How do I interpret this?

[Rick Ransom]

I just checked a longshot method yesterday and got the following results at one track:

30/183 16.39% +102.60 +28.03%

Wow, I thought, there might be something here! So today I checked it at a different track with the following results:

12/142 8.45% -149.20 -52.54%

Ugh, what do we have here? Something that works only at one track, or something that was unusually lucky in my test yesterday. I can't tell you how many times I've seen things like this over the years, and I just never know how to interpret it. I suppose I'll keep checking it a the first track, but the overall results don't give me a great deal of confidence.

What do you think? Is it a real difference or just an illusion?

[Dick Schmidt]

I think you need to check some more races at several other tracks. There are some track specific methods, but most that are valid work everywhere most of the time.

Dick

[baravot]

Rick,

Out of curiosity, did you check them for the exact same timeframe?

[hurrikane]

Rick,
I have several track specific plays performing like that. You don't say how many races you tested over. Many times I will run querys through the previous month, for instance..through the end of May and then see what would have happened if I had make the play in June. It's like working with 2 sets of data. Also you may try running the query for all days except lets say, Tue and Fri...and then run it for just Tue and Fri...see if you get the same results.
You should also restrict payout over say 19-1 to avoid scewing the data, or just look at the pay outs and see if there is 1 or 2 real big pays.

What you are seeing is not necessarily wrong..you just need to investigate further.

[Rick Ransom]

Dick,

This one involves workouts, so it might be different. One was an east coast track, the other a west coast track. In general I agree with you though, except maybe on something that involves track bias.

baravot,

No, different time periods. First test was Monmouth (this meet). Second test was Del Mar from last year.

hurrikane,

183 races at Monmouth, 142 at Del Mar. I was checking it manually to decide whether to include it in my database in the future. Overall it shows a 7.17% loss on 325 races, so it might be useful. It was just a preliminary test of a factor. I didn't even have odds limits. My point is that I was shocked at how different it was in the two samples. Statistical theory says the chance of this happening is about 0.19% or about 1 in 526 of the second sample coming from the same population as the first. But statistics sometimes doesn't work very well on horse racing data because the assumptions behind the mathematics aren't always met in the real world.

[baravot]

Rick,

You wrote: "But statistics sometimes doesn't work very well on horse racing data because the assumptions behind the mathematics aren't always met in the real world."

You've got that right!

[Rick Ransom]

baravot,

Yeah, the math assumes a normal distribution and a stationary process (nonvarying mean). That's convenient mathematically, but way off base in a lot of real-world things. Still, sometimes its useful for a ballpark estimate. I've seen some software that uses multiple regression for their predictions and I can tell you from experience that using it gives you the wrong answers. There are some ways of using statistics that work fairly well however, such as logit models, but you'd probably have to run a different one for each track and you'd need about 1000 races at that track to even get started. So don't worry about the math geniuses, they're no real threat.


By the way, here's today's update on the "sometimes awesome sometimes awful" longshot method. This sample is from part of the last Aqueduct meet:

25/127 19.69% +24.50 +9.65%

So maybe it's the "east coast longshot method". Really though, I'm not too serious about this. I just thought you guys would be amused since you've probably had similar ups and downs. Inexperienced handicappers are at a real disadvantage though, because they'll see something start out good and "bet the farm" just when it turns bad.

[Rick Ransom]

Another update. This one is from the last Santa Anita meet:

19/119 15.97% -33.80 -14.20%

Still better at east coast tracks. The overall results are:

East Coast Tracks:

55/310 17.74% +127.10 +20.50%, Avg Win Pay 13.58

West Coast Tracks:

31/261 11.88% -183.00 -35.06%, Avg Win Pay 10.94

Now, I'm going to make the anti-database guys really happy and point out that you never would have seen this if you had run a query on a database with all four tracks. You would have seen the following:

86/571 15.06% -55.90 -4.89%

These overall results would make you optimistic about the potential of this factor and maybe even play the higher odds horses at the next meet. How about Del Mar? That's coming up soon. Uh oh, you might be making a big mistake if Del Mar this year repeats how it was last year.

My point is that the psychological impact of this information is entirely different when you look at subsets of the data than when you look at the whole. I don't even know for sure that there is a real difference here, but if there isn't I think you'd agree that, unless you're a machine, you'd bet differently depending on which track you started with.

Hey, just to cheer you up, don't forget that in the long run ...... you're dead.

[Larry Hamilton]

that is the neatest expression I've heard in some time....may I "borrow" it?

In the long run, you are dead!

hehehe

[Rick Ransom]

Larry,

It's not original, but I can't remember who said it. Maybe someone else can. Anyway, I've always liked it.

Oh, by the way, the "east coast method" was 3 for 9 today at Louisiana Downs with $12.00 profit, so I guess we'll have to include the south too!

[Rick Ransom]

Ok, I've got it for anyone who's interested. The quote is "in the long run, we are all dead" and it's from economist John Maynard Keynes (1883-1946).

[Larry Hamilton]

Thanks! Reminds me another good one, dont know the author: If you want a new idea, read an old book.

Rick - IMO the sample sizes for all of the varations of your test are way to small to draw any conlcusioins, but I do believe that there are differences between tracks and that with very large samples you still find big swings in results of various queries.

[Tom]

You have to live through the short term to get to the long term. Overall, one HTR factor hits about 79-82% of the top 4 winning races, yet just in the last week it went 0-12. Normally it hits around 22-25% for the top rated hore, so if you geared youur wagering to expect 22% of your wagers to win, you would be drinking some serious beer right now. Things change suddenly and for no apparent reason, and you need to be alert. For me, as soon as I saw one day with 0 results I backed off, but that could easily have killed me if I had been lazy and not updated my model.
Tom

[Lefty]

Rick, Larry, and remember the last line of the song, THE
GAMBLER. "and the gambler he broke even, he died in his sleep"

[Larry Hamilton]

no way I am breaking even, I have some fabulous toys

[Rick Ransom]

Thanks for all the amusing replies, guys.

Tom,

I know what you're saying about small samples, but when you break this down into the east and west groups you have 310 plays at 17.74% and 261 plays at 11.88%. That's a pretty big difference, and we're not talking about 30 races here. My statistics book says the second group is 2.70 standard deviations away from the first, which would occur by chance about 0.35% of the time (odds of about 285 to 1). Despite what I said earlier about not always being able to trust statistics in horse racing, I still think we're either seeing a real difference here or a very rare event.

When I get time I'll check 99 Del Mar and see how that was. I also intend to check the upcoming Del Mar meet. I don't think the method itself is of as much interest as whether differences this large really exist between tracks, and that's what I want to find out.

[Larry Hamilton]

since you have started with the premise that each track may be different, one approach might be to list the variables that exist at each track, such as:

1. Dirt Consistensy (cant do anything about this one)

2. Measurement, dimentions, and shape of track

3. Class of animals that race there, if it can be generalied.

4. I am sure there are more, that's all I could get off the top of me pate.

My point is this: if you cant find the reason for the difference and prove it, there is no difference you can use. All the differences you have found, if you cant explain them, are no better than random distribution

As an example, a couple weeks ago, I built a table that lists all the tracks and the length of their stretch. When I get around to it, I will attempt to calculate the angles around each track then determine if there is some performance limit going around the turns. Of course, each track will have a subset of distance. So, at the moment it is just a pile of data, nothing is proven.

[Dave Schwartz]

Consider the nature of the system...

If it were (for example) a system based upon workouts or workout patterns, that might very much be a "regional" thing. Trainers in the northeast do things differently than trainers in Southern California.

If it is a pace-bias tendancy, that would be potentially track specific.

If it were a class level approach, again, there might be a regional bias to that as that too differes from area to area.

That would be MHO.


Regards,
Dave Schwartz

[Rick Ransom]

I think there are regional differences, at least between the west coast and east coast. The east coast tracks are have a lot more sand in them because of the weather and I think California tracks have more clay. Also, the training patterns must be entirely different because I see many more workouts listed for California tracks in a given period of time than I do for east coast tracks.

I think there may be big differences between major tracks and minor tracks too. This is an area I'm not too familiar with, but I think trainer angles probably work better at small tracks. I'll bet if you look at Jon Worth's TIPS, some of them probably work way better at some tracks than others.

So, IMHO the big differences between tracks are in track bias and trainer intent factors (including workouts). Also, the higher class tracks are more predictable, but that is not necessarily a good thing. Consider a track where every horse always ran exactly the same speed as his last race. It would be impossible for anyone to beat the takeout. Everyone would have the same chance and everyone would lose.