Figure Making

[TGJB]

Bit-- that\'s a very sophisticated post, and some good questions. There are some assumptions involved, and some are correct, and others close to correct.

First of all, it\'s important to understand the process itself. As I said, when we make figures we look at various scenarios, trying to determine which is the most likely. In doing that we look at information outside the race itself, like the data for other races, as well as weather and track maintenance, for clues (as opposed to making dogmatic assumptions, like you-know-who). But most importantly, we look at the relationships within the race, and in doing so we have the advantage of adjusting for weight and ground to a reasonably accurate degree, which gives us as firm ground to stand on as one can have, as a starting point.

I use all this, when looking at the race, to find the single most likely figure scenario. The way you do this is by trying out different ones and seeing how they fit. When you first take your data base from raw figures based on pars to the projection method, you have no position about what percentage of the time 3yos, older horses, or anyone else will run or pair tops (in fact, the Thoro-Patterns are less than a year old). But you do use logic-- for example, you see that by doing a race one way, lots of horses would run in a RANGE they usually run in, and in all the other ways many would not. So in the early stages you find you can frame your decisions within a range.

After a period of time, doing things this way tightens up your data base, makes your figures more and more accurate. (We\'re having an argument here in the office as to whether this constitutes regression analysis as I write this). This enables you to further narrow the range of your decisions. Eventually, you discover that what has happened has created a situation where lots of horses run in a narrow range, and \"pair up\" numbers. Not just tops, but last race, etc.

Now, it\'s important to remember that the relationships between horses in a race are fixed, so you can\'t make this happen unless it really does, or unless you screw around with the relationships within a race, which would be self destructive-- the figures would become inaccurate and thus tougher to win with (losing us customers), tougher to work with in making future figures, and tougher for me to bet with and buy horses with. Making the figures fit together prettily is not an end in itself, it does nothing for you. And if you get it wrong, you\'ll have hell to pay when those horses go up against other horses because the relationships will come out wrong, and you can see where the whole thing would feed on itself, and instead of the data base becoming more and more accurate it would become less and less so-- you wouldn\'t be able to make the figures come out tight.

In terms of your questions, a standard situation that comes up is this-- you have a choice between giving 3 horses pairs of their tops, or deciding that all 3 picked the same day to run not just better than they ever have before, but BY EXACTLY THE SAME AMOUNT. It\'s far more likely that they would all run exactly back to their previous maximum level, than that they would all choose the exact same moment to improve exactly the same amount. Similarly, you often end up with choosing between giving a horse a new top, and three others a pair-up, or giving him a pair-up, and having the other three choose to run EXACTLY the same amount off maximum on EXACTLY the same day-- 98.5% of capacity, or whatever. It\'s far more likely that one horse jumped in that situation.

Aside from the pairups, there is the larger question of ranges, as I said earlier. In general, most horses run a high percentage of races in some range, and you can see that by doing a race one way you are pushing a very high percentage of the field out of that range or to the border of it, where another does not.

The upshot of all this is that those 3yo top percentages are a RESULT of general figure making practices, not a tool we use-- someone actually raised that exact question a few months ago, and I pointed out that they are large population studies, and shoudn\'t be used rigidly for small population studies (one race). I think I said that in a 7 horse field anywhere from 2 to 5 3yos pairing or running new tops would be reasonable, anything outside that would be a stretch. And here again, for figure making purposes, pairs are far more important than new tops-- I wouldn\'t have a problem giving out 2 new tops and 5 pairs. 5 pairs is strong.

As for your questions-- I do think I would get it right if a whole field of 3yos ran an off race, I might or might not if they \"X\"\'d. The chance of the latter is exremely remote-- it\'s maybe 15% for one of them to do it, the chance of 6 of them doing it is 15% to the sixth, which is a really small number. And even if it happened, if it was a day where the track speed was solid I would leave it alone. While I don\'t get situations that extreme, I do have to make decisions something like that, and in deciding what to do I both approach it as I have described, and look at real world circumstances. I had that discussion with David Patent here years ago about figures Ragozin assigned to an older horse stake on Preakness day 01 or 02-- he had nobody running a top, and all but one horse running at least 5 points off it\'s top. I broke the race out-- you can probably find the discussion with a search-- and there were indications from track maintenance, time between races etc. that I was right, independent of figure-logic.

I would also point out what others have pointed out here recently-- that when it comes to the Derby, since 1997, far fewer than 50% of the horses have run a top or better, so obviously I\'m not using that as a guide.

That aside, yes, I think that some errors are inevitable-- there is judgement involved-- and that they are a small and necessary price to pay for having the most accurate data base we can. You can\'t get them all right-- you can get them all \"right\", if you define right as using a dogmatic process with no basis in logic and reality. Then they\'re all right, within a half point.

We posted the Fountain Of Youth figures here on the site (you can find them with a search), and since it had come up and Bandini went forward I looked at again. To date there have been 9 subsequent starts by horses out of the race, with 3 new tops, 4 paired tops as defined in the Thoro-Patterns. It\'s still a little more likely that it\'s right the way I did it, but it\'s maybe 30% that you could take off two points.

I\'m a lot more worried about Afleet Alex\'s 6f race-- that was a situation where you had to give pretty much a whole field NEW tops, and the only question was how much, so I will be reviewing it periodically. There was no way to do the race so that two horses both made sense, let alone more than two.

[BitPlayer]

TGJB -

Thanks for the thorough and thoughtful response.  I think I came away with a misunderstanding about how you do things when last we touched upon these issues (in connection with the Fountain of Youth), but you\'ve cleared it up.  I suppose it\'s hard to really understand the process and how well it works without wallowing in it for a while.

Respectfully,

BitPlayer

[msola1]

Jerry,

You say,

\"It\'s far more likely that they would all run exactly back to their previous maximum level, than that they would all choose the exact same moment to improve exactly the same amount. Similarly, you often end up with choosing between giving a horse a new top, and three others a pair-up, or giving him a pair-up, and having the other three choose to run EXACTLY the same amount off maximum on EXACTLY the same day...\"

Why is it \"far more likely\" that they all run pairs instead of moving ahead or back together? What are the reasoning or statistics behind this? That they have all chosen the exact same moment to pair up strikes me as no more or less probable.

Thanks,

Mike

[NoCarolinaTony]

JB,

Would you say you are using SPC and six sigma type analysis when you prepare your data, (ie Upper and Lower Control limits and distribution  analysis charts)? When the data is presented in its final form is the data normalized to fit statistical control charts/distributions? To me it does sound a lot like linear regresion analysis is being employed. If so, what type of rules are used when the data is outside the contol limits?

At work,we use SPC and SQC statistical data to manage /control our manufacturing processes and finished goods and share that with our customers, and was wondering if in fact what you do to create the final  numbers is similar in methodology? Your explanition to Bitplayer sounds a lot like your are using those techniques. This insight might help me better utilize the TG product data in the future.

Regards,
NC Tony

PS I wear the Hat Proudly on the Golf Course.....Thanks

[TGJB]

NoCar-- you seemed like such a nice guy, and then you went and asked me questions like that. I don\'t have a formal statistics background-- ironically enough Alan used to work for a place that did studies like that, but he is not one of the four here who make figures.

Alan and I just talked about it, and it sounds like you might be describing techniques used for large population studies, or for finding averages (I could be wrong, since I really don\'t know what the hell \"SPC\" and \"six sigma type analysis\" are, and if you are bluffing, nice job). George might want to take a crack at this too, when he gets done laughing at my trying to deal with it. Anyway, if you want to define the terms, I\'ll take a shot at answering it.

I\'m not going to wear the hat on the golf course since I don\'t want anyone to use SPC to equate my ability to get out of sand traps with my ability to make figures.

[TGJB]

Msola-- I didn\'t say they were far more likely to all pair up than move forward or backward together. I said they were more likely to all pair than ALL move forward or backward EXACTLY THE SAME AMOUNT ON THE SAME DAY.

All the thinking in making (as well as betting with) figures is based on the concept that previous figures can be used as a guide in what horses will do subsequently. If you can\'t look at horses previous levels of performance-- ranges, and especially maximum level-- you can\'t make figures.

[NoCarolinaTony]

TGJB,

I think you answered my question .....LOL

None the less, it does for sure sound like you use Linear Regression as described in your response to Bit especially as it relates to races Such as the Blue Grass or The Wood. (If you were using Six Sigma et al I think we should pay you more for your data....LOL). In all honesty it does seem to me that you all do employ a very sophistacted statistical model none the less and I was just trying to correlate to what we do here at work. But the explanation provided to Bit was really very helpful for me undestanding the realtionships. I just ordered Len\'s Book(sorry) to better help me understand what goes on when making figures and will most certainly read it with the idea of just trying truly understand how to make figures for myself sometime.

One last thing......I\'m betting Anthony J tomorrow in AQU race 8 primarily because that is exactly My First Name and Middle Initial. (How sophisticated is that?..) He raced competitively against some quality horses his last two times out...And his Pattern seems to be going the right way and if he won and I didn\'t bet him, I\'d be pissed...I\'m sure he would be some value in exacta\'s.

Thanks,

NC Tony (Anthony J)

[flushedstraight]

Anthony J...

value in exactas? that\'s sounds like a major understatement

he looks to be totally ignored here and a more than decent shot to finish itm imho; considering the weights, modest development, and Perry\'s layoff stats, and what looks like a decent setup. Looks like an interesting race, the big A crowd should be lulled into a chalky stupor by then.

thanks in advance TG...