Rank Ordering: Is this a valid method to score different results?

Question

I have a system for handicapping horseraces. I take a given horserace and rank-order the horses in different factors, or categories (I look at things like speed, odds, etc). Say there are 5 horses I’m looking at: I give the “best” horse in each category/factor a rating of “1” and the worst horse a “5” and so on in between. (I look at something different in each “factor;” sometimes a low number is good, sometimes a higher number is good—the upshot is that the horses for each factor are ranked from 1- 5 (in this case) with 1 being the best.) Occasionally two or more horses have the exact same value in a particular category so they are given the same rating. I calculate a rating for each horse in each category/factor. There may be up to 6 categories. I add up all of the ratings and get a single score for each horse. Now I make a wager on the horse with the lowest score. Here is a simplified example: Horse Factor “A” Rating Factor “B” Rating Factor “C” Rating Score George 6 4 123 4 55 3 11 Ed 3 2 235 3 96 1 6 Sam 5 3 2000 1 42 4 8 Elmo 1 1 333 2 67 2 5 Paul 3 2 50 5 12 5 12

In this example I would bet on the horse called Elmo because it has the lowest score of 5. After scoring 6 categories it becomes clear which is the best horse because it will excel in more than one category and produce the lowest score. Questions: 1. Does this system have any validity? Can rank-ordering different “factors” against one another in several different categories help to identify how one “factor” is superior? 2. Assuming the lowest scored horse is indeed the winner, how can I determine which factor was more important than the others in selecting the winner? And if so, should I give an important factor more (or less) weight than others because of its influence? If so, how?

Answer 1

This idea is unlikely to work well in its current form because, as you indicate awareness of in question 2, your various categories (what we would usually call "independent variables", "predictors", "covariates", or "features") may not be equally important. The bread-and-butter tool in statistics for deciding how a number of independent variables can best be weighted to predict another variable (in this case, winning the race) is regression. I'd suggest reading an introductory text in linear regression, or statistics in general. You can certainly take much fancier approaches to predicting the outcomes of horse races than ordinary linear regression, but that's a good place to start. Don't forget to read existing research on predicting the outcomes of horse races, so you don't reinvent any wheels.