Reader Q&A--Statistical Inquiries on Nicking

re: Reader Q&A--Statistical Inquiries on Nicking

Byron Rogers — Fri, 26 Aug 2011 18:57:20 GMT

Robert,

Unfortunately there are not a lot of options in Australia for stallions with Darshaan/Shirley Heights. Maybe a horse like Spinning World who is by Nureyev out of a Riverman mare might fit? You might want to try running a TrueNicks Broodmare Analysis report with stallions that fit into your price range and see what that returns for you.

Best of luck,

Byron Rogers

re: Reader Q&A--Statistical Inquiries on Nicking

Robert Cooke — Fri, 26 Aug 2011 10:52:43 GMT

I have a mare Knatmeg Scenic/Victorian Opal and am looking for a Darshaan / Shirley Heights type cross in Australia. Does anyone know of a stallion or suggest a mating.

re: Reader Q&A--Statistical Inquiries on Nicking

MurrayK — Sat, 04 Jun 2011 10:39:37 GMT

The proof is in the pudding!!! Three Group 1 European Guineas winners in the UK, France & Ireland in a 2 week period recently for the Galileo x Danehill cross,

Frankel, Golden Lilacs & Roderick O'Connor, was truly an amazing performance, they join a growing list bred on this cross.

re: Reader Q&A--Statistical Inquiries on Nicking

Byron Rogers — Wed, 01 Jun 2011 19:23:55 GMT

Sceptre

No problem. Feel free to email byron@pedigreeconsultants.com

Byron

re: Reader Q&A--Statistical Inquiries on Nicking

sceptre — Wed, 01 Jun 2011 02:00:58 GMT

Hi Byron,

I wrote a lengthly reply to your most recent post, but the system apparently failed and it wasn't transmitted (I retained no copy). I won't attempt to re-create it, and to be honest, I have a whole new set of thoughts that for now I'll refrain from offering-not sure that any will come as news to you. Perhaps, one day we can communicate by phone or in person.

Thanks very much for airing this, and for your detailed and thoughtful replies.

re: Reader Q&A--Statistical Inquiries on Nicking

Byron Rogers — Sat, 28 May 2011 18:51:37 GMT

Sceptre,

We took your comment as read. After all, you have been a long time contributor to this blog and posts like yours do make us really think and evaluate the TrueNicks rating and algorithm like we have here. So in that respect we are always grateful that people like you take the time to make comment. I also apologize for the length of this reply in advance.

We did actually get a PhD statistician to look at the initial numbers to determine firstly if the rules that we created would generate a statistically significant sample size and then if the ratings that were generated had statistical significance themselves. We were quite guided by what he said in that we didn't go back any further than the third generation on the sireline and the fourth generation on the broodmare because of his recommendation that the statistics generated were pretty meaningless.

In regards to the first part, what it came down to was a statistical power analysis whereby we had to determine (a) how large a sample is needed to enable statistical judgments that are accurate and reliable and (b) how likely our statistical test will be to detect effects of a given size in a particular situation. If the sample size is too low, then the test will lack the precision to provide reliable answers, if sample size is too large, time and resources will be wasted, often for minimal gain.

As you may recall we studied some 100,000 horses to create TrueNicks and it turns out that we were guilty of the latter - 100,000 was way too many horses for very little gain. The a priori power assumption was that it would be somewhere around 30,000 horses to make valid assumptions on the total population, but we overpowered the analysis on the basis of it to be 'better safe than sorry'. Subsequent post-hoc analysis proved that we could have done the analysis on just 3,000 random horses and we would have gotten similar results with similar sampling error and had enough power in relation to the population.

Moving on to the calculation itself, and more specifically your question of "how does one determine what numbers are sufficient", it is a difficult question to answer indeed because among other factors it not only boils down to how many starters you need, but how many starts these starters have had to give them the opportunity to perform on the racetrack.

TrueNicks relies on two rules that only calculate on the basis of a number of starters making a significant number of starts. Again, when we were designing the rating we looked at different scenarios of starters (10,15,20,etc) and starts (100,200,300, etc), before settling on the parameters that we use today. This wasn't as easy as it may sound as we had to consider young stallions with their first runners, stallions whose progeny take time to mature and scenario's where old stallions were being bred to very young broodmare sires as well.

We also do a 'faceoff' in within the rules. If you have a hypothetical mating there is just as much relevance, generationally speaking, between a calculation done on the sire of the sire (2 gen) and the broodmare sire (2 gen) and the sire (1 gen) and the sire of the broodmare sire (3 gen). Again, we have a logarithmic calculation to return the most relevant rating.

There was also one deviation. You will notice that from time to time there are calculations made on a small number of runners - i.e Distorted Humor with El Prado mares - where there are two or more stakes winners out of unique mares. These are some of the highest variant scores that you will see. We wrestled a bit with this one as it is surprisingly a rather rare occurrence and we decided that it was better to calculate and show the rating based on two unique mares producing stakes winners on the cross, than not. That is probably the only part that fails statistical significance, but it is still actually quite predictive (i.e subsequent foals bred on the cross still become stakes winners at a higher rate than the alternatives, even if it is much less than the initial success).

Overall, we are quite satisfied as to the sample size we are using to make calculations and the predictive robustness of TrueNicks, but we are never satisfied that we can't improve on TrueNicks to make it an even better product. In line with your question on retrospective ratings, we are looking to develop a historical lookup tool that will allow us to look at larger populations of horses in various countries and have the data collected at various time points.

What does interest us now is to look at the variants, rule #, starters and starts considered when a rating is generated at a particular time or times (say conception, date of birth, 6 months after birth, 12 months after birth) and then see if there are more discreet differences within the rating based on the rule that it has been calculated on and the number of starters and starts, along with some quality indices of the horse to see if there is an even more predictive refinement that we can make. On the stats we have collected, it is possible, and probably not surprising to you and others, that not all "A's" are created equal and there are certain "A's" bred on certain rules with certain starter/starts parameters that may be significantly more predictive of subsequent stakes success than others.

Byron

re: Reader Q&A--Statistical Inquiries on Nicking

sceptre — Sat, 28 May 2011 01:59:32 GMT

Dear Byron,

I'm appreciative and honored that in this blog you chose to focus on my recent questions. Please know that they were indeed honest queries, rather than veiled statements of disagreement. While your Galileo/Danehill vs Galileo/all others "numbers" are what initially prompted my reflection and then questions, their comparative/analysis is but a simple example of the larger issue which continues to perplex-That is, what numbers are sufficient for samples such as these to render an accurate view of reality? I am not a statistician, and I assume that those skilled in this field could shed some light, but my guess is that they would raise also the spectre of variables (of which they may be far less acquainted with respect to the topic at hand) when attempting to answer this. Such a question (above) also applies to your hierarchal nick ratings, among other like things... I did note your attempt to answer my questions, but for me they fell short of what I was hoping for. Incidentally, I state in my first question "...all else equal...", so "...consideration of context (etc.)..." is off the table. Also, Alan specifically mentioned that the "quality" of Danehill mares vs all others was essentially equal. As to my second question, that concerning retrospective analysis, some of your examples, i.e. Unbridled's Song with Storm Cat mares did little more than add fuel to the fire...I fear that what I'm asking-How does one determine what number is sufficient"?-is not easily answered, and that its difficulty is compounded by the degree and complexity of the variables involved (genetic mechanisms themselves, many of which remain unknown, are among such variables). It's apparent to me that TrueNicks continues to fine tune this variable variable (meant to be written twice), but there is also intertwined within this the question of sufficient number. As said, I'm not a statistician, so perhaps/or perhaps not my questions are valid ones, and a true statistician might have more questions to pose. I did raise the retrospective issue, because I felt that a proper retrospective analysis of large numbers might evoke more confidence in the quantity of numbers you deem as sufficient presently.

re: Reader Q&A--Statistical Inquiries on Nicking

Ian Tapp — Fri, 27 May 2011 18:48:01 GMT

Hi Ryan,

To clarify the comment about Northern Dancer: in general, crossing a sire and broodmare from the same immediate sire line under-performs opportunity.

A good current example is Smart Strike/Mr. Prospector. This is a Mr. P/Mr. P cross; there are 64 starters by Smart Strike and sons out of Mr. P line mares, but zero SW.

With further generations certain crosses branch off, especially between contrasting examples of that sire line--for example, we wouldn't call Galileo/Danehill a "Northern Dancer/Northern Dancer" cross since it has distinguished itself specifically.

Furthermore, calling it a "Nearco/Nearco" cross would be completely irrelevant due to genetic divergence. Nicking is only useful when comparing groups of horses with some degree of genetic similarity. This is why TrueNicks only looks back as far as [grandsire of the sire] and [grandsire of the broodmare sire]. Encompassing more distant relations would make nicking comparisons impossible, as genetic divergence brings the group closer to breed average.

Regarding your comment on "top sires of sires", I'm not one to take stock in the "sire of sires" argument--I think that's a myth. A sire is either good or not. While sons of a particular stallion usually maintain similar genetic affinities to their sire, a stallion's individual success or failure at stud is not due simply to his sire or cross, but rather his total genetic profile and a multitude of environmental factors (including human decision-making).

re: Reader Q&A--Statistical Inquiries on Nicking

Ryan — Fri, 27 May 2011 15:07:45 GMT

Ian,

On an earlier thread you mentioned that the cross of all Northern Dancer sires over Northern Dancer mares produces a below average strike rate of stakes winners. Can you provide the strike rate (and frequency in the population if you can) of the most popular sire of sire cross. AP, MrP, ND, Roberto, and Blushing. This should produce the following permutations: APxMrP, APxND, APxRob, APxBlush, MrPxAP, NDxAP, RobxAP, BlushxAP, MrPxND, MrPxRob, MrPxBlush, NDxMrP, RobxMrP, RobxMrP, NDxRob, NDxBlush, RobxND, BlushxND, RobxBlush, BlushxRob, ApxAp, MrPxMrP, NDxND (already completed), RobxRob, BlushxBlush

This would identify the top sire of sires TrueNick. I'm sure most people would of thought that ND over ND would of been the top Sire of Sires TrueNick, but I was surprised to read that this is not the case.

I do understand that this can be done with some creative matings using your online resource but I thought it would be nice if you would provide for readers.

Europeans: This is a cry to please stop complaining about distance. Fans have been doing this for three centuries and it is old. Horses in America are primarily bred to win the Classics on dirt from 8 to 10 furlongs. To do this horses need to be extremely fast (fastest in the world) and carry their distance. Biomechanics, DNA, Nicks, cardio etc... prevent them from getting the Classic distances so we have a number of six furlong races. Get over it!