Multiple regression analysis

[redeye007]

I have an older handicapping program that performs a regression analysis after the top 4 finishers of the race are known and reweighs the handicapping varibles to more closely approximate the actual results and saves that system to disk for future predictions. I wonder if anyone has ever thought of taking it a step further and run a regression analysis of all the possible combinations of 3 or 4 horses in the race without knowing the results and coming up with a consensus of the predicted order of finish and would something like this produce more precise results?

[ranchwest]

I don't have data to back this up, but I think regression analysis points to the optimal horse, but the optimal horse is seldom at a good price and usually doesn't win. It would probably be much better as a spot play or used as a portion of a predicted odds formula.

[CBedo]

I have an older handicapping program that performs a regression analysis after the top 4 finishers of the race are known and reweighs the handicapping varibles to more closely approximate the actual results and saves that system to disk for future predictions. I wonder if anyone has ever thought of taking it a step further and run a regression analysis of all the possible combinations of 3 or 4 horses in the race without knowing the results and coming up with a consensus of the predicted order of finish and would something like this produce more precise results?There are lots of issues (good and bad) with using regression to predict horse finishes. I'm not exactly understanding your "combinations of 3 or 4 horses concept" but I'll note a few of the biggest issues to me. 1) Usually, in regression analysis, you are using some handicapping factors that are specific to the horse, so you come up with a "score" so to speak, instead of actually comparing horses in the race. You can get around this by using rankings or relative factors, but often this leads to issue 2) Factors aren't really independent (pace and speed are related) and if factors are correlated, then the regression analysis will be flawed. 3) The more factors you use, and the more specific you are in your race samples, you will get better and better 'test" results that more than likely won't hold up in the real world. You're usually better off with fewer, more significant factors to get better forecasts.

[Robert Goren]

I have done several of these. The first thing I found was that Log (odds+1) has a pretty good correlation. Then finding something that would improve upon it very much was hard. I suggest that you try some "off the wall" things that betters do not normally take into account. I wish luck, but I think you are :bang:.

[46zilzal]

The assumption here (in regression analysis) would be that ALL horse racing is somewhat the same and truthfully IT IS NOT

[CBedo]

The assumption here (in regression analysis) would be that ALL horse racing is somewhat the same and truthfully IT IS NOTWow! That took way longer than I expected for 46 to bust out the "horses are individiuals" speech. :lol:

[cmoore]

Wow! That took way longer than I expected for 46 to bust out the "horses are individiuals" speech. :lol:

Laughing my ass off!!!!

[garyoz]

Been many discussions about multiple regression and logistical regression or multivariate analysis on this board. Search and you'll find more than you want to know. There are statistical issues, don't even need to get into the validity question of the underlying variable construction.