on breeding and generally stay well clear of that subject as I have little interest in it. Perhaps we like Australia have a breeding population skewed towards speed?

    1. karel says:

      Jay
      1. One reason I can think of that I use a bottom cut-off (AR >0) – that could exclude a bunch of raced (bad) horses, which would have an effect on the SD %
      Another reason could be that the normal distribution is skewed (mean and median are different, even in this large population) – maybe it’s a lognormal instead of normal?
      2. It could be the length/distance translation, in combination with the fact that sprint are usually true run races. With us a very large proportion of races isn’t true run, which affects length/distance translation and weight-for-age allowance (like when races are decided in a sprint home). I might be clutching at straws here.

      There’s another thing, which might have an influence. We rate very conservatively, especially when races aren’t true run.
      The reason is that to err on caution has little effect on future ratings.
      Going too high when it isn’t deserved can have a serious long-term effect and is much more difficult to eradicate.

  4. Jay August says:

    Karel,

    1) I’ve compared all data with ratings of R>0, so that is a constant. The mean and the median in all cases is pretty similar so there is not much to be implied from that.There are however differences in skewness and kurtosis (flatness) of the various databases. Our MRs are slightly left-skewed while the other countries have slight right skewness. Our MRs have a high positive kurtosis while the UK, Ireland and Dubai are slightly negative in that regard.

    2) The pace of races is a possibility although it would be very difficult to tease that out easily. I’m not sure how reliable the current sectional data is in SA but I’ll look at it in a years time when there is a larger amount of data available. I’ll probably overlay it onto my ratings database, add a pace variable (shape of race rather than time par) and see what what useful trends it yields.

    I rate very conservatively as well although I am not so sure that is always the case with the MR ratings.

    There is much going on lately in the realm of data analysis in horse racing. We have not even touched on stride analysis which is now starting to become an avenue of assessment in the UK.

    Given the normal run of technology the next decade should produce a myriad of new tools to analyse horses and races. Hopefully racing remains viable as a sport over that time!

Leave a Comment

‹ Previous

SA Champions Season – The Countdown

Next ›

Back To Kenilworth Today

Popular Posts