Stats-Lies-Stats. A Few Ways to Tell the Difference

Jeff Platt - a full time horseplayer  - had this to say about Barry Meadow's recent article in Horseplayer Monthly (Page 1 here).

"That article by Barry Meadow should be required reading for anyone who does data mining."

I would agree, but I think even if you don't data mine, it's well worth reading too.

Oftentimes we are prey to the little bitty numbers at the bottom of the PP's or if we're looking at Trackmaster programs for clues on what to bet.  If we do a few things and use some common sense, we'll be much better off.

1. If we are datamining, Barry makes a few points about backfitting. Backfitting is simply looking to exclude horses out of a sample that makes your ROI or win percentage go up. This rarely holds moving forward. Why? Because we are excluding in a capricious fashion.

2. Use Common Sense: When we are trying to build something that will work for us - in harness or thoroughbred - our rules need to make sense.

For example, a case was giving on page two of the Horseplayer Monthly about maiden specials dropping to maiden claimers. Many handicapping books talk about horses having shown speed, dropping, as better bets and they are correct. This makes perfect sense. What about then using trainer stats to help us out even more.

MSW to MCL 14% over large sample, ROI $0.75
MSW to MCL w/ Horse showing top "E" number 25%, ROI $0.92
MSW to MCL w/ Top "E" number with a set of trainers who excel at the move 32% ROI $1.02

That progression of modeling follows a handicapping pattern.

3. We can still use our eyes. You see a trainer with a five year rate off the claim of 22-104. That's fair, but that's what the trainers win percentage is. What if, suddenly we see a run where the guy or gal goes 2 for 4? That's meaningless, too. Now, let's say out of those four, the horses raced unbelievably well - like I mean really well  - then we may be onto something with this trainer. New vet? Something sinister? It could be, and the stats don't tell us how they are off the claim until we see how the horses race. I like to use my eyes to ensure this trainer is worth watching off barn changes.

3. Subsets. Further, in both standardbred and thoroughbred racing, off claim moves happen regularly, and there is gold in them there hills. Subsetting off claim trainers by time off is one way that tends to work over time. For example, in harness there was a horse a few weeks ago who won off the claim at 9-2. The horses was a decent horse and it was not a shock, but a few things made this an interesting bet:

• The horse was off 25 days
• Lasix was added
• Several equipment changes, including the hopples out an amazing 4.5 inches
• The trainer won like this once or twice before

That's compelling, and not a one week wonder everyone and their brother is on board with (at a probable 4-5).

Similarly, in thoroughbred racing we see the same thing. With some trainers (if they don't like the horse) he is back in quickly. If they do see some potential, and they give the horse a little time, the horse will be good. Some trainers > 45 days but less than 60 days off the claim are tremendous.

You won't see that solely in the numbers, we have to use our eyes, like in number two.

Anyway, give Barry's article a read. It, and the rest of the magazine (it's free) had some interesting insight.

Have a great day.