Quote:
Originally Posted by ReplayRandall
What is your version of a "Large enough sample"?
|
For a simple test of accuracy between different sets of figures I would suggest 1500 races would be enough, using a straightforward metric of best figure last race. The chance that less accurate numbers would have a higher win rate then would seem very unlikely with a sample of that size. I'm not interested in having anyone pay for second rate stuff so if there's something more accurate out there definitely use that instead. Of course not everyone uses ratings in their handicapping anyway so the amount of benefit to using something like this as an input to your own individual handicapping process is going to vary.